Add optional TLS support (#340)

* Add optional TLS support

Introduce HTTPS support with net/http Server.ListenAndServeTLS.

This should enable the option of serving via HTTPS without a reverse
proxy.

Add two flags:
- tls-cert-file (path to the TLS certificate file)
- tls-key-file (path to the TLS private key file)

Both flags must be supplied together; otherwise exit with error.

If both flags are present, call srv.ListenAndServeTLS.
If not, fall back to the existing srv.ListenAndServe (HTTP); no changes
to existing non‑TLS behavior.

.coderabbit.yaml

@@ -0,0 +1,15 @@
+# yaml-language-server: $schema=https://coderabbit.ai/integrations/schema.v2.json
+language: "en-US"
+early_access: false
+reviews:
+  profile: "chill"
+  request_changes_workflow: false
+  high_level_summary: true
+  poem: false
+  review_status: true
+  collapse_walkthrough: false
+  auto_review:
+    enabled: true
+    drafts: false
+chat:
+  auto_reply: true

.github/ISSUE_TEMPLATE/bug-report.md

@@ -0,0 +1,39 @@
+---
+name: Bug Report
+about: I found a defect
+title: ''
+labels: 'unconfirmed bug'
+assignees: ''
+
+---
+> [!IMPORTANT]
+> If you have questions about llama-swap please post in the Q&A in Discussions. Use bug reports when you've found a defect and wish to discuss a fix.
+
+**Describe the bug**
+A clear and concise description of what the bug is.
+
+**Expected behaviour**
+A clear and concise description of what you expected to happen.
+
+**Operating system and version**
+
+- OS: (linux, osx, windows, freebsd, etc)
+- GPUs: (list architecture)
+
+**My Configuration**
+
+```yaml
+# copy / paste your configuration here
+```
+
+**Proxy Logs**
+
+```
+# copy / paste from /logs
+```
+
+**Upstream Logs**
+
+```
+# copy/paste from /logs
+```

.github/workflows/closeinactive.yml

@@ -0,0 +1,23 @@
+# https://docs.github.com/en/actions/use-cases-and-examples/project-management/closing-inactive-issues
+name: Close inactive issues
+on:
+  schedule:
+    - cron: "32 1 * * *"
+
+jobs:
+  close-issues:
+    runs-on: ubuntu-latest
+    permissions:
+      issues: write
+      pull-requests: write
+    steps:
+      - uses: actions/stale@v9
+        with:
+          days-before-issue-stale: 14
+          days-before-issue-close: 14
+          stale-issue-label: "stale"
+          stale-issue-message: "This issue is stale because it has been open for 2 weeks with no activity."
+          close-issue-message: "This issue was closed because it has been inactive for 2 weeks since being marked as stale."
+          days-before-pr-stale: -1
+          days-before-pr-close: -1
+          repo-token: ${{ secrets.GITHUB_TOKEN }}

.github/workflows/containers.yml

@@ -0,0 +1,46 @@
+name: Build Containers
+
+on:
+  # time has no specific meaning, trying to time it after
+  # the llama.cpp daily packages are published
+  # https://github.com/ggml-org/llama.cpp/blob/master/.github/workflows/docker.yml
+  schedule:
+    - cron: "37 5 * * *"
+
+  # Allows manual triggering of the workflow
+  workflow_dispatch:
+
+jobs:
+  build-and-push:
+    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        platform: [intel, cuda, vulkan, cpu, musa]
+      fail-fast: false
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+
+      - name: Log in to GitHub Container Registry
+        uses: docker/login-action@v2
+        with:
+          registry: ghcr.io
+          username: ${{ github.actor }}
+          password: ${{ secrets.GITHUB_TOKEN }}
+
+      - name: Run build-container
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        run: ./docker/build-container.sh ${{ matrix.platform }} true
+
+  # note make sure mostlygeek/llama-swap has admin rights to the llama-swap package
+  # see: https://github.com/actions/delete-package-versions/issues/74
+  delete-untagged-containers:
+    needs: build-and-push
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/delete-package-versions@v5
+        with:
+          package-name: 'llama-swap'
+          package-type: 'container'
+          delete-only-untagged-versions: 'true'

.github/workflows/go-ci-windows.yml

@@ -0,0 +1,50 @@
+name: Windows CI
+
+on:
+  push:
+    branches: [ "main" ]
+
+  pull_request:
+    branches: [ "main" ]
+
+  # Allows manual triggering of the workflow
+  workflow_dispatch:
+
+jobs:
+
+  run-tests:
+    runs-on: windows-latest
+    steps:
+    - uses: actions/checkout@v4
+
+    - name: Set up Go
+      uses: actions/setup-go@v4
+      with:
+        go-version: '1.23'
+
+    # cache simple-responder to save the build time
+    - name: Restore Simple Responder
+      id: restore-simple-responder
+      uses: actions/cache/restore@v4
+      with:
+        path: ./build
+        key: ${{ runner.os }}-simple-responder-${{ hashFiles('misc/simple-responder/simple-responder.go') }}
+
+    # necessary for testing proxy/Process swapping
+    - name: Create simple-responder
+      if: steps.restore-simple-responder.outputs.cache-hit != 'true'
+      shell: bash
+      run: make simple-responder-windows
+
+    - name: Save Simple Responder
+      # nothing new to save ... skip this step
+      if: steps.restore-simple-responder.outputs.cache-hit != 'true'
+      id: save-simple-responder
+      uses: actions/cache/save@v4
+      with:
+        path: ./build
+        key: ${{ runner.os }}-simple-responder-${{ hashFiles('misc/simple-responder/simple-responder.go') }}
+
+    - name: Test all
+      shell: bash
+      run: make test-all

.github/workflows/go-ci.yml

@@ -0,0 +1,54 @@
+name: Linux CI
+
+on:
+  push:
+    branches: [ "main" ]
+
+  pull_request:
+    branches: [ "main" ]
+
+  # Allows manual triggering of the workflow
+  workflow_dispatch:
+
+jobs:
+
+  run-tests:
+    runs-on: ubuntu-latest
+    steps:
+    - uses: actions/checkout@v4
+
+    - name: Set up Go
+      uses: actions/setup-go@v4
+      with:
+        go-version: '1.23'
+
+    # Only run in this linux based runner
+    - name: Check Formatting
+      run: |
+        if [ "$(gofmt -l . | grep -v 'event/.*_test.go' | wc -l)" -gt 0 ]; then
+          gofmt -l . | grep -v 'event/.*_test.go'
+          exit 1
+        fi
+    # cache simple-responder to save the build time
+    - name: Restore Simple Responder
+      id: restore-simple-responder
+      uses: actions/cache/restore@v4
+      with:
+        path: ./build
+        key: ${{ runner.os }}-simple-responder-${{ hashFiles('misc/simple-responder/simple-responder.go') }}
+
+    # necessary for testing proxy/Process swapping
+    - name: Create simple-responder
+      run: make simple-responder
+
+    - name: Save Simple Responder
+      # nothing new to save ... skip this step
+      if: steps.restore-simple-responder.outputs.cache-hit != 'true'
+      id: save-simple-responder
+      uses: actions/cache/save@v4
+      with:
+        path: ./build
+        key: ${{ runner.os }}-simple-responder-${{ hashFiles('misc/simple-responder/simple-responder.go') }}
+
+    - name: Test all
+      run: make test-all

.github/workflows/release.yml

@@ -5,6 +5,13 @@ on:
     tags:
       - '*'
 
+  # Allows manual triggering of the workflow
+  workflow_dispatch:
+    inputs:
+      tag:
+        description: 'Tag version to release (e.g. v144)'
+        required: true
+
 permissions:
   contents: write
 
@@ -17,9 +24,22 @@ jobs:
         uses: actions/checkout@v4
         with:
           fetch-depth: 0
+          ref: ${{ github.event.inputs.tag || github.ref }}
       -
         name: Set up Go
         uses: actions/setup-go@v5
+      -
+        name: Set up Node.js
+        uses: actions/setup-node@v4
+        with:
+          node-version: '23'
+      -
+        name: Install dependencies and build UI
+        run: |
+          cd ui
+          npm ci
+          npm run build
+
       -
         name: Run GoReleaser
         uses: goreleaser/goreleaser-action@v6
@@ -31,3 +51,29 @@ jobs:
           args: release --clean
         env:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+
+  trigger-tap-update:
+    runs-on: ubuntu-latest
+    needs: goreleaser
+    steps:
+      - name: "Resolve tag to dispatch"
+        id: tag
+        run: |
+          if [[ "${{ github.event_name }}" == "workflow_dispatch" ]]; then
+            echo "tag=${{ github.event.inputs.tag }}" >> "$GITHUB_OUTPUT"
+          else
+            echo "tag=${{ github.ref_name }}" >> "$GITHUB_OUTPUT"
+          fi
+
+      - name: "Trigger tap repository update"
+        uses: peter-evans/repository-dispatch@v2
+        with:
+          token: ${{ secrets.TAP_REPO_PAT }}
+          repository: mostlygeek/homebrew-llama-swap
+          event-type: new-release
+          client-payload: |
+            {
+              "release": {
+                "tag_name": "${{ steps.tag.outputs.tag }}"
+              }
+            }

.gitignore

@@ -2,3 +2,6 @@
 .env
 build/
 dist/
+.vscode
+.DS_Store
+.dev/

.goreleaser.yaml

@@ -0,0 +1,32 @@
+version: 2
+
+builds:
+  - env:
+      - CGO_ENABLED=0
+    goos:
+      - linux
+      - darwin
+      - freebsd
+      - windows
+    goarch:
+      - amd64
+      - arm64
+    ignore:
+      - goos: freebsd
+        goarch: arm64
+      - goos: windows
+        goarch: arm64
+
+archives:
+  - id: default
+    formats:
+      - tar.gz
+    name_template: "{{ .ProjectName }}_{{ .Version }}_{{ .Os }}_{{ .Arch }}"
+    builds_info:
+      group: root
+      owner: root
+    format_overrides:
+      # use zip format for windows
+      - goos: windows
+        formats:
+          - zip

CLAUDE.md

@@ -0,0 +1,43 @@
+# Project: llama-swap
+
+## Project Description:
+
+llama-swap is a light weight, transparent proxy server that provides automatic model swapping to llama.cpp's server.
+
+## Tech stack
+
+- golang
+- typescript, vite and react for UI (ui/)
+
+## Testing
+
+- `make test-dev` - Use this when making iterative changes. Runs `go test` and `staticcheck`. Fix any static checking errors.
+- `make test-all` - runs at the end before completing work. Includes long running concurrency tests.
+
+## Workflow Tasks
+
+### Plan Improvements
+
+Work plans are located in ai-plans/. Plans written by the user may be incomplete, contain inconsistencies or errors.
+
+When the user asks to improve a plan follow these guidelines for expanding and improving it.
+
+- Identify any inconsistencies.
+- Expand plans out to be detailed specification of requirements and changes to be made.
+- Plans should have at least these sections:
+  - Title - very short, describes changes
+  - Overview: A more detailed summary of goal and outcomes desired
+  - Design Requirements: Detailed descriptions of what needs to be done
+  - Testing Plan: Tests to be implemented
+  - Checklist: A detailed list of changes to be made
+
+Look for "plan expansion" as explicit instructions to improve a plan.
+
+### Implementation of plans
+
+When the user says "paint it", respond with "commencing automated assembly". Then implement the changes as described by the plan. Update the checklist as you complete items.
+
+## General Rules
+
+- when summarizing changes only include details that require further action (action items)
+- when there are no action items, just say "Done."

Makefile

@@ -2,6 +2,16 @@
 APP_NAME = llama-swap
 BUILD_DIR = build
 
+# Get the current Git hash
+GIT_HASH := $(shell git rev-parse --short HEAD)
+ifneq ($(shell git status --porcelain),)
+    # There are untracked changes
+    GIT_HASH := $(GIT_HASH)+
+endif
+
+# Capture the current build date in RFC3339 format
+BUILD_DATE := $(shell date -u +"%Y-%m-%dT%H:%M:%SZ")
+
 # Default target: Builds binaries for both OSX and Linux
 all: mac linux simple-responder
 
@@ -9,24 +19,72 @@ all: mac linux simple-responder
 clean:
 	rm -rf $(BUILD_DIR)
 
+proxy/ui_dist/placeholder.txt:
+	mkdir -p proxy/ui_dist
+	touch $@
+
+# use cached test results while developing
+test-dev: proxy/ui_dist/placeholder.txt
+	go test -short ./proxy/...
+	staticcheck ./proxy/... || true
+
+test: proxy/ui_dist/placeholder.txt
+	go test -short -count=1 ./proxy/...
+
+# for CI - full test (takes longer)
+test-all: proxy/ui_dist/placeholder.txt
+	go test -race -count=1 ./proxy/...
+
+ui/node_modules:
+	cd ui && npm install
+
+# build react UI
+ui: ui/node_modules
+	cd ui && npm run build
+
 # Build OSX binary
-mac:
+mac: ui
 	@echo "Building Mac binary..."
-	GOOS=darwin GOARCH=arm64 go build -o $(BUILD_DIR)/$(APP_NAME)-darwin-arm64
+	GOOS=darwin GOARCH=arm64 go build -ldflags="-X main.commit=${GIT_HASH} -X main.version=local_${GIT_HASH} -X main.date=${BUILD_DATE}" -o $(BUILD_DIR)/$(APP_NAME)-darwin-arm64
 
 # Build Linux binary
-linux:
+linux: ui
 	@echo "Building Linux binary..."
-	GOOS=linux GOARCH=amd64 go build -o $(BUILD_DIR)/$(APP_NAME)-linux-amd64
+	GOOS=linux GOARCH=amd64 go build -ldflags="-X main.commit=${GIT_HASH} -X main.version=local_${GIT_HASH} -X main.date=${BUILD_DATE}" -o $(BUILD_DIR)/$(APP_NAME)-linux-amd64
+	GOOS=linux GOARCH=arm64 go build -ldflags="-X main.commit=${GIT_HASH} -X main.version=local_${GIT_HASH} -X main.date=${BUILD_DATE}" -o $(BUILD_DIR)/$(APP_NAME)-linux-arm64
 
-# for testing things
+# Build Windows binary
+windows: ui
+	@echo "Building Windows binary..."
+	GOOS=windows GOARCH=amd64 go build -ldflags="-X main.commit=${GIT_HASH} -X main.version=local_${GIT_HASH} -X main.date=${BUILD_DATE}" -o $(BUILD_DIR)/$(APP_NAME)-windows-amd64.exe
+
+# for testing proxy.Process
 simple-responder:
 	@echo "Building simple responder"
-	go build -o $(BUILD_DIR)/simple-responder misc/simple-responder/simple-responder.go
+	GOOS=darwin GOARCH=arm64 go build -o $(BUILD_DIR)/simple-responder_darwin_arm64 misc/simple-responder/simple-responder.go
+	GOOS=linux GOARCH=amd64 go build -o $(BUILD_DIR)/simple-responder_linux_amd64 misc/simple-responder/simple-responder.go
+
+simple-responder-windows:
+	@echo "Building simple responder for windows"
+	GOOS=windows GOARCH=amd64 go build -o $(BUILD_DIR)/simple-responder.exe misc/simple-responder/simple-responder.go
 
 # Ensure build directory exists
 $(BUILD_DIR):
 	mkdir -p $(BUILD_DIR)
 
+# Create a new release tag
+release:
+	@echo "Checking for unstaged changes..."
+	@if [ -n "$(shell git status --porcelain)" ]; then \
+		echo "Error: There are unstaged changes. Please commit or stash your changes before creating a release tag." >&2; \
+		exit 1; \
+	fi
+
+# Get the highest tag in v{number} format, increment it, and create a new tag
+	@highest_tag=$$(git tag --sort=-v:refname | grep -E '^v[0-9]+$$' | head -n 1 || echo "v0"); \
+	new_tag="v$$(( $${highest_tag#v} + 1 ))"; \
+	echo "tagging new version: $$new_tag"; \
+	git tag "$$new_tag";
+
 # Phony targets
-.PHONY: all clean osx linux
+.PHONY: all clean ui mac linux windows simple-responder test test-all test-dev

README.md

@@ -1,77 +1,237 @@
+![llama-swap header image](header2.png)
+![GitHub Downloads (all assets, all releases)](https://img.shields.io/github/downloads/mostlygeek/llama-swap/total)
+![GitHub Actions Workflow Status](https://img.shields.io/github/actions/workflow/status/mostlygeek/llama-swap/go-ci.yml)
+![GitHub Repo stars](https://img.shields.io/github/stars/mostlygeek/llama-swap)
+
 # llama-swap
 
-[llama.cpp's server](https://github.com/ggerganov/llama.cpp/tree/master/examples/server) can't swap models, so let's swap llama-server instead!
+llama-swap is a light weight, transparent proxy server that provides automatic model swapping to llama.cpp's server.
 
-llama-swap is a proxy server that sits in front of llama-server. When a request for `/v1/chat/completions` comes in it will extract the `model` requested and change the underlying llama-server automatically.
+Written in golang, it is very easy to install (single binary with no dependencies) and configure (single yaml file). To get started, download a pre-built binary, a provided docker images or Homebrew.
 
-- ✅ easy to deploy: single binary with no dependencies
+## Features:
-- ✅ full control over llama-server's startup settings
+
-- ✅ ❤️ for nvidia P40 users who are rely on llama.cpp for inference
+- ✅ Easy to deploy: single binary with no dependencies
+- ✅ Easy to config: single yaml file
+- ✅ On-demand model switching
+- ✅ OpenAI API supported endpoints:
+  - `v1/completions`
+  - `v1/chat/completions`
+  - `v1/embeddings`
+  - `v1/audio/speech` ([#36](https://github.com/mostlygeek/llama-swap/issues/36))
+  - `v1/audio/transcriptions` ([docs](https://github.com/mostlygeek/llama-swap/issues/41#issuecomment-2722637867))
+- ✅ llama-server (llama.cpp) supported endpoints:
+  - `v1/rerank`, `v1/reranking`, `/rerank`
+  - `/infill` - for code infilling
+  - `/completion` - for completion endpoint
+- ✅ llama-swap custom API endpoints
+  - `/ui` - web UI
+  - `/log` - remote log monitoring
+  - `/upstream/:model_id` - direct access to upstream HTTP server ([demo](https://github.com/mostlygeek/llama-swap/pull/31))
+  - `/unload` - manually unload running models ([#58](https://github.com/mostlygeek/llama-swap/issues/58))
+  - `/running` - list currently running models ([#61](https://github.com/mostlygeek/llama-swap/issues/61))
+  - `/health` - just returns "OK"
+- ✅ Run multiple models at once with `Groups` ([#107](https://github.com/mostlygeek/llama-swap/issues/107))
+- ✅ Automatic unloading of models after timeout by setting a `ttl`
+- ✅ Use any local OpenAI compatible server (llama.cpp, vllm, tabbyAPI, etc)
+- ✅ Reliable Docker and Podman support using `cmd` and `cmdStop` together
+- ✅ Full control over server settings per model
+- ✅ Preload models on startup with `hooks` ([#235](https://github.com/mostlygeek/llama-swap/pull/235))
+
+## How does llama-swap work?
+
+When a request is made to an OpenAI compatible endpoint, llama-swap will extract the `model` value and load the appropriate server configuration to serve it. If the wrong upstream server is running, it will be replaced with the correct one. This is where the "swap" part comes in. The upstream server is automatically swapped to the correct one to serve the request.
+
+In the most basic configuration llama-swap handles one model at a time. For more advanced use cases, the `groups` feature allows multiple models to be loaded at the same time. You have complete control over how your system resources are used.
 
 ## config.yaml
 
-llama-swap's configuration purposefully simple.
+llama-swap is managed entirely through a yaml configuration file.
+
+It can be very minimal to start:
 
 ```yaml
-# Seconds to wait for llama.cpp to load and be ready to serve requests
-# Default (and minimum) is 15 seconds
-healthCheckTimeout: 60
-
-# define valid model values and the upstream server start
 models:
-  "llama":
+  "qwen2.5":
-    cmd: "llama-server --port 8999 -m Llama-3.2-1B-Instruct-Q4_K_M.gguf"
+    cmd: |
-
+      /path/to/llama-server
-    # Where to proxy to, important it matches this format
+      -hf bartowski/Qwen2.5-0.5B-Instruct-GGUF:Q4_K_M
-    proxy: "http://127.0.0.1:8999"
+      --port ${PORT}
-
-    # aliases model names to use this configuration for
-    aliases:
-    - "gpt-4o-mini"
-    - "gpt-3.5-turbo"
-
-  "qwen":
-    # environment variables to pass to the command
-    env:
-      - "CUDA_VISIBLE_DEVICES=0"
-    cmd: "llama-server --port 8999 -m path/to/Qwen2.5-1.5B-Instruct-Q4_K_M.gguf"
-    proxy: "http://127.0.0.1:8999"
 ```
 
+However, there are many more capabilities that llama-swap supports:
+
+- `groups` to run multiple models at once
+- `ttl` to automatically unload models
+- `macros` for reusable snippets
+- `aliases` to use familiar model names (e.g., "gpt-4o-mini")
+- `env` to pass custom environment variables to inference servers
+- `cmdStop` for to gracefully stop Docker/Podman containers
+- `useModelName` to override model names sent to upstream servers
+- `healthCheckTimeout` to control model startup wait times
+- `${PORT}` automatic port variables for dynamic port assignment
+
+See the [configuration documentation](https://github.com/mostlygeek/llama-swap/wiki/Configuration) in the wiki all options and examples.
+
+## Reverse Proxy Configuration (nginx)
+
+If you deploy llama-swap behind nginx, disable response buffering for streaming endpoints. By default, nginx buffers responses which breaks Server‑Sent Events (SSE) and streaming chat completion. ([#236](https://github.com/mostlygeek/llama-swap/issues/236))
+
+Recommended nginx configuration snippets:
+
+```nginx
+# SSE for UI events/logs
+location /api/events {
+    proxy_pass http://your-llama-swap-backend;
+    proxy_buffering off;
+    proxy_cache off;
+}
+
+# Streaming chat completions (stream=true)
+location /v1/chat/completions {
+    proxy_pass http://your-llama-swap-backend;
+    proxy_buffering off;
+    proxy_cache off;
+}
+```
+
+As a safeguard, llama-swap also sets `X-Accel-Buffering: no` on SSE responses. However, explicitly disabling `proxy_buffering` at your reverse proxy is still recommended for reliable streaming behavior.
+
+## Web UI
+
+llama-swap includes a real time web interface for monitoring logs and models:
+
+<img width="1360" height="963" alt="image" src="https://github.com/user-attachments/assets/adef4a8e-de0b-49db-885a-8f6dedae6799" />
+
+The Activity Page shows recent requests:
+
+<img width="1360" height="963" alt="image" src="https://github.com/user-attachments/assets/5f3edee6-d03a-4ae5-ae06-b20ac1f135bd" />
+
 ## Installation
 
-1. Create a configuration file, see [config.example.yaml](config.example.yaml)
+llama-swap can be installed in multiple ways
+
+1. Docker
+2. Homebrew (OSX and Linux)
+3. From release binaries
+4. From source
+
+### Docker Install ([download images](https://github.com/mostlygeek/llama-swap/pkgs/container/llama-swap))
+
+Docker images with llama-swap and llama-server are built nightly.
+
+```shell
+# use CPU inference comes with the example config above
+$ docker run -it --rm -p 9292:8080 ghcr.io/mostlygeek/llama-swap:cpu
+
+# qwen2.5 0.5B
+$ curl -s http://localhost:9292/v1/chat/completions \
+    -H "Content-Type: application/json" \
+    -H "Authorization: Bearer no-key" \
+    -d '{"model":"qwen2.5","messages": [{"role": "user","content": "tell me a joke"}]}' | \
+    jq -r '.choices[0].message.content'
+
+# SmolLM2 135M
+$ curl -s http://localhost:9292/v1/chat/completions \
+    -H "Content-Type: application/json" \
+    -H "Authorization: Bearer no-key" \
+    -d '{"model":"smollm2","messages": [{"role": "user","content": "tell me a joke"}]}' | \
+    jq -r '.choices[0].message.content'
+```
+
+<details>
+<summary>Docker images are built nightly with llama-server for cuda, intel, vulcan and musa.</summary>
+
+They include:
+
+- `ghcr.io/mostlygeek/llama-swap:cpu`
+- `ghcr.io/mostlygeek/llama-swap:cuda`
+- `ghcr.io/mostlygeek/llama-swap:intel`
+- `ghcr.io/mostlygeek/llama-swap:vulkan`
+- ROCm disabled until fixed in llama.cpp container
+
+Specific versions are also available and are tagged with the llama-swap, architecture and llama.cpp versions. For example: `ghcr.io/mostlygeek/llama-swap:v89-cuda-b4716`
+
+Beyond the demo you will likely want to run the containers with your downloaded models and custom configuration.
+
+```shell
+$ docker run -it --rm --runtime nvidia -p 9292:8080 \
+  -v /path/to/models:/models \
+  -v /path/to/custom/config.yaml:/app/config.yaml \
+  ghcr.io/mostlygeek/llama-swap:cuda
+```
+
+</details>
+
+### Homebrew Install (macOS/Linux)
+
+The latest release of `llama-swap` can be installed via [Homebrew](https://brew.sh).
+
+```shell
+# Set up tap and install formula
+brew tap mostlygeek/llama-swap
+brew install llama-swap
+# Run llama-swap
+llama-swap --config path/to/config.yaml --listen localhost:8080
+```
+
+This will install the `llama-swap` binary and make it available in your path. See the [configuration documentation](https://github.com/mostlygeek/llama-swap/wiki/Configuration)
+
+### Pre-built Binaries ([download](https://github.com/mostlygeek/llama-swap/releases))
+
+Binaries are available for Linux, Mac, Windows and FreeBSD. These are automatically published and are likely a few hours ahead of the docker releases. The binary install works with any OpenAI compatible server, not just llama-server.
+
 1. Download a [release](https://github.com/mostlygeek/llama-swap/releases) appropriate for your OS and architecture.
-    * _Note: Windows currently untested._
+1. Create a configuration file, see the [configuration documentation](https://github.com/mostlygeek/llama-swap/wiki/Configuration).
-1. Run the binary with `llama-swap --config path/to/config.yaml`
+1. Run the binary with `llama-swap --config path/to/config.yaml --listen localhost:8080`.
+   Available flags:
+   - `--config`: Path to the configuration file (default: `config.yaml`).
+   - `--listen`: Address and port to listen on (default: `:8080`).
+   - `--version`: Show version information and exit.
+   - `--watch-config`: Automatically reload the configuration file when it changes. This will wait for in-flight requests to complete then stop all running models (default: `false`).
 
-## Systemd Unit Files
+### Building from source
 
-Use this unit file to start llama-swap on boot. This is only tested on Ubuntu.
+1. Build requires golang and nodejs for the user interface.
+1. `git clone https://github.com/mostlygeek/llama-swap.git`
+1. `make clean all`
+1. Binaries will be in `build/` subdirectory
 
-`/etc/systemd/system/llama-swap.service`
+## Monitoring Logs
-```
-[Unit]
-Description=llama-swap
-After=network.target
 
-[Service]
+Open the `http://<host>:<port>/` with your browser to get a web interface with streaming logs.
-User=nobody
 
-# set this to match your environment
+CLI access is also supported:
-ExecStart=/path/to/llama-swap --config /path/to/llama-swap.config.yml
 
-Restart=on-failure
+```shell
-RestartSec=3
+# sends up to the last 10KB of logs
-StartLimitBurst=3
+curl http://host/logs'
-StartLimitInterval=30
 
-[Install]
+# streams combined logs
-WantedBy=multi-user.target
+curl -Ns 'http://host/logs/stream'
+
+# just llama-swap's logs
+curl -Ns 'http://host/logs/stream/proxy'
+
+# just upstream's logs
+curl -Ns 'http://host/logs/stream/upstream'
+
+# stream and filter logs with linux pipes
+curl -Ns http://host/logs/stream | grep 'eval time'
+
+# skips history and just streams new log entries
+curl -Ns 'http://host/logs/stream?no-history'
 ```
 
-## Building from Source
+## Do I need to use llama.cpp's server (llama-server)?
 
-1. Install golang for your system
+Any OpenAI compatible server would work. llama-swap was originally designed for llama-server and it is the best supported.
-1. run `make clean all`
+
-1. binaries will be built into `build/` directory
+For Python based inference servers like vllm or tabbyAPI it is recommended to run them via podman or docker. This provides clean environment isolation as well as responding correctly to `SIGTERM` signals to shutdown.
+
+## Star History
+
+> [!NOTE]
+> ⭐️ Star this project to help others discover it! 
+
+[![Star History Chart](https://api.star-history.com/svg?repos=mostlygeek/llama-swap&type=Date)](https://www.star-history.com/#mostlygeek/llama-swap&Date)

ai-plans/issue-264-add-metadata.md

@@ -0,0 +1,292 @@
+# Add Model Metadata Support with Typed Macros
+
+## Overview
+
+Implement support for arbitrary metadata on model configurations that can be exposed through the `/v1/models` API endpoint. This feature extends the existing macro system to support scalar types (string, int, float, bool) instead of only strings, enabling type-safe metadata values.
+
+The metadata will be schemaless, allowing users to define any key-value pairs they need. Macro substitution will work within metadata values, preserving types when macros are used directly and converting to strings when macros are interpolated within strings.
+
+## Design Requirements
+
+### 1. Enhanced Macro System
+
+**Current State:**
+
+- Macros are defined as `map[string]string` at both global and model levels
+- Only string substitution is supported
+- Macros are replaced in: `cmd`, `cmdStop`, `proxy`, `checkEndpoint`, `filters.stripParams`
+
+**Required Changes:**
+
+- Change `MacroList` type from `map[string]string` to `map[string]any`
+- Support scalar types: `string`, `int`, `float64`, `bool`
+- Implement type-preserving macro substitution:
+  - Direct macro usage (`key: ${macro}`) preserves the macro's type
+  - Interpolated usage (`key: "text ${macro}"`) converts to string
+- Add validation to ensure macro values are scalar types only
+- Update existing macro substitution logic in [proxy/config/config.go](proxy/config/config.go) to handle `any` types
+
+**Implementation Details:**
+
+- Create a generic helper function to perform macro substitution that:
+  - Takes a value of type `any`
+  - Recursively processes maps, slices, and scalar values
+  - Replaces `${macro_name}` patterns with macro values
+  - Preserves types for direct substitution
+  - Converts to strings for interpolated substitution
+- Update `validateMacro()` function to accept `any` type and validate scalar types
+- Maintain backward compatibility with existing string-only macros
+
+### 2. Metadata Field in ModelConfig
+
+**Location:** [proxy/config/model_config.go](proxy/config/model_config.go)
+
+**Required Changes:**
+
+- Add `Metadata map[string]any` field to `ModelConfig` struct
+- Support YAML unmarshaling of arbitrary structures (maps, arrays, scalars)
+- Apply macro substitution to metadata values during config loading
+
+**Schema Requirements:**
+
+- Metadata is optional (default: empty/nil map)
+- Supports nested structures (objects within objects, arrays, etc.)
+- All string values within metadata undergo macro substitution
+- Type preservation rules apply as described above
+
+### 3. Macro Substitution in Metadata
+
+**Location:** [proxy/config/config.go](proxy/config/config.go) in `LoadConfigFromReader()`
+
+**Process Flow:**
+
+1. After loading YAML configuration
+2. After model-level and global macro merging
+3. Apply macro substitution to `ModelConfig.Metadata` field
+4. Use the same merged macros available to `cmd`, `proxy`, etc.
+5. Process recursively through all nested structures
+
+**Substitution Rules:**
+
+- `port: ${PORT}` → keeps integer type from PORT macro
+- `temperature: ${temp}` → keeps float type from temp macro
+- `note: "Running on ${PORT}"` → converts to string `"Running on 10001"`
+- Arrays and nested objects are processed recursively
+- Unknown macros should cause configuration load error (consistent with existing behavior)
+
+### 4. API Response Updates
+
+**Location:** [proxy/proxymanager.go:350](proxy/proxymanager.go#L350) `listModelsHandler()`
+
+**Current Behavior:**
+
+- Returns model records with: `id`, `object`, `created`, `owned_by`
+- Optionally includes: `name`, `description`
+
+**Required Changes:**
+
+- Add metadata to each model record under the key `llamaswap_meta`
+- Only include `llamaswap_meta` if metadata is non-empty
+- Preserve all types when marshaling to JSON
+- Maintain existing sorting by model ID
+
+**Example Response:**
+
+```json
+{
+  "object": "list",
+  "data": [
+    {
+      "id": "llama",
+      "object": "model",
+      "created": 1234567890,
+      "owned_by": "llama-swap",
+      "name": "llama 3.1 8B",
+      "description": "A small but capable model",
+      "llamaswap_meta": {
+        "port": 10001,
+        "temperature": 0.7,
+        "note": "The llama is running on port 10001 temp=0.7, context=16384",
+        "a_list": [1, 1.23, "macros are OK in list and dictionary types: llama"],
+        "an_obj": {
+          "a": "1",
+          "b": 2,
+          "c": [0.7, false, "model: llama"]
+        }
+      }
+    }
+  ]
+}
+```
+
+### 5. Validation and Error Handling
+
+**Macro Validation:**
+
+- Extend `validateMacro()` to accept values of type `any`
+- Verify macro values are scalar types: `string`, `int`, `float64`, `bool`
+- Reject complex types (maps, slices, structs) as macro values
+- Maintain existing validation for macro names and lengths
+
+**Configuration Loading:**
+
+- Fail fast if unknown macros are found in metadata
+- Provide clear error messages indicating which model and field contains errors
+- Ensure macros in metadata follow same rules as macros in cmd/proxy fields
+
+## Testing Plan
+
+### Test 1: Model-Level Macros with Different Types
+
+**File:** [proxy/config/model_config_test.go](proxy/config/model_config_test.go)
+
+**Test Cases:**
+
+- Define model with macros of each scalar type
+- Verify metadata correctly substitutes and preserves types
+- Test direct substitution (`port: ${PORT}`)
+- Test string interpolation (`note: "Port is ${PORT}"`)
+- Verify nested objects and arrays work correctly
+
+### Test 2: Global and Model Macro Precedence
+
+**File:** [proxy/config/config_test.go](proxy/config/config_test.go)
+
+**Test Cases:**
+
+- Define same macro at global and model level with different types
+- Verify model-level macro takes precedence
+- Test metadata uses correct macro value
+- Verify type is preserved from the winning macro
+
+### Test 3: Macro Validation
+
+**File:** [proxy/config/config_test.go](proxy/config/config_test.go)
+
+**Test Cases:**
+
+- Test that complex types (maps, arrays) are rejected as macro values
+  - Verify error message includes: macro name and type that was rejected
+- Test that scalar types (string, int, float, bool) are accepted
+  - Each type should load without error
+- Test macro name validation still works with `any` types
+  - Invalid characters, reserved names, length limits should still be enforced
+
+### Test 4: Metadata in API Response
+
+**File:** [proxy/proxymanager_test.go](proxy/proxymanager_test.go)
+
+**Existing Test:** `TestProxyManager_ListModelsHandler`
+
+**Test Cases:**
+
+- Model with metadata → verify `llamaswap_meta` key appears
+- Model without metadata → verify `llamaswap_meta` key is absent
+- Verify all types are correctly marshaled to JSON
+- Verify nested structures are preserved
+- Verify macro substitution has occurred before serialization
+
+### Test 5: Unknown Macros in Metadata
+
+**File:** [proxy/config/config_test.go](proxy/config/config_test.go)
+
+**Test Cases:**
+
+- Use undefined macro in metadata
+- Verify configuration loading fails with clear error
+- Error should indicate model name and that macro is undefined
+
+### Test 6: Recursive Substitution
+
+**File:** [proxy/config/config_test.go](proxy/config/config_test.go)
+
+**Test Cases:**
+
+- Metadata with deeply nested structures
+- Arrays containing objects with macros
+- Objects containing arrays with macros
+- Mixed string interpolation and direct substitution at various nesting levels
+
+## Checklist
+
+### Configuration Schema Changes
+
+- [x] Change `MacroList` type from `map[string]string` to `map[string]any` in [proxy/config/config.go:19](proxy/config/config.go#L19)
+- [x] Add `Metadata map[string]any` field to `ModelConfig` struct in [proxy/config/model_config.go:37](proxy/config/model_config.go#L37)
+- [x] Update `validateMacro()` function signature to accept `any` type for values
+- [x] Add validation logic to ensure macro values are scalar types only
+
+### Macro Substitution Logic
+
+- [x] Create generic recursive function `substituteMetadataMacros()` to handle `any` types
+- [x] Implement type-preserving direct substitution logic
+- [x] Implement string interpolation with type conversion
+- [x] Handle maps: recursively process all values
+- [x] Handle slices: recursively process all elements
+- [x] Handle scalar types: perform string-based macro substitution if value is string
+- [x] Integrate macro substitution into `LoadConfigFromReader()` after existing macro expansion
+- [x] Update existing macro substitution calls to use merged macros with correct types
+
+### API Response Changes
+
+- [x] Modify `listModelsHandler()` in [proxy/proxymanager.go:350](proxy/proxymanager.go#L350)
+- [x] Add `llamaswap_meta` field to model records when metadata exists
+- [x] Ensure empty metadata results in omitted `llamaswap_meta` key
+- [x] Verify JSON marshaling preserves all types correctly
+
+### Testing - Config Package
+
+- [x] Add test for string macros in metadata: [proxy/config/config_test.go](proxy/config/config_test.go)
+- [x] Add test for int macros in metadata: [proxy/config/config_test.go](proxy/config/config_test.go)
+- [x] Add test for float macros in metadata: [proxy/config/config_test.go](proxy/config/config_test.go)
+- [x] Add test for bool macros in metadata: [proxy/config/config_test.go](proxy/config/config_test.go)
+- [x] Add test for string interpolation in metadata: [proxy/config/config_test.go](proxy/config/config_test.go)
+- [x] Add test for model-level macro precedence: [proxy/config/config_test.go](proxy/config/config_test.go)
+- [x] Add test for nested structures in metadata: [proxy/config/config_test.go](proxy/config/config_test.go)
+- [x] Add test for unknown macro in metadata (should error): [proxy/config/config_test.go](proxy/config/config_test.go)
+- [x] Add test for invalid macro type validation: [proxy/config/config_test.go](proxy/config/config_test.go)
+
+### Testing - Model Config Package
+
+- [x] Add test cases to [proxy/config/model_config_test.go](proxy/config/model_config_test.go) for metadata unmarshaling
+- [x] Test metadata with various scalar types
+- [x] Test metadata with nested objects and arrays
+
+### Testing - Proxy Manager
+
+- [x] Update `TestProxyManager_ListModelsHandler` in [proxy/proxymanager_test.go](proxy/proxymanager_test.go)
+- [x] Add test case for model with metadata
+- [x] Add test case for model without metadata
+- [x] Verify `llamaswap_meta` key presence/absence
+- [x] Verify type preservation in JSON output
+- [x] Verify macro substitution has occurred
+
+### Documentation
+
+- [x] Verify [config.example.yaml](config.example.yaml) already has complete metadata examples (lines 149-171)
+- [x] No additional documentation needed per project instructions
+
+## Known Issues and Considerations
+
+### Inconsistencies
+
+None identified. The plan references the correct existing example in [config.example.yaml:149-171](config.example.yaml#L149-L171).
+
+### Design Decisions
+
+1. **Why `llamaswap_meta` instead of merging into record?**
+
+   - Avoids potential collisions with OpenAI API standard fields
+   - Makes it clear this is llama-swap specific metadata
+   - Easier for clients to distinguish standard vs. custom fields
+
+2. **Why support nested structures?**
+
+   - Provides maximum flexibility for users
+   - Aligns with the schemaless design principle
+   - Example config already demonstrates this capability
+
+3. **Why validate macro types?**
+   - Prevents confusing behavior (e.g., substituting a map)
+   - Makes configuration errors explicit at load time
+   - Simpler implementation and testing

ai-plans/issue-336-macro-in-macro.md

@@ -0,0 +1,397 @@
+# Improve macro-in-macro support
+
+**Status: COMPLETED ✅**
+
+## Title
+
+Fix macro substitution ordering by preserving definition order using ordered YAML parsing
+
+## Overview
+
+The current macro implementation uses `map[string]any` which does not preserve insertion order. This causes issues when macros reference other macros - if macro `B` contains `${A}` but `B` is processed before `A`, the reference won't be substituted, leading to "unknown macro" errors.
+
+**Goal:** Ensure macros are substituted in definition order (LIFO - last in, first out) to allow macros to reliably reference previously-defined macros.
+
+**Outcomes:**
+- Macros can reference other macros defined earlier in the config
+- Macro substitution is deterministic and order-dependent
+- Single-pass substitution prevents circular dependencies
+- Use `yaml.Node` from `gopkg.in/yaml.v3` to preserve macro definition order
+- All existing tests pass
+- New tests validate substitution order and self-reference detection
+
+## Design Requirements
+
+### 1. YAML Parsing Strategy
+- **Continue using:** `gopkg.in/yaml.v3` (current library)
+- **Use:** `yaml.Node` for ordered parsing of macros
+- **Reason:** `yaml.Node` preserves document structure and order, avoiding need for migration
+
+### 2. Data Structure Changes
+
+#### Current Implementation (config.go:19)
+```go
+type MacroList map[string]any
+```
+
+#### New Implementation
+```go
+type MacroList []MacroEntry
+
+type MacroEntry struct {
+    Name  string
+    Value any
+}
+```
+
+**Implementation Note:** Parse macros using `yaml.Node` to extract key-value pairs in document order, then construct the ordered `MacroList`.
+
+### 3. Macro Substitution Order Rules
+
+The substitution must follow this hierarchy (from most specific to least):
+
+1. **Reserved macros** (last): `PORT`, `MODEL_ID` - substituted last, highest priority
+2. **Model-level macros** (middle): Defined in specific model config, overrides global
+3. **Global macros** (first): Defined at config root level
+
+Within each level, macros are substituted in **reverse definition order** (LIFO):
+- The last macro defined is substituted first
+- This allows later macros to reference earlier ones
+- Single-pass substitution prevents circular dependencies
+
+### 4. Macro Reference Rules
+
+**Allowed:**
+- Macro can reference any macro defined **before** it (earlier in the file)
+- Model macros can reference global macros
+- Macros can reference reserved macros (`${PORT}`, `${MODEL_ID}`)
+
+**Prohibited:**
+- Macro cannot reference itself (e.g., `foo: "value ${foo}"`)
+- Macro cannot reference macros defined **after** it
+- No circular references (prevented by single-pass, ordered substitution)
+
+### 5. Validation Requirements
+
+Add validation to detect:
+- **Self-references:** Macro value contains reference to its own name
+- **Unknown macros:** After substitution, any remaining `${...}` references
+
+Error messages should be clear:
+```
+macro 'foo' contains self-reference
+unknown macro '${bar}' in model.cmd
+```
+
+### 6. Implementation Changes
+
+#### Files to Modify
+
+1. **[proxy/config/config.go](proxy/config/config.go)**
+   - Line 19: Change `MacroList` type definition
+   - Line 69: Update `Macros MacroList` field
+   - Line 153-157: Update macro validation loop to work with ordered structure
+   - Line 175-188: Update model-level macro validation
+   - Line 181-188: **NEW** Implement proper macro merging respecting order
+   - Line 193-202: **NEW** Implement ordered macro substitution in LIFO order
+   - Line 389-415: Update `validateMacro` to detect self-references
+   - Line 420-475: Update `substituteMetadataMacros` to accept ordered MacroList
+
+2. **[proxy/config/model_config.go](proxy/config/model_config.go)**
+   - Line 33: Update `Macros MacroList` field type
+
+3. **All test files**
+   - Update test fixtures to use ordered macro definitions
+   - Ensure tests specify macro order explicitly
+
+#### Core Algorithm
+
+Replace the macro substitution logic in [config.go:181-252](proxy/config/config.go#L181-L252) with:
+
+```go
+// Merge global config and model macros. Model macros take precedence
+mergedMacros := make(MacroList, 0, len(config.Macros)+len(modelConfig.Macros)+2)
+
+// Add global macros first
+for _, entry := range config.Macros {
+	mergedMacros = append(mergedMacros, entry)
+}
+
+// Add model macros (can override global)
+for _, entry := range modelConfig.Macros {
+	// Remove any existing global macro with same name
+	found := false
+	for i, existing := range mergedMacros {
+		if existing.Name == entry.Name {
+			mergedMacros[i] = entry // Override
+			found = true
+			break
+		}
+	}
+	if !found {
+		mergedMacros = append(mergedMacros, entry)
+	}
+}
+
+// Add reserved MODEL_ID macro at the end
+mergedMacros = append(mergedMacros, MacroEntry{Name: "MODEL_ID", Value: modelId})
+
+// Check if PORT macro is needed
+if strings.Contains(modelConfig.Cmd, "${PORT}") || strings.Contains(modelConfig.Proxy, "${PORT}") || strings.Contains(modelConfig.CmdStop, "${PORT}") {
+	// enforce ${PORT} used in both cmd and proxy
+	if !strings.Contains(modelConfig.Cmd, "${PORT}") && strings.Contains(modelConfig.Proxy, "${PORT}") {
+		return Config{}, fmt.Errorf("model %s: proxy uses ${PORT} but cmd does not - ${PORT} is only available when used in cmd", modelId)
+	}
+
+	// Add PORT macro to the end (highest priority)
+	mergedMacros = append(mergedMacros, MacroEntry{Name: "PORT", Value: nextPort})
+	nextPort++
+}
+
+// Single-pass substitution: Substitute all macros in LIFO order (last defined first)
+// This allows later macros to reference earlier ones
+for i := len(mergedMacros) - 1; i >= 0; i-- {
+	entry := mergedMacros[i]
+	macroSlug := fmt.Sprintf("${%s}", entry.Name)
+	macroStr := fmt.Sprintf("%v", entry.Value)
+
+	// Substitute in command fields
+	modelConfig.Cmd = strings.ReplaceAll(modelConfig.Cmd, macroSlug, macroStr)
+	modelConfig.CmdStop = strings.ReplaceAll(modelConfig.CmdStop, macroSlug, macroStr)
+	modelConfig.Proxy = strings.ReplaceAll(modelConfig.Proxy, macroSlug, macroStr)
+	modelConfig.CheckEndpoint = strings.ReplaceAll(modelConfig.CheckEndpoint, macroSlug, macroStr)
+	modelConfig.Filters.StripParams = strings.ReplaceAll(modelConfig.Filters.StripParams, macroSlug, macroStr)
+
+	// Substitute in metadata (recursive)
+	if len(modelConfig.Metadata) > 0 {
+		var err error
+		modelConfig.Metadata, err = substituteMacroInValue(modelConfig.Metadata, entry.Name, entry.Value)
+		if err != nil {
+			return Config{}, fmt.Errorf("model %s metadata: %s", modelId, err.Error())
+		}
+	}
+}
+```
+
+Add this new helper function to replace `substituteMetadataMacros`:
+
+```go
+// substituteMacroInValue recursively substitutes a single macro in a value structure
+// This is called once per macro, allowing LIFO substitution order
+func substituteMacroInValue(value any, macroName string, macroValue any) (any, error) {
+	macroSlug := fmt.Sprintf("${%s}", macroName)
+	macroStr := fmt.Sprintf("%v", macroValue)
+
+	switch v := value.(type) {
+	case string:
+		// Check if this is a direct macro substitution
+		if v == macroSlug {
+			return macroValue, nil
+		}
+		// Handle string interpolation
+		if strings.Contains(v, macroSlug) {
+			return strings.ReplaceAll(v, macroSlug, macroStr), nil
+		}
+		return v, nil
+
+	case map[string]any:
+		// Recursively process map values
+		newMap := make(map[string]any)
+		for key, val := range v {
+			newVal, err := substituteMacroInValue(val, macroName, macroValue)
+			if err != nil {
+				return nil, err
+			}
+			newMap[key] = newVal
+		}
+		return newMap, nil
+
+	case []any:
+		// Recursively process slice elements
+		newSlice := make([]any, len(v))
+		for i, val := range v {
+			newVal, err := substituteMacroInValue(val, macroName, macroValue)
+			if err != nil {
+				return nil, err
+			}
+			newSlice[i] = newVal
+		}
+		return newSlice, nil
+
+	default:
+		// Return scalar types as-is
+		return value, nil
+	}
+}
+```
+
+### 7. Self-Reference Detection
+
+Add to `validateMacro` function:
+
+```go
+func validateMacro(name string, value any) error {
+    // ... existing validation ...
+
+    // Check for self-reference
+    if str, ok := value.(string); ok {
+        macroSlug := fmt.Sprintf("${%s}", name)
+        if strings.Contains(str, macroSlug) {
+            return fmt.Errorf("macro '%s' contains self-reference", name)
+        }
+    }
+
+    return nil
+}
+```
+
+## Testing Plan
+
+### 1. Migration Tests
+- **Test:** All existing macro tests still pass after YAML library migration
+- **Files:** All `*_test.go` files with macro tests
+
+### 2. Macro Order Tests
+
+#### Test: Macro-in-macro substitution order
+```yaml
+macros:
+  "A": "value-A"
+  "B": "prefix-${A}-suffix"
+
+models:
+  test:
+    cmd: "echo ${B}"
+```
+**Expected:** `cmd` becomes `"echo prefix-value-A-suffix"`
+
+#### Test: LIFO substitution order
+```yaml
+macros:
+  "base": "/models"
+  "path": "${base}/llama"
+  "full": "${path}/model.gguf"
+
+models:
+  test:
+    cmd: "load ${full}"
+```
+**Expected:** `cmd` becomes `"load /models/llama/model.gguf"`
+
+#### Test: Model macro overrides global
+```yaml
+macros:
+  "tag": "global"
+  "msg": "value-${tag}"
+
+models:
+  test:
+    macros:
+      "tag": "model-level"
+    cmd: "echo ${msg}"
+```
+**Expected:** `cmd` becomes `"echo value-model-level"` (model macro overrides global)
+
+### 3. Reserved Macro Tests
+
+#### Test: MODEL_ID substituted in macro
+```yaml
+macros:
+  "podman-llama": "podman run --name ${MODEL_ID} ghcr.io/ggml-org/llama.cpp:server-cuda"
+
+models:
+  my-model:
+    cmd: "${podman-llama} -m model.gguf"
+```
+**Expected:** `cmd` becomes `"podman run --name my-model ghcr.io/ggml-org/llama.cpp:server-cuda -m model.gguf"`
+
+### 4. Error Detection Tests
+
+#### Test: Self-reference detection
+```yaml
+macros:
+  "recursive": "value-${recursive}"
+```
+**Expected:** Error: `macro 'recursive' contains self-reference`
+
+#### Test: Undefined macro reference
+```yaml
+macros:
+  "A": "value-${UNDEFINED}"
+```
+**Expected:** Error: `unknown macro '${UNDEFINED}' found in macros.A` (or similar)
+
+### 5. Regression Tests
+- Run all existing macro tests: `TestConfig_MacroReplacement`, `TestConfig_MacroReservedNames`, etc.
+- Ensure all pass without modification (except test fixtures if needed)
+
+## Checklist
+
+### Phase 1: Data Structure Changes
+- [ ] Implement custom `UnmarshalYAML` method for `MacroList` that uses `yaml.Node`
+- [ ] Define new ordered `MacroList` type as `[]MacroEntry`
+- [ ] Update `MacroList` type definition in [config.go](proxy/config/config.go#L19)
+- [ ] Update `Config.Macros` field type in [config.go](proxy/config/config.go#L69)
+- [ ] Update `ModelConfig.Macros` field type in [model_config.go](proxy/config/model_config.go#L33)
+- [ ] Implement helper functions:
+  - [ ] `func (ml MacroList) Get(name string) (any, bool)` - lookup by name
+  - [ ] `func (ml MacroList) Set(name string, value any) MacroList` - add/override entry
+  - [ ] `func (ml MacroList) ToMap() map[string]any` - convert to map if needed
+
+### Phase 2: Macro Validation Updates
+- [ ] Update macro validation loop at [config.go:153-157](proxy/config/config.go#L153-L157)
+- [ ] Update model macro validation at [config.go:175-179](proxy/config/config.go#L175-L179)
+- [ ] Add self-reference detection to `validateMacro` function [config.go:389](proxy/config/config.go#L389)
+- [ ] Test self-reference detection with new test case
+
+### Phase 3: Macro Substitution Algorithm
+- [ ] Implement ordered macro merging (global → model → reserved) at [config.go:181-188](proxy/config/config.go#L181-L188)
+- [ ] Implement single-pass LIFO substitution loop (reverse iteration) at [config.go:193-202](proxy/config/config.go#L193-L202)
+  - [ ] Substitute in all string fields (cmd, cmdStop, proxy, checkEndpoint, stripParams)
+  - [ ] Substitute in metadata within same loop
+- [ ] Ensure `MODEL_ID` is added to merged macros before substitution
+- [ ] Ensure `PORT` is added after port assignment (if needed)
+- [ ] Replace `substituteMetadataMacros` with new `substituteMacroInValue` function that processes one macro at a time [config.go:420](proxy/config/config.go#L420)
+- [ ] Remove old metadata substitution code that was separate from main loop [config.go:245-251](proxy/config/config.go#L245-L251)
+
+### Phase 4: Testing
+- [ ] Run `make test-dev` - fix any static checking errors
+- [ ] Add test: macro-in-macro basic substitution
+- [ ] Add test: LIFO substitution order with 3+ macro levels
+- [ ] Add test: MODEL_ID in global macro used by model
+- [ ] Add test: PORT in global macro used by model
+- [ ] Add test: model macro overrides global macro in substitution
+- [ ] Add test: self-reference detection error
+- [ ] Add test: undefined macro reference error
+- [ ] Verify all existing macro tests pass: `TestConfig_Macro*`
+- [ ] Run `make test-all` - ensure all tests including concurrency tests pass
+
+### Phase 5: Documentation
+- [ ] Update plan status in this file (mark completed)
+- [ ] Update CLAUDE.md if macro behavior needs documentation
+- [ ] Verify no new error messages need user documentation
+
+## Bug Example (Original Issue)
+
+```yaml
+macros:
+  "podman-llama": >
+    podman run --name ${MODEL_ID}
+    --init --rm -p ${PORT}:8080 -v /home/alex/ai/models:/models:z --gpus=all
+    ghcr.io/ggml-org/llama.cpp:server-cuda
+
+  "standard-options": >
+    --no-mmap --jinja
+
+  "kv8": >
+    -fa on -ctk q8_0 -ctv q8_0
+```
+
+**Current Bug:**
+- During macro substitution, if `${MODEL_ID}` is processed before `${podman-llama}`, the `${MODEL_ID}` reference inside `podman-llama` remains unsubstituted
+- Results in error: `unknown macro '${MODEL_ID}' found in model.cmd`
+
+**After Fix:**
+- Macros substituted in LIFO order: `kv8` → `standard-options` → `podman-llama`
+- `MODEL_ID` is a reserved macro, substituted last (after all user macros)
+- `${MODEL_ID}` inside `podman-llama` is correctly replaced with the model name

config.example.yaml

@@ -1,31 +1,295 @@
-# Seconds to wait for llama.cpp to be available to serve requests
+# llama-swap YAML configuration example
-# Default (and minimum): 15 seconds
+# -------------------------------------
-healthCheckTimeout: 60
+#
+# 💡 Tip - Use an LLM with this file!
+# ====================================
+#  This example configuration is written to be LLM friendly. Try
+#  copying this file into an LLM and asking it to explain or generate
+#  sections for you.
+# ====================================
 
+# Usage notes:
+# - Below are all the available configuration options for llama-swap.
+# - Settings noted as "required" must be in your configuration file
+# - Settings noted as "optional" can be omitted
+
+# healthCheckTimeout: number of seconds to wait for a model to be ready to serve requests
+# - optional, default: 120
+# - minimum value is 15 seconds, anything less will be set to this value
+healthCheckTimeout: 500
+
+# logLevel: sets the logging value
+# - optional, default: info
+# - Valid log levels: debug, info, warn, error
+logLevel: info
+
+# metricsMaxInMemory: maximum number of metrics to keep in memory
+# - optional, default: 1000
+# - controls how many metrics are stored in memory before older ones are discarded
+# - useful for limiting memory usage when processing large volumes of metrics
+metricsMaxInMemory: 1000
+
+# startPort: sets the starting port number for the automatic ${PORT} macro.
+# - optional, default: 5800
+# - the ${PORT} macro can be used in model.cmd and model.proxy settings
+# - it is automatically incremented for every model that uses it
+startPort: 10001
+
+# macros: a dictionary of string substitutions
+# - optional, default: empty dictionary
+# - macros are reusable snippets
+# - used in a model's cmd, cmdStop, proxy, checkEndpoint, filters.stripParams
+# - useful for reducing common configuration settings
+# - macro names are strings and must be less than 64 characters
+# - macro names must match the regex ^[a-zA-Z0-9_-]+$
+# - macro names must not be a reserved name: PORT or MODEL_ID
+# - macro values can be numbers, bools, or strings
+# - macros can contain other macros, but they must be defined before they are used
+macros:
+  # Example of a multi-line macro
+  "latest-llama": >
+    /path/to/llama-server/llama-server-ec9e0301
+    --port ${PORT}
+
+  "default_ctx": 4096
+
+  # Example of macro-in-macro usage. macros can contain other macros
+  # but they must be previously declared.
+  "default_args": "--ctx-size ${default_ctx}"
+
+# models: a dictionary of model configurations
+# - required
+# - each key is the model's ID, used in API requests
+# - model settings have default values that are used if they are not defined here
+# - the model's ID is available in the ${MODEL_ID} macro, also available in macros defined above
+# - below are examples of the all the settings a model can have
 models:
+
+  # keys are the model names used in API requests
   "llama":
-    cmd: "models/llama-server-osx --port 8999 -m models/Llama-3.2-1B-Instruct-Q4_K_M.gguf"
+    # macros: a dictionary of string substitutions specific to this model
-    proxy: "http://127.0.0.1:8999"
+    # - optional, default: empty dictionary
+    # - macros defined here override macros defined in the global macros section
+    # - model level macros follow the same rules as global macros
+    macros:
+      "default_ctx": 16384
+      "temp": 0.7
 
-    # list of model name aliases this llama.cpp instance can serve
+    # cmd: the command to run to start the inference server.
-    aliases:
+    # - required
-    - "gpt-4o-mini"
+    # - it is just a string, similar to what you would run on the CLI
-  "qwen":
+    # - using `|` allows for comments in the command, these will be parsed out
-    cmd: "models/llama-server-osx --port 8999 -m models/Qwen2.5-1.5B-Instruct-Q4_K_M.gguf"
+    # - macros can be used within cmd
-    proxy: "http://127.0.0.1:8999"
+    cmd: |
-    aliases:
+      # ${latest-llama} is a macro that is defined above
-    - "gpt-3.5-turbo"
+      ${latest-llama}
+      --model path/to/llama-8B-Q4_K_M.gguf
+      --ctx-size ${default_ctx}
+      --temperature ${temp}
 
-  "simple":
+    # name: a display name for the model
-    # example of setting environment variables
+    # - optional, default: empty string
+    # - if set, it will be used in the v1/models API response
+    # - if not set, it will be omitted in the JSON model record
+    name: "llama 3.1 8B"
+
+    # description: a description for the model
+    # - optional, default: empty string
+    # - if set, it will be used in the v1/models API response
+    # - if not set, it will be omitted in the JSON model record
+    description: "A small but capable model used for quick testing"
+
+    # env: define an array of environment variables to inject into cmd's environment
+    # - optional, default: empty array
+    # - each value is a single string
+    # - in the format: ENV_NAME=value
     env:
-      - "CUDA_VISIBLE_DEVICES=0,1"
+      - "CUDA_VISIBLE_DEVICES=0,1,2"
-      - "env1=hello"
-    cmd: "build/simple-responder --port 8999"
-    proxy: "http://127.0.0.1:8999"
 
-  # don't use this, just for testing if things are broken
+    # proxy: the URL where llama-swap routes API requests
-  "broken":
+    # - optional, default: http://localhost:${PORT}
-    cmd: "models/llama-server-osx --port 8999 -m models/doesnotexist.gguf"
+    # - if you used ${PORT} in cmd this can be omitted
-    proxy: "http://127.0.0.1:8999"
+    # - if you use a custom port in cmd this *must* be set
+    proxy: http://127.0.0.1:8999
 
+    # aliases: alternative model names that this model configuration is used for
+    # - optional, default: empty array
+    # - aliases must be unique globally
+    # - useful for impersonating a specific model
+    aliases:
+      - "gpt-4o-mini"
+      - "gpt-3.5-turbo"
+
+    # checkEndpoint: URL path to check if the server is ready
+    # - optional, default: /health
+    # - endpoint is expected to return an HTTP 200 response
+    # - all requests wait until the endpoint is ready or fails
+    # - use "none" to skip endpoint health checking
+    checkEndpoint: /custom-endpoint
+
+    # ttl: automatically unload the model after ttl seconds
+    # - optional, default: 0
+    # - ttl values must be a value greater than 0
+    # - a value of 0 disables automatic unloading of the model
+    ttl: 60
+
+    # useModelName: override the model name that is sent to upstream server
+    # - optional, default: ""
+    # - useful for when the upstream server expects a specific model name that
+    #   is different from the model's ID
+    useModelName: "qwen:qwq"
+
+    # filters: a dictionary of filter settings
+    # - optional, default: empty dictionary
+    # - only stripParams is currently supported
+    filters:
+      # stripParams: a comma separated list of parameters to remove from the request
+      # - optional, default: ""
+      # - useful for server side enforcement of sampling parameters
+      # - the `model` parameter can never be removed
+      # - can be any JSON key in the request body
+      # - recommended to stick to sampling parameters
+      stripParams: "temperature, top_p, top_k"
+
+    # metadata: a dictionary of arbitrary values that are included in /v1/models
+    # - optional, default: empty dictionary
+    # - while metadata can contains complex types it is recommended to keep it simple
+    # - metadata is only passed through in /v1/models responses
+    metadata:
+      # port will remain an integer
+      port: ${PORT}
+
+      # the ${temp} macro will remain a float
+      temperature: ${temp}
+      note: "The ${MODEL_ID} is running on port ${PORT} temp=${temp}, context=${default_ctx}"
+
+      a_list:
+        - 1
+        - 1.23
+        - "macros are OK in list and dictionary types: ${MODEL_ID}"
+
+      an_obj:
+        a: "1"
+        b: 2
+        # objects can contain complex types with macro substitution
+        # becomes: c: [0.7, false, "model: llama"]
+        c: ["${temp}", false, "model: ${MODEL_ID}"]
+
+    # concurrencyLimit: overrides the allowed number of active parallel requests to a model
+    # - optional, default: 0
+    # - useful for limiting the number of active parallel requests a model can process
+    # - must be set per model
+    # - any number greater than 0 will override the internal default value of 10
+    # - any requests that exceeds the limit will receive an HTTP 429 Too Many Requests response
+    # - recommended to be omitted and the default used
+    concurrencyLimit: 0
+
+  # Unlisted model example:
+  "qwen-unlisted":
+    # unlisted: boolean, true or false
+    # - optional, default: false
+    # - unlisted models do not show up in /v1/models api requests
+    # - can be requested as normal through all apis
+    unlisted: true
+    cmd: llama-server --port ${PORT} -m Llama-3.2-1B-Instruct-Q4_K_M.gguf -ngl 0
+
+  # Docker example:
+  # container runtimes like Docker and Podman can be used reliably with
+  # a combination of cmd, cmdStop, and ${MODEL_ID}
+  "docker-llama":
+    proxy: "http://127.0.0.1:${PORT}"
+    cmd: |
+      docker run --name ${MODEL_ID}
+      --init --rm -p ${PORT}:8080 -v /mnt/nvme/models:/models
+      ghcr.io/ggml-org/llama.cpp:server
+      --model '/models/Qwen2.5-Coder-0.5B-Instruct-Q4_K_M.gguf'
+
+    # cmdStop: command to run to stop the model gracefully
+    # - optional, default: ""
+    # - useful for stopping commands managed by another system
+    # - the upstream's process id is available in the ${PID} macro
+    #
+    # When empty, llama-swap has this default behaviour:
+    # - on POSIX systems: a SIGTERM signal is sent
+    # - on Windows, calls taskkill to stop the process
+    # - processes have 5 seconds to shutdown until forceful termination is attempted
+    cmdStop: docker stop ${MODEL_ID}
+
+# groups: a dictionary of group settings
+# - optional, default: empty dictionary
+# - provides advanced controls over model swapping behaviour
+# - using groups some models can be kept loaded indefinitely, while others are swapped out
+# - model IDs must be defined in the Models section
+# - a model can only be a member of one group
+# - group behaviour is controlled via the `swap`, `exclusive` and `persistent` fields
+# - see issue #109 for details
+#
+# NOTE: the example below uses model names that are not defined above for demonstration purposes
+groups:
+  # group1 works the same as the default behaviour of llama-swap where only one model is allowed
+  # to run a time across the whole llama-swap instance
+  "group1":
+    # swap: controls the model swapping behaviour in within the group
+    # - optional, default: true
+    # - true : only one model is allowed to run at a time
+    # - false: all models can run together, no swapping
+    swap: true
+
+    # exclusive: controls how the group affects other groups
+    # - optional, default: true
+    # - true: causes all other groups to unload when this group runs a model
+    # - false: does not affect other groups
+    exclusive: true
+
+    # members references the models defined above
+    # required
+    members:
+      - "llama"
+      - "qwen-unlisted"
+
+  # Example:
+  # - in group2 all models can run at the same time
+  # - when a different group is loaded it causes all running models in this group to unload
+  "group2":
+    swap: false
+
+    # exclusive: false does not unload other groups when a model in group2 is requested
+    # - the models in group2 will be loaded but will not unload any other groups
+    exclusive: false
+    members:
+      - "docker-llama"
+      - "modelA"
+      - "modelB"
+
+  # Example:
+  # - a persistent group, prevents other groups from unloading it
+  "forever":
+    # persistent: prevents over groups from unloading the models in this group
+    # - optional, default: false
+    # - does not affect individual model behaviour
+    persistent: true
+
+    # set swap/exclusive to false to prevent swapping inside the group
+    # and the unloading of other groups
+    swap: false
+    exclusive: false
+    members:
+      - "forever-modelA"
+      - "forever-modelB"
+      - "forever-modelc"
+
+# hooks: a dictionary of event triggers and actions
+# - optional, default: empty dictionary
+# - the only supported hook is on_startup
+hooks:
+  # on_startup: a dictionary of actions to perform on startup
+  # - optional, default: empty dictionary
+  # - the only supported action is preload
+  on_startup:
+        # preload: a list of model ids to load on startup
+        # - optional, default: empty list
+        # - model names must match keys in the models sections
+        # - when preloading multiple models at once, define a group
+        #   otherwise models will be loaded and swapped out
+    preload:
+      - "llama"

docker/build-container.sh

@@ -0,0 +1,55 @@
+#!/bin/bash
+
+cd $(dirname "$0")
+
+ARCH=$1
+PUSH_IMAGES=${2:-false}
+
+# List of allowed architectures
+ALLOWED_ARCHS=("intel" "vulkan" "musa" "cuda" "cpu")
+
+# Check if ARCH is in the allowed list
+if [[ ! " ${ALLOWED_ARCHS[@]} " =~ " ${ARCH} " ]]; then
+  echo "Error: ARCH must be one of the following: ${ALLOWED_ARCHS[@]}"
+  exit 1
+fi
+
+# Check if GITHUB_TOKEN is set and not empty
+if [[ -z "$GITHUB_TOKEN" ]]; then
+  echo "Error: GITHUB_TOKEN is not set or is empty."
+  exit 1
+fi
+
+# the most recent llama-swap tag
+# have to strip out the 'v' due to .tar.gz file naming
+LS_VER=$(curl -s https://api.github.com/repos/mostlygeek/llama-swap/releases/latest | jq -r .tag_name | sed 's/v//')
+
+if [ "$ARCH" == "cpu" ]; then
+    # cpu only containers just use the latest available
+    CONTAINER_LATEST="ghcr.io/mostlygeek/llama-swap:cpu"
+    echo "Building ${CONTAINER_LATEST} $LS_VER"
+    docker build -f llama-swap.Containerfile --build-arg BASE_TAG=server --build-arg LS_VER=${LS_VER} -t ${CONTAINER_LATEST} .
+    if [ "$PUSH_IMAGES" == "true" ]; then
+      docker push ${CONTAINER_LATEST}
+    fi
+else
+    LCPP_TAG=$(curl -s -H "Authorization: Bearer $GITHUB_TOKEN" \
+        "https://api.github.com/users/ggml-org/packages/container/llama.cpp/versions" \
+        | jq -r --arg arch "$ARCH" '.[] | select(.metadata.container.tags[] | startswith("server-\($arch)")) | .metadata.container.tags[]' \
+        | sort -r | head -n1 | awk -F '-' '{print $3}')
+
+    # Abort if LCPP_TAG is empty.
+    if [[ -z "$LCPP_TAG" ]]; then
+        echo "Abort: Could not find llama-server container for arch: $ARCH"
+        exit 1
+    fi
+
+    CONTAINER_TAG="ghcr.io/mostlygeek/llama-swap:v${LS_VER}-${ARCH}-${LCPP_TAG}"
+    CONTAINER_LATEST="ghcr.io/mostlygeek/llama-swap:${ARCH}"
+    echo "Building ${CONTAINER_TAG} $LS_VER"
+    docker build -f llama-swap.Containerfile --build-arg BASE_TAG=server-${ARCH}-${LCPP_TAG} --build-arg LS_VER=${LS_VER} -t ${CONTAINER_TAG} -t ${CONTAINER_LATEST} .
+    if [ "$PUSH_IMAGES" == "true" ]; then
+      docker push ${CONTAINER_TAG}
+      docker push ${CONTAINER_LATEST}
+    fi
+fi

docker/config.example.yaml

@@ -0,0 +1,18 @@
+healthCheckTimeout: 300
+logRequests: true
+metricsMaxInMemory: 1000
+
+models:
+  "qwen2.5":
+    proxy: "http://127.0.0.1:9999"
+    cmd: >
+      /app/llama-server
+      -hf bartowski/Qwen2.5-0.5B-Instruct-GGUF:Q4_K_M
+      --port 9999
+
+  "smollm2":
+    proxy: "http://127.0.0.1:9999"
+    cmd: >
+      /app/llama-server
+      -hf bartowski/SmolLM2-135M-Instruct-GGUF:Q4_K_M
+      --port 9999

docker/llama-swap.Containerfile

@@ -0,0 +1,16 @@
+ARG BASE_TAG=server-cuda
+FROM ghcr.io/ggml-org/llama.cpp:${BASE_TAG}
+
+# has to be after the FROM
+ARG LS_VER=89
+
+WORKDIR /app
+RUN \
+    curl -LO https://github.com/mostlygeek/llama-swap/releases/download/v"${LS_VER}"/llama-swap_"${LS_VER}"_linux_amd64.tar.gz && \
+    tar -zxf llama-swap_"${LS_VER}"_linux_amd64.tar.gz && \
+    rm llama-swap_"${LS_VER}"_linux_amd64.tar.gz
+
+COPY config.example.yaml /app/config.yaml
+
+HEALTHCHECK CMD curl -f http://localhost:8080/ || exit 1
+ENTRYPOINT [ "/app/llama-swap", "-config", "/app/config.yaml" ]

event/README.md

@@ -0,0 +1,3 @@
+The code in `event` was originally a part of https://github.com/kelindar/event (v1.5.2)
+
+The original code uses a `time.Ticker` to process the event queue which caused a large increase in CPU usage ([#189](https://github.com/mostlygeek/llama-swap/issues/189)). This code was ported to remove the ticker and instead be more event driven.

event/default.go

@@ -0,0 +1,30 @@
+// Copyright (c) Roman Atachiants and contributore. All rights reserved.
+// Licensed under the MIT license. See LICENSE file in the project root for detaile.
+
+package event
+
+import (
+	"context"
+)
+
+// Default initializes a default in-process dispatcher
+var Default = NewDispatcherConfig(25000)
+
+// On subscribes to an event, the type of the event will be automatically
+// inferred from the provided type. Must be constant for this to work. This
+// functions same way as Subscribe() but uses the default dispatcher instead.
+func On[T Event](handler func(T)) context.CancelFunc {
+	return Subscribe(Default, handler)
+}
+
+// OnType subscribes to an event with the specified event type. This functions
+// same way as SubscribeTo() but uses the default dispatcher instead.
+func OnType[T Event](eventType uint32, handler func(T)) context.CancelFunc {
+	return SubscribeTo(Default, eventType, handler)
+}
+
+// Emit writes an event into the dispatcher. This functions same way as
+// Publish() but uses the default dispatcher instead.
+func Emit[T Event](ev T) {
+	Publish(Default, ev)
+}

event/default_test.go

@@ -0,0 +1,54 @@
+// Copyright (c) Roman Atachiants and contributore. All rights reserved.
+// Licensed under the MIT license. See LICENSE file in the project root for detaile.
+
+package event
+
+import (
+	"sync"
+	"sync/atomic"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+)
+
+/*
+cpu: 13th Gen Intel(R) Core(TM) i7-13700K
+BenchmarkSubcribeConcurrent-24    	 1826686	       606.3 ns/op	    1648 B/op	       5 allocs/op
+*/
+func BenchmarkSubscribeConcurrent(b *testing.B) {
+	d := NewDispatcher()
+	b.ReportAllocs()
+	b.ResetTimer()
+
+	b.RunParallel(func(pb *testing.PB) {
+		for pb.Next() {
+			unsub := Subscribe(d, func(ev MyEvent1) {})
+			unsub()
+		}
+	})
+}
+
+func TestDefaultPublish(t *testing.T) {
+	var wg sync.WaitGroup
+
+	// Subscribe
+	var count int64
+	defer On(func(ev MyEvent1) {
+		atomic.AddInt64(&count, 1)
+		wg.Done()
+	})()
+
+	defer OnType(TypeEvent1, func(ev MyEvent1) {
+		atomic.AddInt64(&count, 1)
+		wg.Done()
+	})()
+
+	// Publish
+	wg.Add(4)
+	Emit(MyEvent1{})
+	Emit(MyEvent1{})
+
+	// Wait and check
+	wg.Wait()
+	assert.Equal(t, int64(4), count)
+}

event/event.go

@@ -0,0 +1,324 @@
+// Copyright (c) Roman Atachiants and contributors. All rights reserved.
+// Licensed under the MIT license. See LICENSE file in the project root for details.
+
+package event
+
+import (
+	"context"
+	"fmt"
+	"reflect"
+	"sort"
+	"strings"
+	"sync"
+	"sync/atomic"
+)
+
+// Event represents an event contract
+type Event interface {
+	Type() uint32
+}
+
+// registry holds an immutable sorted array of event mappings
+type registry struct {
+	keys []uint32 // Event types (sorted)
+	grps []any    // Corresponding subscribers
+}
+
+// ------------------------------------- Dispatcher -------------------------------------
+
+// Dispatcher represents an event dispatcher.
+type Dispatcher struct {
+	subs     atomic.Pointer[registry] // Atomic pointer to immutable array
+	done     chan struct{}            // Cancellation
+	maxQueue int                      // Maximum queue size per consumer
+	mu       sync.Mutex               // Only for writes (subscribe/unsubscribe)
+}
+
+// NewDispatcher creates a new dispatcher of events.
+func NewDispatcher() *Dispatcher {
+	return NewDispatcherConfig(50000)
+}
+
+// NewDispatcherConfig creates a new dispatcher with configurable max queue size
+func NewDispatcherConfig(maxQueue int) *Dispatcher {
+	d := &Dispatcher{
+		done:     make(chan struct{}),
+		maxQueue: maxQueue,
+	}
+
+	d.subs.Store(&registry{
+		keys: make([]uint32, 0, 16),
+		grps: make([]any, 0, 16),
+	})
+	return d
+}
+
+// Close closes the dispatcher
+func (d *Dispatcher) Close() error {
+	close(d.done)
+	return nil
+}
+
+// isClosed returns whether the dispatcher is closed or not
+func (d *Dispatcher) isClosed() bool {
+	select {
+	case <-d.done:
+		return true
+	default:
+		return false
+	}
+}
+
+// findGroup performs a lock-free binary search for the event type
+func (d *Dispatcher) findGroup(eventType uint32) any {
+	reg := d.subs.Load()
+	keys := reg.keys
+
+	// Inlined binary search for better cache locality
+	left, right := 0, len(keys)
+	for left < right {
+		mid := left + (right-left)/2
+		if keys[mid] < eventType {
+			left = mid + 1
+		} else {
+			right = mid
+		}
+	}
+
+	if left < len(keys) && keys[left] == eventType {
+		return reg.grps[left]
+	}
+	return nil
+}
+
+// Subscribe subscribes to an event, the type of the event will be automatically
+// inferred from the provided type. Must be constant for this to work.
+func Subscribe[T Event](broker *Dispatcher, handler func(T)) context.CancelFunc {
+	var event T
+	return SubscribeTo(broker, event.Type(), handler)
+}
+
+// SubscribeTo subscribes to an event with the specified event type.
+func SubscribeTo[T Event](broker *Dispatcher, eventType uint32, handler func(T)) context.CancelFunc {
+	if broker.isClosed() {
+		panic(errClosed)
+	}
+
+	broker.mu.Lock()
+	defer broker.mu.Unlock()
+
+	// Check if group already exists
+	if existing := broker.findGroup(eventType); existing != nil {
+		grp := groupOf[T](eventType, existing)
+		sub := grp.Add(handler)
+		return func() {
+			grp.Del(sub)
+		}
+	}
+
+	// Create new group
+	grp := &group[T]{cond: sync.NewCond(new(sync.Mutex)), maxQueue: broker.maxQueue}
+	sub := grp.Add(handler)
+
+	// Copy-on-write: insert new entry in sorted position
+	old := broker.subs.Load()
+	idx := sort.Search(len(old.keys), func(i int) bool {
+		return old.keys[i] >= eventType
+	})
+
+	// Create new arrays with space for one more element
+	newKeys := make([]uint32, len(old.keys)+1)
+	newGrps := make([]any, len(old.grps)+1)
+
+	// Copy elements before insertion point
+	copy(newKeys[:idx], old.keys[:idx])
+	copy(newGrps[:idx], old.grps[:idx])
+
+	// Insert new element
+	newKeys[idx] = eventType
+	newGrps[idx] = grp
+
+	// Copy elements after insertion point
+	copy(newKeys[idx+1:], old.keys[idx:])
+	copy(newGrps[idx+1:], old.grps[idx:])
+
+	// Atomically store the new registry (mutex ensures no concurrent writers)
+	newReg := &registry{keys: newKeys, grps: newGrps}
+	broker.subs.Store(newReg)
+
+	return func() {
+		grp.Del(sub)
+	}
+}
+
+// Publish writes an event into the dispatcher
+func Publish[T Event](broker *Dispatcher, ev T) {
+	eventType := ev.Type()
+	if sub := broker.findGroup(eventType); sub != nil {
+		group := groupOf[T](eventType, sub)
+		group.Broadcast(ev)
+	}
+}
+
+// Count counts the number of subscribers, this is for testing only.
+func (d *Dispatcher) count(eventType uint32) int {
+	if group := d.findGroup(eventType); group != nil {
+		return group.(interface{ Count() int }).Count()
+	}
+	return 0
+}
+
+// groupOf casts the subscriber group to the specified generic type
+func groupOf[T Event](eventType uint32, subs any) *group[T] {
+	if group, ok := subs.(*group[T]); ok {
+		return group
+	}
+
+	panic(errConflict[T](eventType, subs))
+}
+
+// ------------------------------------- Subscriber -------------------------------------
+
+// consumer represents a consumer with a message queue
+type consumer[T Event] struct {
+	queue []T  // Current work queue
+	stop  bool // Stop signal
+}
+
+// Listen listens to the event queue and processes events
+func (s *consumer[T]) Listen(c *sync.Cond, fn func(T)) {
+	pending := make([]T, 0, 128)
+
+	for {
+		c.L.Lock()
+		for len(s.queue) == 0 {
+			switch {
+			case s.stop:
+				c.L.Unlock()
+				return
+			default:
+				c.Wait()
+			}
+		}
+
+		// Swap buffers and reset the current queue
+		temp := s.queue
+		s.queue = pending[:0]
+		pending = temp
+		c.L.Unlock()
+
+		// Outside of the critical section, process the work
+		for _, event := range pending {
+			fn(event)
+		}
+
+		// Notify potential publishers waiting due to backpressure
+		c.Broadcast()
+	}
+}
+
+// ------------------------------------- Subscriber Group -------------------------------------
+
+// group represents a consumer group
+type group[T Event] struct {
+	cond     *sync.Cond
+	subs     []*consumer[T]
+	maxQueue int // Maximum queue size per consumer
+	maxLen   int // Current maximum queue length across all consumers
+}
+
+// Broadcast sends an event to all consumers
+func (s *group[T]) Broadcast(ev T) {
+	s.cond.L.Lock()
+	defer s.cond.L.Unlock()
+
+	// Calculate current maximum queue length
+	s.maxLen = 0
+	for _, sub := range s.subs {
+		if len(sub.queue) > s.maxLen {
+			s.maxLen = len(sub.queue)
+		}
+	}
+
+	// Backpressure: wait if queues are full
+	for s.maxLen >= s.maxQueue {
+		s.cond.Wait()
+
+		// Recalculate after wakeup
+		s.maxLen = 0
+		for _, sub := range s.subs {
+			if len(sub.queue) > s.maxLen {
+				s.maxLen = len(sub.queue)
+			}
+		}
+	}
+
+	// Add event to all queues and track new maximum
+	newMax := 0
+	for _, sub := range s.subs {
+		sub.queue = append(sub.queue, ev)
+		if len(sub.queue) > newMax {
+			newMax = len(sub.queue)
+		}
+	}
+	s.maxLen = newMax
+	s.cond.Broadcast() // Wake consumers
+}
+
+// Add adds a subscriber to the list
+func (s *group[T]) Add(handler func(T)) *consumer[T] {
+	sub := &consumer[T]{
+		queue: make([]T, 0, 64),
+	}
+
+	// Add the consumer to the list of active consumers
+	s.cond.L.Lock()
+	s.subs = append(s.subs, sub)
+	s.cond.L.Unlock()
+
+	// Start listening
+	go sub.Listen(s.cond, handler)
+	return sub
+}
+
+// Del removes a subscriber from the list
+func (s *group[T]) Del(sub *consumer[T]) {
+	s.cond.L.Lock()
+	defer s.cond.L.Unlock()
+
+	// Search and remove the subscriber
+	sub.stop = true
+	for i, v := range s.subs {
+		if v == sub {
+			copy(s.subs[i:], s.subs[i+1:])
+			s.subs = s.subs[:len(s.subs)-1]
+			break
+		}
+	}
+}
+
+// ------------------------------------- Debugging -------------------------------------
+
+var errClosed = fmt.Errorf("event dispatcher is closed")
+
+// Count returns the number of subscribers in this group
+func (s *group[T]) Count() int {
+	return len(s.subs)
+}
+
+// String returns string representation of the type
+func (s *group[T]) String() string {
+	typ := reflect.TypeOf(s).String()
+	idx := strings.LastIndex(typ, "/")
+	typ = typ[idx+1 : len(typ)-1]
+	return typ
+}
+
+// errConflict returns a conflict message
+func errConflict[T any](eventType uint32, existing any) string {
+	var want T
+	return fmt.Sprintf(
+		"conflicting event type, want=<%T>, registered=<%s>, event=0x%v",
+		want, existing, eventType,
+	)
+}

event/event_test.go

@@ -0,0 +1,324 @@
+// Copyright (c) Roman Atachiants and contributore. All rights reserved.
+// Licensed under the MIT license. See LICENSE file in the project root for detaile.
+
+package event
+
+import (
+	"fmt"
+	"sync"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/assert"
+)
+
+func TestPublish(t *testing.T) {
+	d := NewDispatcher()
+	var wg sync.WaitGroup
+
+	// Subscribe, must be received in order
+	var count int64
+	defer Subscribe(d, func(ev MyEvent1) {
+		assert.Equal(t, int(atomic.AddInt64(&count, 1)), ev.Number)
+		wg.Done()
+	})()
+
+	// Publish
+	wg.Add(3)
+	Publish(d, MyEvent1{Number: 1})
+	Publish(d, MyEvent1{Number: 2})
+	Publish(d, MyEvent1{Number: 3})
+
+	// Wait and check
+	wg.Wait()
+	assert.Equal(t, int64(3), count)
+}
+
+func TestUnsubscribe(t *testing.T) {
+	d := NewDispatcher()
+	assert.Equal(t, 0, d.count(TypeEvent1))
+	unsubscribe := Subscribe(d, func(ev MyEvent1) {
+		// Nothing
+	})
+
+	assert.Equal(t, 1, d.count(TypeEvent1))
+	unsubscribe()
+	assert.Equal(t, 0, d.count(TypeEvent1))
+}
+
+func TestConcurrent(t *testing.T) {
+	const max = 1000000
+	var count int64
+	var wg sync.WaitGroup
+	wg.Add(1)
+
+	d := NewDispatcher()
+	defer Subscribe(d, func(ev MyEvent1) {
+		if current := atomic.AddInt64(&count, 1); current == max {
+			wg.Done()
+		}
+	})()
+
+	// Asynchronously publish
+	go func() {
+		for i := 0; i < max; i++ {
+			Publish(d, MyEvent1{})
+		}
+	}()
+
+	defer Subscribe(d, func(ev MyEvent1) {
+		// Subscriber that does nothing
+	})()
+
+	wg.Wait()
+	assert.Equal(t, max, int(count))
+}
+
+func TestSubscribeDifferentType(t *testing.T) {
+	d := NewDispatcher()
+	assert.Panics(t, func() {
+		SubscribeTo(d, TypeEvent1, func(ev MyEvent1) {})
+		SubscribeTo(d, TypeEvent1, func(ev MyEvent2) {})
+	})
+}
+
+func TestPublishDifferentType(t *testing.T) {
+	d := NewDispatcher()
+	assert.Panics(t, func() {
+		SubscribeTo(d, TypeEvent1, func(ev MyEvent2) {})
+		Publish(d, MyEvent1{})
+	})
+}
+
+func TestCloseDispatcher(t *testing.T) {
+	d := NewDispatcher()
+	defer SubscribeTo(d, TypeEvent1, func(ev MyEvent2) {})()
+
+	assert.NoError(t, d.Close())
+	assert.Panics(t, func() {
+		SubscribeTo(d, TypeEvent1, func(ev MyEvent2) {})
+	})
+}
+
+func TestMatrix(t *testing.T) {
+	const amount = 1000
+	for _, subs := range []int{1, 10, 100} {
+		for _, topics := range []int{1, 10} {
+			expected := subs * topics * amount
+			t.Run(fmt.Sprintf("%dx%d", topics, subs), func(t *testing.T) {
+				var count atomic.Int64
+				var wg sync.WaitGroup
+				wg.Add(expected)
+
+				d := NewDispatcher()
+				for i := 0; i < subs; i++ {
+					for id := 0; id < topics; id++ {
+						defer SubscribeTo(d, uint32(id), func(ev MyEvent3) {
+							count.Add(1)
+							wg.Done()
+						})()
+					}
+				}
+
+				for n := 0; n < amount; n++ {
+					for id := 0; id < topics; id++ {
+						go Publish(d, MyEvent3{ID: id})
+					}
+				}
+
+				wg.Wait()
+				assert.Equal(t, expected, int(count.Load()))
+			})
+		}
+	}
+}
+
+func TestConcurrentSubscriptionRace(t *testing.T) {
+	// This test specifically targets the race condition that occurs when multiple
+	// goroutines try to subscribe to different event types simultaneously.
+	// Without the CAS loop, subscriptions could be lost due to registry corruption.
+
+	const numGoroutines = 100
+	const numEventTypes = 50
+
+	d := NewDispatcher()
+	defer d.Close()
+
+	var wg sync.WaitGroup
+	var receivedCount int64
+	var subscribedTypes sync.Map // Thread-safe map
+
+	wg.Add(numGoroutines)
+
+	// Start multiple goroutines that subscribe to different event types concurrently
+	for i := 0; i < numGoroutines; i++ {
+		go func(goroutineID int) {
+			defer wg.Done()
+
+			// Each goroutine subscribes to a unique event type
+			eventType := uint32(goroutineID%numEventTypes + 1000) // Offset to avoid collision with other tests
+
+			// Subscribe to the event type
+			SubscribeTo(d, eventType, func(ev MyEvent3) {
+				atomic.AddInt64(&receivedCount, 1)
+			})
+
+			// Record that this type was subscribed
+			subscribedTypes.Store(eventType, true)
+		}(i)
+	}
+
+	// Wait for all subscriptions to complete
+	wg.Wait()
+
+	// Count the number of unique event types subscribed
+	expectedTypes := 0
+	subscribedTypes.Range(func(key, value interface{}) bool {
+		expectedTypes++
+		return true
+	})
+
+	// Small delay to ensure all subscriptions are fully processed
+	time.Sleep(10 * time.Millisecond)
+
+	// Publish events to each subscribed type
+	subscribedTypes.Range(func(key, value interface{}) bool {
+		eventType := key.(uint32)
+		Publish(d, MyEvent3{ID: int(eventType)})
+		return true
+	})
+
+	// Wait for all events to be processed
+	time.Sleep(50 * time.Millisecond)
+
+	// Verify that we received at least the expected number of events
+	// (there might be more if multiple goroutines subscribed to the same event type)
+	received := atomic.LoadInt64(&receivedCount)
+	assert.GreaterOrEqual(t, int(received), expectedTypes,
+		"Should have received at least %d events, got %d", expectedTypes, received)
+
+	// Verify that we have the expected number of unique event types
+	assert.Equal(t, numEventTypes, expectedTypes,
+		"Should have exactly %d unique event types", numEventTypes)
+}
+
+func TestConcurrentHandlerRegistration(t *testing.T) {
+	const numGoroutines = 100
+
+	// Test concurrent subscriptions to the same event type
+	t.Run("SameEventType", func(t *testing.T) {
+		d := NewDispatcher()
+		var handlerCount int64
+		var wg sync.WaitGroup
+
+		// Start multiple goroutines subscribing to the same event type (0x1)
+		for i := 0; i < numGoroutines; i++ {
+			wg.Add(1)
+			go func() {
+				defer wg.Done()
+				SubscribeTo(d, uint32(0x1), func(ev MyEvent1) {
+					atomic.AddInt64(&handlerCount, 1)
+				})
+			}()
+		}
+
+		wg.Wait()
+
+		// Verify all handlers were registered by publishing an event
+		atomic.StoreInt64(&handlerCount, 0)
+		Publish(d, MyEvent1{})
+
+		// Small delay to ensure all handlers have executed
+		time.Sleep(10 * time.Millisecond)
+
+		assert.Equal(t, int64(numGoroutines), atomic.LoadInt64(&handlerCount),
+			"Not all handlers were registered due to race condition")
+	})
+
+	// Test concurrent subscriptions to different event types
+	t.Run("DifferentEventTypes", func(t *testing.T) {
+		d := NewDispatcher()
+		var wg sync.WaitGroup
+		receivedEvents := make(map[uint32]*int64)
+
+		// Create multiple event types and subscribe concurrently
+		for i := 0; i < numGoroutines; i++ {
+			eventType := uint32(100 + i)
+			counter := new(int64)
+			receivedEvents[eventType] = counter
+
+			wg.Add(1)
+			go func(et uint32, cnt *int64) {
+				defer wg.Done()
+				SubscribeTo(d, et, func(ev MyEvent3) {
+					atomic.AddInt64(cnt, 1)
+				})
+			}(eventType, counter)
+		}
+
+		wg.Wait()
+
+		// Publish events to all types
+		for eventType := uint32(100); eventType < uint32(100+numGoroutines); eventType++ {
+			Publish(d, MyEvent3{ID: int(eventType)})
+		}
+
+		// Small delay to ensure all handlers have executed
+		time.Sleep(10 * time.Millisecond)
+
+		// Verify all event types received their events
+		for eventType, counter := range receivedEvents {
+			assert.Equal(t, int64(1), atomic.LoadInt64(counter),
+				"Event type %d did not receive its event", eventType)
+		}
+	})
+}
+
+func TestBackpressure(t *testing.T) {
+	d := NewDispatcher()
+	d.maxQueue = 10
+
+	var processedCount int64
+	unsub := SubscribeTo(d, uint32(0x200), func(ev MyEvent3) {
+		atomic.AddInt64(&processedCount, 1)
+	})
+	defer unsub()
+
+	const eventsToPublish = 1000
+	for i := 0; i < eventsToPublish; i++ {
+		Publish(d, MyEvent3{ID: 0x200})
+	}
+
+	time.Sleep(100 * time.Millisecond)
+
+	// Verify all events were eventually processed
+	finalProcessed := atomic.LoadInt64(&processedCount)
+	assert.Equal(t, int64(eventsToPublish), finalProcessed)
+	t.Logf("Events processed: %d/%d", finalProcessed, eventsToPublish)
+}
+
+// ------------------------------------- Test Events -------------------------------------
+
+const (
+	TypeEvent1 = 0x1
+	TypeEvent2 = 0x2
+)
+
+type MyEvent1 struct {
+	Number int
+}
+
+func (t MyEvent1) Type() uint32 { return TypeEvent1 }
+
+type MyEvent2 struct {
+	Text string
+}
+
+func (t MyEvent2) Type() uint32 { return TypeEvent2 }
+
+type MyEvent3 struct {
+	ID int
+}
+
+func (t MyEvent3) Type() uint32 { return uint32(t.ID) }

examples/README.md

@@ -0,0 +1,6 @@
+# Example Configs and Use Cases
+
+A collections of usecases and examples for getting the most out of llama-swap.
+
+* [Speculative Decoding](speculative-decoding/README.md) - using a small draft model can increase inference speeds from 20% to 40%. This example includes a configurations Qwen2.5-Coder-32B (2.5x increase) and Llama-3.1-70B (1.4x increase) in the best cases.
+* [Optimizing Code Generation](benchmark-snakegame/README.md) - find the optimal settings for your machine. This example demonstrates defining multiple configurations and testing which one is fastest.

examples/aider-qwq-coder/README.md

@@ -0,0 +1,153 @@
+# aider, QwQ, Qwen-Coder 2.5 and llama-swap
+
+This guide show how to use aider and llama-swap to get a 100% local coding co-pilot setup. The focus is on the trickest part which is configuring aider, llama-swap and llama-server to work together.
+
+## Here's what you you need:
+
+- aider - [installation docs](https://aider.chat/docs/install.html)
+- llama-server - [download latest release](https://github.com/ggml-org/llama.cpp/releases)
+- llama-swap - [download latest release](https://github.com/mostlygeek/llama-swap/releases)
+- [QwQ 32B](https://huggingface.co/bartowski/Qwen_QwQ-32B-GGUF) and [Qwen Coder 2.5 32B](https://huggingface.co/bartowski/Qwen2.5-Coder-32B-Instruct-GGUF) models
+- 24GB VRAM video card
+
+## Running aider
+
+The goal is getting this command line to work:
+
+```sh
+aider --architect \
+    --no-show-model-warnings \
+    --model openai/QwQ \
+    --editor-model openai/qwen-coder-32B \
+    --model-settings-file aider.model.settings.yml \
+    --openai-api-key "sk-na" \
+    --openai-api-base "http://10.0.1.24:8080/v1" \
+```
+
+Set `--openai-api-base` to the IP and port where your llama-swap is running.
+
+## Create an aider model settings file
+
+```yaml
+# aider.model.settings.yml
+
+#
+# !!! important: model names must match llama-swap configuration names !!!
+#
+
+- name: "openai/QwQ"
+  edit_format: diff
+  extra_params:
+    max_tokens: 16384
+    top_p: 0.95
+    top_k: 40
+    presence_penalty: 0.1
+    repetition_penalty: 1
+    num_ctx: 16384
+  use_temperature: 0.6
+  reasoning_tag: think
+  weak_model_name: "openai/qwen-coder-32B"
+  editor_model_name: "openai/qwen-coder-32B"
+
+- name: "openai/qwen-coder-32B"
+  edit_format: diff
+  extra_params:
+    max_tokens: 16384
+    top_p: 0.8
+    top_k: 20
+    repetition_penalty: 1.05
+  use_temperature: 0.6
+  reasoning_tag: think
+  editor_edit_format: editor-diff
+  editor_model_name: "openai/qwen-coder-32B"
+```
+
+## llama-swap configuration
+
+```yaml
+# config.yaml
+
+# The parameters are tweaked to fit model+context into 24GB VRAM GPUs
+models:
+  "qwen-coder-32B":
+    proxy: "http://127.0.0.1:8999"
+    cmd: >
+      /path/to/llama-server
+      --host 127.0.0.1 --port 8999 --flash-attn --slots
+      --ctx-size 16000
+      --cache-type-k q8_0 --cache-type-v q8_0
+       -ngl 99
+      --model /path/to/Qwen2.5-Coder-32B-Instruct-Q4_K_M.gguf
+
+  "QwQ":
+    proxy: "http://127.0.0.1:9503"
+    cmd: >
+      /path/to/llama-server
+      --host 127.0.0.1 --port 9503 --flash-attn --metrics--slots
+      --cache-type-k q8_0 --cache-type-v q8_0
+      --ctx-size 32000
+      --samplers "top_k;top_p;min_p;temperature;dry;typ_p;xtc"
+      --temp 0.6 --repeat-penalty 1.1 --dry-multiplier 0.5
+      --min-p 0.01 --top-k 40 --top-p 0.95
+      -ngl 99
+      --model /mnt/nvme/models/bartowski/Qwen_QwQ-32B-Q4_K_M.gguf
+```
+
+## Advanced, Dual GPU Configuration
+
+If you have _dual 24GB GPUs_ you can use llama-swap profiles to avoid swapping between QwQ and Qwen Coder.
+
+In llama-swap's configuration file:
+
+1. add a `profiles` section with `aider` as the profile name
+2. using the `env` field to specify the GPU IDs for each model
+
+```yaml
+# config.yaml
+
+# Add a profile for aider
+profiles:
+  aider:
+    - qwen-coder-32B
+    - QwQ
+
+models:
+  "qwen-coder-32B":
+    # manually set the GPU to run on
+    env:
+      - "CUDA_VISIBLE_DEVICES=0"
+    proxy: "http://127.0.0.1:8999"
+    cmd: /path/to/llama-server ...
+
+  "QwQ":
+    # manually set the GPU to run on
+    env:
+      - "CUDA_VISIBLE_DEVICES=1"
+    proxy: "http://127.0.0.1:9503"
+    cmd: /path/to/llama-server ...
+```
+
+Append the profile tag, `aider:`, to the model names in the model settings file
+
+```yaml
+# aider.model.settings.yml
+- name: "openai/aider:QwQ"
+  weak_model_name: "openai/aider:qwen-coder-32B-aider"
+  editor_model_name: "openai/aider:qwen-coder-32B-aider"
+
+- name: "openai/aider:qwen-coder-32B"
+  editor_model_name: "openai/aider:qwen-coder-32B-aider"
+```
+
+Run aider with:
+
+```sh
+$ aider --architect \
+    --no-show-model-warnings \
+    --model openai/aider:QwQ \
+    --editor-model openai/aider:qwen-coder-32B \
+    --config aider.conf.yml \
+    --model-settings-file aider.model.settings.yml
+    --openai-api-key "sk-na" \
+    --openai-api-base "http://10.0.1.24:8080/v1"
+```

examples/aider-qwq-coder/aider.model.settings.dualgpu.yml

@@ -0,0 +1,28 @@
+# this makes use of llama-swap's profile feature to
+# keep the architect and editor models in VRAM on different GPUs
+
+- name: "openai/aider:QwQ"
+  edit_format: diff
+  extra_params:
+    max_tokens: 16384
+    top_p: 0.95
+    top_k: 40
+    presence_penalty: 0.1
+    repetition_penalty: 1
+    num_ctx: 16384
+  use_temperature: 0.6
+  reasoning_tag: think
+  weak_model_name: "openai/aider:qwen-coder-32B"
+  editor_model_name: "openai/aider:qwen-coder-32B"
+
+- name: "openai/aider:qwen-coder-32B"
+  edit_format: diff
+  extra_params:
+    max_tokens: 16384
+    top_p: 0.8
+    top_k: 20
+    repetition_penalty: 1.05
+  use_temperature: 0.6
+  reasoning_tag: think
+  editor_edit_format: editor-diff
+  editor_model_name: "openai/aider:qwen-coder-32B"

examples/aider-qwq-coder/aider.model.settings.yml

@@ -0,0 +1,26 @@
+- name: "openai/QwQ"
+  edit_format: diff
+  extra_params:
+    max_tokens: 16384
+    top_p: 0.95
+    top_k: 40
+    presence_penalty: 0.1
+    repetition_penalty: 1
+    num_ctx: 16384
+  use_temperature: 0.6
+  reasoning_tag: think
+  weak_model_name: "openai/qwen-coder-32B"
+  editor_model_name: "openai/qwen-coder-32B"
+
+- name: "openai/qwen-coder-32B"
+  edit_format: diff
+  extra_params:
+    max_tokens: 16384
+    top_p: 0.8
+    top_k: 20
+    repetition_penalty: 1.05
+  use_temperature: 0.6
+  reasoning_tag: think
+  editor_edit_format: editor-diff
+  editor_model_name: "openai/qwen-coder-32B"
+

examples/aider-qwq-coder/llama-swap.yaml

@@ -0,0 +1,49 @@
+healthCheckTimeout: 300
+logLevel: debug
+
+profiles:
+    aider:
+      - qwen-coder-32B
+      - QwQ
+
+models:
+  "qwen-coder-32B":
+    env:
+      - "CUDA_VISIBLE_DEVICES=0"
+    aliases:
+      - coder
+    proxy: "http://127.0.0.1:8999"
+
+    # set appropriate paths for your environment
+    cmd: >
+      /path/to/llama-server
+      --host 127.0.0.1 --port 8999 --flash-attn --slots
+      --ctx-size 16000
+      --ctx-size-draft 16000
+      --model /path/to/Qwen2.5-Coder-32B-Instruct-Q4_K_M.gguf
+      --model-draft /path/to/Qwen2.5-Coder-1.5B-Instruct-Q8_0.gguf
+      -ngl 99 -ngld 99
+      --draft-max 16 --draft-min 4 --draft-p-min 0.4
+      --cache-type-k q8_0 --cache-type-v q8_0
+  "QwQ":
+    env:
+      - "CUDA_VISIBLE_DEVICES=1"
+    proxy: "http://127.0.0.1:9503"
+
+    # set appropriate paths for your environment
+    cmd: >
+      /path/to/llama-server
+      --host 127.0.0.1 --port 9503
+      --flash-attn --metrics
+      --slots
+      --model /path/to/Qwen_QwQ-32B-Q4_K_M.gguf
+      --cache-type-k q8_0 --cache-type-v q8_0
+      --ctx-size 32000
+      --samplers "top_k;top_p;min_p;temperature;dry;typ_p;xtc"
+      --temp 0.6
+      --repeat-penalty 1.1
+      --dry-multiplier 0.5
+      --min-p 0.01
+      --top-k 40
+      --top-p 0.95
+      -ngl 99 -ngld 99

examples/benchmark-snakegame/README.md

@@ -0,0 +1,123 @@
+# Optimizing Code Generation with llama-swap
+
+Finding the best mix of settings for your hardware can be time consuming. This example demonstrates using a custom configuration file to automate testing different scenarios to find the an optimal configuration.
+
+The benchmark writes a snake game in Python, TypeScript, and Swift using the Qwen 2.5 Coder models. The experiments were done using a 3090 and a P40.
+
+**Benchmark Scenarios**
+
+Three scenarios are tested:
+
+- 3090-only: Just the main model on the 3090
+- 3090-with-draft: the main and draft models on the 3090
+- 3090-P40-draft: the main model on the 3090 with the draft model offloaded to the P40
+
+**Available Devices**
+
+Use the following command to list available devices IDs for the configuration:
+
+```
+$ /mnt/nvme/llama-server/llama-server-f3252055 --list-devices
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 4 CUDA devices:
+  Device 0: NVIDIA GeForce RTX 3090, compute capability 8.6, VMM: yes
+  Device 1: Tesla P40, compute capability 6.1, VMM: yes
+  Device 2: Tesla P40, compute capability 6.1, VMM: yes
+  Device 3: Tesla P40, compute capability 6.1, VMM: yes
+Available devices:
+  CUDA0: NVIDIA GeForce RTX 3090 (24154 MiB, 406 MiB free)
+  CUDA1: Tesla P40 (24438 MiB, 22942 MiB free)
+  CUDA2: Tesla P40 (24438 MiB, 24144 MiB free)
+  CUDA3: Tesla P40 (24438 MiB, 24144 MiB free)
+```
+
+**Configuration**
+
+The configuration file, `benchmark-config.yaml`, defines the three scenarios:
+
+```yaml
+models:
+  "3090-only":
+    proxy: "http://127.0.0.1:9503"
+    cmd: >
+      /mnt/nvme/llama-server/llama-server-f3252055
+      --host 127.0.0.1 --port 9503
+      --flash-attn
+      --slots
+
+      --model /mnt/nvme/models/Qwen2.5-Coder-32B-Instruct-Q4_K_M.gguf
+      -ngl 99
+      --device CUDA0
+
+      --ctx-size 32768
+      --cache-type-k q8_0 --cache-type-v q8_0
+
+  "3090-with-draft":
+    proxy: "http://127.0.0.1:9503"
+    # --ctx-size 28500 max that can fit on 3090 after draft model
+    cmd: >
+      /mnt/nvme/llama-server/llama-server-f3252055
+      --host 127.0.0.1 --port 9503
+      --flash-attn
+      --slots
+
+      --model /mnt/nvme/models/Qwen2.5-Coder-32B-Instruct-Q4_K_M.gguf
+      -ngl 99
+      --device CUDA0
+
+      --model-draft /mnt/nvme/models/Qwen2.5-Coder-0.5B-Instruct-Q8_0.gguf
+      -ngld 99
+      --draft-max 16
+      --draft-min 4
+      --draft-p-min 0.4
+      --device-draft CUDA0
+
+      --ctx-size 28500
+      --cache-type-k q8_0 --cache-type-v q8_0
+
+  "3090-P40-draft":
+    proxy: "http://127.0.0.1:9503"
+    cmd: >
+      /mnt/nvme/llama-server/llama-server-f3252055
+      --host 127.0.0.1 --port 9503
+      --flash-attn --metrics
+      --slots
+      --model /mnt/nvme/models/Qwen2.5-Coder-32B-Instruct-Q4_K_M.gguf
+      -ngl 99
+      --device CUDA0
+
+      --model-draft /mnt/nvme/models/Qwen2.5-Coder-0.5B-Instruct-Q8_0.gguf
+      -ngld 99
+      --draft-max 16
+      --draft-min 4
+      --draft-p-min 0.4
+      --device-draft CUDA1
+
+      --ctx-size 32768
+      --cache-type-k q8_0 --cache-type-v q8_0
+```
+
+> Note in the `3090-with-draft` scenario the `--ctx-size` had to be reduced from 32768 to to accommodate the draft model.
+
+
+**Running the Benchmark**
+
+To run the benchmark, execute the following commands:
+
+1. `llama-swap -config benchmark-config.yaml`
+1. `./run-benchmark.sh http://localhost:8080 "3090-only" "3090-with-draft" "3090-P40-draft"`
+
+The [benchmark script](run-benchmark.sh) generates a CSV output of the results, which can be converted to a Markdown table for readability.
+
+**Results (tokens/second)**
+
+| model           | python | typescript | swift |
+|-----------------|--------|------------|-------|
+| 3090-only       | 34.03  | 34.01      | 34.01 |
+| 3090-with-draft | 106.65 | 70.48      | 57.89 |
+| 3090-P40-draft  | 81.54  | 60.35      | 46.50 |
+
+Many different factors, like the programming language, can have big impacts on the performance gains. However, with a custom configuration file for benchmarking it is easy to test the different variations to discover what's best for your hardware.
+
+Happy coding!

examples/benchmark-snakegame/run-benchmark.sh

@@ -0,0 +1,40 @@
+#!/usr/bin/env bash
+
+# This script generates a CSV file showing the token/second for generating a Snake Game in python, typescript and swift
+# It was created to test the effects of speculative decoding and the various draft settings on performance.
+#
+# Writing code with a low temperature seems to provide fairly consistent logic.
+#
+# Usage: ./benchmark.sh <url> <model1> [model2 ...]
+# Example: ./benchmark.sh http://localhost:8080 model1 model2
+
+if [ "$#" -lt 2 ]; then
+    echo "Usage: $0 <url> <model1> [model2 ...]"
+    exit 1
+fi
+
+url=$1; shift
+
+echo "model,python,typescript,swift"
+
+for model in "$@"; do
+
+    echo -n "$model,"
+
+    for lang in "python" "typescript" "swift"; do
+        # expects a llama.cpp after PR https://github.com/ggerganov/llama.cpp/pull/10548
+        # (Dec 3rd/2024)
+        time=$(curl -s --url "$url/v1/chat/completions" -d "{\"messages\": [{\"role\": \"system\", \"content\": \"you only write code.\"}, {\"role\": \"user\", \"content\": \"write snake game in $lang\"}], \"top_k\": 1, \"timings_per_token\":true, \"model\":\"$model\"}" | jq -r .timings.predicted_per_second)
+
+        if [ $? -ne 0 ]; then
+            time="error"
+            exit 1
+        fi
+
+        if [ "$lang" != "swift" ]; then
+            printf "%0.2f tps," $time
+        else
+            printf "%0.2f tps\n" $time
+        fi
+    done
+done

examples/restart-on-config-change/README.md

@@ -0,0 +1,51 @@
+# Restart llama-swap on config change
+
+Sometimes editing the configuration file can take a bit of trail and error to get a model configuration tuned just right. The `watch-and-restart.sh` script can be used to watch `config.yaml` for changes and restart `llama-swap` when it detects a change.
+
+```bash
+#!/bin/bash
+#
+# A simple watch and restart llama-swap when its configuration
+# file changes. Useful for trying out configuration changes
+# without manually restarting the server each time.
+if [ -z "$1" ]; then
+    echo "Usage: $0 <path to config.yaml>"
+    exit 1
+fi
+
+while true; do
+    # Start the process again
+    ./llama-swap-linux-amd64 -config $1 -listen :1867 &
+    PID=$!
+    echo "Started llama-swap with PID $PID"
+
+    # Wait for modifications in the specified directory or file
+    inotifywait -e modify "$1"
+
+    # Check if process exists before sending signal
+    if kill -0 $PID 2>/dev/null; then
+        echo "Sending SIGTERM to $PID"
+        kill -SIGTERM $PID
+        wait $PID
+    else
+        echo "Process $PID no longer exists"
+    fi
+    sleep 1
+done
+```
+
+## Usage and output example
+
+```bash
+$ ./watch-and-restart.sh config.yaml
+Started llama-swap with PID 495455
+Setting up watches.
+Watches established.
+llama-swap listening on :1867
+Sending SIGTERM to 495455
+Shutting down llama-swap
+Started llama-swap with PID 495486
+Setting up watches.
+Watches established.
+llama-swap listening on :1867
+```

examples/speculative-decoding/README.md

@@ -0,0 +1,124 @@
+# Speculative Decoding
+
+Speculative decoding can significantly improve the tokens per second. However, this comes at the cost of increased VRAM usage for the draft model. The examples provided are based on a server with three P40s and one 3090.
+
+## Coding Use Case
+
+This example uses Qwen2.5 Coder 32B with the 0.5B model as a draft. A quantization of Q8_0 was chosen for the draft model, as quantization has a greater impact on smaller models.
+
+The models used are:
+
+* [Bartowski Qwen2.5-Coder-32B-Instruct](https://huggingface.co/bartowski/Qwen2.5-Coder-32B-Instruct-GGUF)
+* [Bartowski Qwen2.5-Coder-0.5B-Instruct](https://huggingface.co/bartowski/Qwen2.5-Coder-0.5B-Instruct-GGUF)
+
+The llama-swap configuration is as follows:
+
+```yaml
+models:
+  "qwen-coder-32b-q4":
+    # main model on 3090, draft on P40 #1
+    cmd: >
+      /mnt/nvme/llama-server/llama-server-be0e35
+      --host 127.0.0.1 --port 9503
+      --flash-attn --metrics
+      --slots
+      --model /mnt/nvme/models/Qwen2.5-Coder-32B-Instruct-Q4_K_M.gguf
+      -ngl 99
+      --ctx-size 19000
+      --model-draft /mnt/nvme/models/Qwen2.5-Coder-0.5B-Instruct-Q8_0.gguf
+      -ngld 99
+      --draft-max 16
+      --draft-min 4
+      --draft-p-min 0.4
+      --device CUDA0
+      --device-draft CUDA1
+    proxy: "http://127.0.0.1:9503"
+```
+
+In this configuration, two GPUs are used: a 3090 (CUDA0) for the main model and a P40 (CUDA1) for the draft model. Although both models can fit on the 3090, relocating the draft model to the P40 freed up space for a larger context size. Despite the P40 being about 1/3rd the speed of the 3090, the small model still improved tokens per second.
+
+Multiple tests were run with various parameters, and the fastest result was chosen for the configuration. In all tests, the 0.5B model produced the largest improvements to tokens per second.
+
+Baseline: 33.92 tokens/second on 3090 without a draft model.
+
+| draft-max | draft-min | draft-p-min | python | TS | swift |
+|-----------|-----------|-------------|--------|----|-------|
+| 16 | 1 | 0.9 | 71.64 | 55.55 | 48.06 |
+| 16 | 1 | 0.4 | 83.21 | 58.55 | 45.50 |
+| 16 | 1 | 0.1 | 79.72 | 55.66 | 43.94 |
+| 16 | 2 | 0.9 | 68.47 | 55.13 | 43.12 |
+| 16 | 2 | 0.4 | 82.82 | 57.42 | 48.83 |
+| 16 | 2 | 0.1 | 81.68 | 51.37 | 45.72 |
+| 16 | 4 | 0.9 | 66.44 | 48.49 | 42.40 |
+| 16 | 4 | 0.4 | _83.62_ (fastest)| _58.29_ | _50.17_ |
+| 16 | 4 | 0.1 | 82.46 | 51.45 | 40.71 |
+| 8 | 1 | 0.4 | 67.07 | 55.17 | 48.46 |
+| 4 | 1 | 0.4 | 50.13 | 44.96 | 40.79 |
+
+The test script can be found in this [gist](https://gist.github.com/mostlygeek/da429769796ac8a111142e75660820f1). It is a simple curl script that prompts generating a snake game in Python, TypeScript, or Swift. Evaluation metrics were pulled from llama.cpp's logs.
+
+```bash
+for lang in "python" "typescript" "swift"; do
+    echo "Generating Snake Game in $lang using $model"
+    curl -s --url http://localhost:8080/v1/chat/completions -d "{\"messages\": [{\"role\": \"system\", \"content\": \"you only write code.\"}, {\"role\": \"user\", \"content\": \"write snake game in $lang\"}], \"temperature\": 0.1, \"model\":\"$model\"}" > /dev/null
+done
+```
+
+Python consistently outperformed Swift in all tests, likely due to the 0.5B draft model being more proficient in generating Python code accepted by the larger 32B model.
+
+## Chat
+
+This configuration is for a regular chat use case. It produces approximately 13 tokens/second in typical use, up from ~9 tokens/second with only the 3xP40s. This is great news for P40 owners.
+
+The models used are:
+
+* [Bartowski Meta-Llama-3.1-70B-Instruct-GGUF](https://huggingface.co/bartowski/Meta-Llama-3.1-70B-Instruct-GGUF)
+* [Bartowski Llama-3.2-3B-Instruct-GGUF](https://huggingface.co/bartowski/Llama-3.2-3B-Instruct-GGUF)
+
+```yaml
+models:
+  "llama-70B":
+    cmd: >
+      /mnt/nvme/llama-server/llama-server-be0e35
+      --host 127.0.0.1 --port 9602
+      --flash-attn --metrics
+      --split-mode row
+      --ctx-size 80000
+      --model /mnt/nvme/models/Meta-Llama-3.1-70B-Instruct-Q4_K_L.gguf
+      -ngl 99
+      --model-draft /mnt/nvme/models/Llama-3.2-3B-Instruct-Q4_K_M.gguf
+      -ngld 99
+      --draft-max 16
+      --draft-min 1
+      --draft-p-min 0.4
+      --device-draft CUDA0
+      --tensor-split 0,1,1,1
+```
+
+In this configuration, Llama-3.1-70B is split across three P40s, and Llama-3.2-3B is on the 3090.
+
+Some flags deserve further explanation:
+
+* `--split-mode row` - increases inference speeds using multiple P40s by about 30%. This is a P40-specific feature.
+* `--tensor-split 0,1,1,1` - controls how the main model is split across the GPUs. This means 0% on the 3090 and an even split across the P40s. A value of `--tensor-split 0,5,4,1` would mean 0% on the 3090, 50%, 40%, and 10% respectively across the other P40s. However, this would exceed the available VRAM.
+* `--ctx-size 80000` - the maximum context size that can fit in the remaining VRAM.
+
+## What is CUDA0, CUDA1, CUDA2, CUDA3?
+
+These devices are the IDs used by llama.cpp.
+
+```bash
+$ ./llama-server --list-devices
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 4 CUDA devices:
+  Device 0: NVIDIA GeForce RTX 3090, compute capability 8.6, VMM: yes
+  Device 1: Tesla P40, compute capability 6.1, VMM: yes
+  Device 2: Tesla P40, compute capability 6.1, VMM: yes
+  Device 3: Tesla P40, compute capability 6.1, VMM: yes
+Available devices:
+  CUDA0: NVIDIA GeForce RTX 3090 (24154 MiB, 23892 MiB free)
+  CUDA1: Tesla P40 (24438 MiB, 24290 MiB free)
+  CUDA2: Tesla P40 (24438 MiB, 24290 MiB free)
+  CUDA3: Tesla P40 (24438 MiB, 24290 MiB free)
+```

go.mod

@@ -2,4 +2,44 @@ module github.com/mostlygeek/llama-swap
 
 go 1.23.0
 
-require gopkg.in/yaml.v3 v3.0.1
+require (
+	github.com/billziss-gh/golib v0.2.0
+	github.com/fsnotify/fsnotify v1.9.0
+	github.com/gin-gonic/gin v1.10.0
+	github.com/stretchr/testify v1.9.0
+	github.com/tidwall/gjson v1.18.0
+	github.com/tidwall/sjson v1.2.5
+	gopkg.in/yaml.v3 v3.0.1
+)
+
+require (
+	github.com/bytedance/sonic v1.11.6 // indirect
+	github.com/bytedance/sonic/loader v0.1.1 // indirect
+	github.com/cloudwego/base64x v0.1.4 // indirect
+	github.com/cloudwego/iasm v0.2.0 // indirect
+	github.com/davecgh/go-spew v1.1.1 // indirect
+	github.com/gabriel-vasile/mimetype v1.4.3 // indirect
+	github.com/gin-contrib/sse v0.1.0 // indirect
+	github.com/go-playground/locales v0.14.1 // indirect
+	github.com/go-playground/universal-translator v0.18.1 // indirect
+	github.com/go-playground/validator/v10 v10.20.0 // indirect
+	github.com/goccy/go-json v0.10.2 // indirect
+	github.com/json-iterator/go v1.1.12 // indirect
+	github.com/klauspost/cpuid/v2 v2.2.7 // indirect
+	github.com/leodido/go-urn v1.4.0 // indirect
+	github.com/mattn/go-isatty v0.0.20 // indirect
+	github.com/modern-go/concurrent v0.0.0-20180306012644-bacd9c7ef1dd // indirect
+	github.com/modern-go/reflect2 v1.0.2 // indirect
+	github.com/pelletier/go-toml/v2 v2.2.2 // indirect
+	github.com/pmezard/go-difflib v1.0.0 // indirect
+	github.com/tidwall/match v1.1.1 // indirect
+	github.com/tidwall/pretty v1.2.1 // indirect
+	github.com/twitchyliquid64/golang-asm v0.15.1 // indirect
+	github.com/ugorji/go/codec v1.2.12 // indirect
+	golang.org/x/arch v0.8.0 // indirect
+	golang.org/x/crypto v0.36.0 // indirect
+	golang.org/x/net v0.38.0 // indirect
+	golang.org/x/sys v0.31.0 // indirect
+	golang.org/x/text v0.23.0 // indirect
+	google.golang.org/protobuf v1.34.1 // indirect
+)

go.sum

@@ -1,4 +1,103 @@
+github.com/billziss-gh/golib v0.2.0 h1:NyvcAQdfvM8xokKkKotiligKjKXzuQD4PPykg1nKc/8=
+github.com/billziss-gh/golib v0.2.0/go.mod h1:mZpUYANXZkDKSnyYbX9gfnyxwe0ddRhUtfXcsD5r8dw=
+github.com/bytedance/sonic v1.11.6 h1:oUp34TzMlL+OY1OUWxHqsdkgC/Zfc85zGqw9siXjrc0=
+github.com/bytedance/sonic v1.11.6/go.mod h1:LysEHSvpvDySVdC2f87zGWf6CIKJcAvqab1ZaiQtds4=
+github.com/bytedance/sonic/loader v0.1.1 h1:c+e5Pt1k/cy5wMveRDyk2X4B9hF4g7an8N3zCYjJFNM=
+github.com/bytedance/sonic/loader v0.1.1/go.mod h1:ncP89zfokxS5LZrJxl5z0UJcsk4M4yY2JpfqGeCtNLU=
+github.com/cloudwego/base64x v0.1.4 h1:jwCgWpFanWmN8xoIUHa2rtzmkd5J2plF/dnLS6Xd/0Y=
+github.com/cloudwego/base64x v0.1.4/go.mod h1:0zlkT4Wn5C6NdauXdJRhSKRlJvmclQ1hhJgA0rcu/8w=
+github.com/cloudwego/iasm v0.2.0 h1:1KNIy1I1H9hNNFEEH3DVnI4UujN+1zjpuk6gwHLTssg=
+github.com/cloudwego/iasm v0.2.0/go.mod h1:8rXZaNYT2n95jn+zTI1sDr+IgcD2GVs0nlbbQPiEFhY=
+github.com/davecgh/go-spew v1.1.0/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
+github.com/davecgh/go-spew v1.1.1 h1:vj9j/u1bqnvCEfJOwUhtlOARqs3+rkHYY13jYWTU97c=
+github.com/davecgh/go-spew v1.1.1/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
+github.com/fsnotify/fsnotify v1.9.0 h1:2Ml+OJNzbYCTzsxtv8vKSFD9PbJjmhYF14k/jKC7S9k=
+github.com/fsnotify/fsnotify v1.9.0/go.mod h1:8jBTzvmWwFyi3Pb8djgCCO5IBqzKJ/Jwo8TRcHyHii0=
+github.com/gabriel-vasile/mimetype v1.4.3 h1:in2uUcidCuFcDKtdcBxlR0rJ1+fsokWf+uqxgUFjbI0=
+github.com/gabriel-vasile/mimetype v1.4.3/go.mod h1:d8uq/6HKRL6CGdk+aubisF/M5GcPfT7nKyLpA0lbSSk=
+github.com/gin-contrib/sse v0.1.0 h1:Y/yl/+YNO8GZSjAhjMsSuLt29uWRFHdHYUb5lYOV9qE=
+github.com/gin-contrib/sse v0.1.0/go.mod h1:RHrZQHXnP2xjPF+u1gW/2HnVO7nvIa9PG3Gm+fLHvGI=
+github.com/gin-gonic/gin v1.10.0 h1:nTuyha1TYqgedzytsKYqna+DfLos46nTv2ygFy86HFU=
+github.com/gin-gonic/gin v1.10.0/go.mod h1:4PMNQiOhvDRa013RKVbsiNwoyezlm2rm0uX/T7kzp5Y=
+github.com/go-playground/assert/v2 v2.2.0 h1:JvknZsQTYeFEAhQwI4qEt9cyV5ONwRHC+lYKSsYSR8s=
+github.com/go-playground/assert/v2 v2.2.0/go.mod h1:VDjEfimB/XKnb+ZQfWdccd7VUvScMdVu0Titje2rxJ4=
+github.com/go-playground/locales v0.14.1 h1:EWaQ/wswjilfKLTECiXz7Rh+3BjFhfDFKv/oXslEjJA=
+github.com/go-playground/locales v0.14.1/go.mod h1:hxrqLVvrK65+Rwrd5Fc6F2O76J/NuW9t0sjnWqG1slY=
+github.com/go-playground/universal-translator v0.18.1 h1:Bcnm0ZwsGyWbCzImXv+pAJnYK9S473LQFuzCbDbfSFY=
+github.com/go-playground/universal-translator v0.18.1/go.mod h1:xekY+UJKNuX9WP91TpwSH2VMlDf28Uj24BCp08ZFTUY=
+github.com/go-playground/validator/v10 v10.20.0 h1:K9ISHbSaI0lyB2eWMPJo+kOS/FBExVwjEviJTixqxL8=
+github.com/go-playground/validator/v10 v10.20.0/go.mod h1:dbuPbCMFw/DrkbEynArYaCwl3amGuJotoKCe95atGMM=
+github.com/goccy/go-json v0.10.2 h1:CrxCmQqYDkv1z7lO7Wbh2HN93uovUHgrECaO5ZrCXAU=
+github.com/goccy/go-json v0.10.2/go.mod h1:6MelG93GURQebXPDq3khkgXZkazVtN9CRI+MGFi0w8I=
+github.com/google/go-cmp v0.5.5 h1:Khx7svrCpmxxtHBq5j2mp/xVjsi8hQMfNLvJFAlrGgU=
+github.com/google/go-cmp v0.5.5/go.mod h1:v8dTdLbMG2kIc/vJvl+f65V22dbkXbowE6jgT/gNBxE=
+github.com/google/gofuzz v1.0.0/go.mod h1:dBl0BpW6vV/+mYPU4Po3pmUjxk6FQPldtuIdl/M65Eg=
+github.com/json-iterator/go v1.1.12 h1:PV8peI4a0ysnczrg+LtxykD8LfKY9ML6u2jnxaEnrnM=
+github.com/json-iterator/go v1.1.12/go.mod h1:e30LSqwooZae/UwlEbR2852Gd8hjQvJoHmT4TnhNGBo=
+github.com/klauspost/cpuid/v2 v2.0.9/go.mod h1:FInQzS24/EEf25PyTYn52gqo7WaD8xa0213Md/qVLRg=
+github.com/klauspost/cpuid/v2 v2.2.7 h1:ZWSB3igEs+d0qvnxR/ZBzXVmxkgt8DdzP6m9pfuVLDM=
+github.com/klauspost/cpuid/v2 v2.2.7/go.mod h1:Lcz8mBdAVJIBVzewtcLocK12l3Y+JytZYpaMropDUws=
+github.com/knz/go-libedit v1.10.1/go.mod h1:MZTVkCWyz0oBc7JOWP3wNAzd002ZbM/5hgShxwh4x8M=
+github.com/leodido/go-urn v1.4.0 h1:WT9HwE9SGECu3lg4d/dIA+jxlljEa1/ffXKmRjqdmIQ=
+github.com/leodido/go-urn v1.4.0/go.mod h1:bvxc+MVxLKB4z00jd1z+Dvzr47oO32F/QSNjSBOlFxI=
+github.com/mattn/go-isatty v0.0.20 h1:xfD0iDuEKnDkl03q4limB+vH+GxLEtL/jb4xVJSWWEY=
+github.com/mattn/go-isatty v0.0.20/go.mod h1:W+V8PltTTMOvKvAeJH7IuucS94S2C6jfK/D7dTCTo3Y=
+github.com/modern-go/concurrent v0.0.0-20180228061459-e0a39a4cb421/go.mod h1:6dJC0mAP4ikYIbvyc7fijjWJddQyLn8Ig3JB5CqoB9Q=
+github.com/modern-go/concurrent v0.0.0-20180306012644-bacd9c7ef1dd h1:TRLaZ9cD/w8PVh93nsPXa1VrQ6jlwL5oN8l14QlcNfg=
+github.com/modern-go/concurrent v0.0.0-20180306012644-bacd9c7ef1dd/go.mod h1:6dJC0mAP4ikYIbvyc7fijjWJddQyLn8Ig3JB5CqoB9Q=
+github.com/modern-go/reflect2 v1.0.2 h1:xBagoLtFs94CBntxluKeaWgTMpvLxC4ur3nMaC9Gz0M=
+github.com/modern-go/reflect2 v1.0.2/go.mod h1:yWuevngMOJpCy52FWWMvUC8ws7m/LJsjYzDa0/r8luk=
+github.com/pelletier/go-toml/v2 v2.2.2 h1:aYUidT7k73Pcl9nb2gScu7NSrKCSHIDE89b3+6Wq+LM=
+github.com/pelletier/go-toml/v2 v2.2.2/go.mod h1:1t835xjRzz80PqgE6HHgN2JOsmgYu/h4qDAS4n929Rs=
+github.com/pmezard/go-difflib v1.0.0 h1:4DBwDE0NGyQoBHbLQYPwSUPoCMWR5BEzIk/f1lZbAQM=
+github.com/pmezard/go-difflib v1.0.0/go.mod h1:iKH77koFhYxTK1pcRnkKkqfTogsbg7gZNVY4sRDYZ/4=
+github.com/stretchr/objx v0.1.0/go.mod h1:HFkY916IF+rwdDfMAkV7OtwuqBVzrE8GR6GFx+wExME=
+github.com/stretchr/objx v0.4.0/go.mod h1:YvHI0jy2hoMjB+UWwv71VJQ9isScKT/TqJzVSSt89Yw=
+github.com/stretchr/objx v0.5.0/go.mod h1:Yh+to48EsGEfYuaHDzXPcE3xhTkx73EhmCGUpEOglKo=
+github.com/stretchr/objx v0.5.2/go.mod h1:FRsXN1f5AsAjCGJKqEizvkpNtU+EGNCLh3NxZ/8L+MA=
+github.com/stretchr/testify v1.3.0/go.mod h1:M5WIy9Dh21IEIfnGCwXGc5bZfKNJtfHm1UVUgZn+9EI=
+github.com/stretchr/testify v1.7.0/go.mod h1:6Fq8oRcR53rry900zMqJjRRixrwX3KX962/h/Wwjteg=
+github.com/stretchr/testify v1.7.1/go.mod h1:6Fq8oRcR53rry900zMqJjRRixrwX3KX962/h/Wwjteg=
+github.com/stretchr/testify v1.8.0/go.mod h1:yNjHg4UonilssWZ8iaSj1OCr/vHnekPRkoO+kdMU+MU=
+github.com/stretchr/testify v1.8.1/go.mod h1:w2LPCIKwWwSfY2zedu0+kehJoqGctiVI29o6fzry7u4=
+github.com/stretchr/testify v1.8.4/go.mod h1:sz/lmYIOXD/1dqDmKjjqLyZ2RngseejIcXlSw2iwfAo=
+github.com/stretchr/testify v1.9.0 h1:HtqpIVDClZ4nwg75+f6Lvsy/wHu+3BoSGCbBAcpTsTg=
+github.com/stretchr/testify v1.9.0/go.mod h1:r2ic/lqez/lEtzL7wO/rwa5dbSLXVDPFyf8C91i36aY=
+github.com/tidwall/gjson v1.14.2/go.mod h1:/wbyibRr2FHMks5tjHJ5F8dMZh3AcwJEMf5vlfC0lxk=
+github.com/tidwall/gjson v1.18.0 h1:FIDeeyB800efLX89e5a8Y0BNH+LOngJyGrIWxG2FKQY=
+github.com/tidwall/gjson v1.18.0/go.mod h1:/wbyibRr2FHMks5tjHJ5F8dMZh3AcwJEMf5vlfC0lxk=
+github.com/tidwall/match v1.1.1 h1:+Ho715JplO36QYgwN9PGYNhgZvoUSc9X2c80KVTi+GA=
+github.com/tidwall/match v1.1.1/go.mod h1:eRSPERbgtNPcGhD8UCthc6PmLEQXEWd3PRB5JTxsfmM=
+github.com/tidwall/pretty v1.2.0/go.mod h1:ITEVvHYasfjBbM0u2Pg8T2nJnzm8xPwvNhhsoaGGjNU=
+github.com/tidwall/pretty v1.2.1 h1:qjsOFOWWQl+N3RsoF5/ssm1pHmJJwhjlSbZ51I6wMl4=
+github.com/tidwall/pretty v1.2.1/go.mod h1:ITEVvHYasfjBbM0u2Pg8T2nJnzm8xPwvNhhsoaGGjNU=
+github.com/tidwall/sjson v1.2.5 h1:kLy8mja+1c9jlljvWTlSazM7cKDRfJuR/bOJhcY5NcY=
+github.com/tidwall/sjson v1.2.5/go.mod h1:Fvgq9kS/6ociJEDnK0Fk1cpYF4FIW6ZF7LAe+6jwd28=
+github.com/twitchyliquid64/golang-asm v0.15.1 h1:SU5vSMR7hnwNxj24w34ZyCi/FmDZTkS4MhqMhdFk5YI=
+github.com/twitchyliquid64/golang-asm v0.15.1/go.mod h1:a1lVb/DtPvCB8fslRZhAngC2+aY1QWCk3Cedj/Gdt08=
+github.com/ugorji/go/codec v1.2.12 h1:9LC83zGrHhuUA9l16C9AHXAqEV/2wBQ4nkvumAE65EE=
+github.com/ugorji/go/codec v1.2.12/go.mod h1:UNopzCgEMSXjBc6AOMqYvWC1ktqTAfzJZUZgYf6w6lg=
+golang.org/x/arch v0.0.0-20210923205945-b76863e36670/go.mod h1:5om86z9Hs0C8fWVUuoMHwpExlXzs5Tkyp9hOrfG7pp8=
+golang.org/x/arch v0.8.0 h1:3wRIsP3pM4yUptoR96otTUOXI367OS0+c9eeRi9doIc=
+golang.org/x/arch v0.8.0/go.mod h1:FEVrYAQjsQXMVJ1nsMoVVXPZg6p2JE2mx8psSWTDQys=
+golang.org/x/crypto v0.36.0 h1:AnAEvhDddvBdpY+uR+MyHmuZzzNqXSe/GvuDeob5L34=
+golang.org/x/crypto v0.36.0/go.mod h1:Y4J0ReaxCR1IMaabaSMugxJES1EpwhBHhv2bDHklZvc=
+golang.org/x/net v0.38.0 h1:vRMAPTMaeGqVhG5QyLJHqNDwecKTomGeqbnfZyKlBI8=
+golang.org/x/net v0.38.0/go.mod h1:ivrbrMbzFq5J41QOQh0siUuly180yBYtLp+CKbEaFx8=
+golang.org/x/sys v0.5.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
+golang.org/x/sys v0.6.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
+golang.org/x/sys v0.31.0 h1:ioabZlmFYtWhL+TRYpcnNlLwhyxaM9kWTDEmfnprqik=
+golang.org/x/sys v0.31.0/go.mod h1:BJP2sWEmIv4KK5OTEluFJCKSidICx8ciO85XgH3Ak8k=
+golang.org/x/text v0.23.0 h1:D71I7dUrlY+VX0gQShAThNGHFxZ13dGLBHQLVl1mJlY=
+golang.org/x/text v0.23.0/go.mod h1:/BLNzu4aZCJ1+kcD0DNRotWKage4q2rGVAg4o22unh4=
+golang.org/x/xerrors v0.0.0-20191204190536-9bdfabe68543 h1:E7g+9GITq07hpfrRu66IVDexMakfv52eLZ2CXBWiKr4=
+golang.org/x/xerrors v0.0.0-20191204190536-9bdfabe68543/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
+google.golang.org/protobuf v1.34.1 h1:9ddQBjfCyZPOHPUiPxpYESBLc+T8P3E+Vo4IbKZgFWg=
+google.golang.org/protobuf v1.34.1/go.mod h1:c6P6GXX6sHbq/GpV6MGZEdwhWPcYBgnhAHhKbcUYpos=
 gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405 h1:yhCVgyC4o1eVCa2tZl7eS0r+SDo693bJlVdllGtEeKM=
 gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0=
+gopkg.in/yaml.v3 v3.0.0-20200313102051-9f266ea9e77c/go.mod h1:K4uyk7z7BCEPqu6E+C64Yfv1cQ7kz7rIZviUmN+EgEM=
 gopkg.in/yaml.v3 v3.0.1 h1:fxVm/GzAzEWqLHuvctI91KS9hhNmmWOoWu0XTYJS7CA=
 gopkg.in/yaml.v3 v3.0.1/go.mod h1:K4uyk7z7BCEPqu6E+C64Yfv1cQ7kz7rIZviUmN+EgEM=
+nullprogram.com/x/optparse v1.0.0/go.mod h1:KdyPE+Igbe0jQUrVfMqDMeJQIJZEuyV7pjYmp6pbG50=
+rsc.io/pdf v0.1.1/go.mod h1:n8OzWcQ6Sp37PL01nO98y4iUCRdTGarVfzxY20ICaU4=

llama-swap.go

@@ -1,33 +1,215 @@
 package main
 
 import (
+	"context"
 	"flag"
 	"fmt"
+	"log"
 	"net/http"
 	"os"
+	"os/signal"
+	"path/filepath"
+	"syscall"
+	"time"
 
+	"github.com/fsnotify/fsnotify"
+	"github.com/gin-gonic/gin"
+	"github.com/mostlygeek/llama-swap/event"
 	"github.com/mostlygeek/llama-swap/proxy"
+	"github.com/mostlygeek/llama-swap/proxy/config"
+)
+
+var (
+	version string = "0"
+	commit  string = "abcd1234"
+	date    string = "unknown"
 )
 
 func main() {
 	// Define a command-line flag for the port
 	configPath := flag.String("config", "config.yaml", "config file name")
-	listenStr := flag.String("listen", ":8080", "listen ip/port")
+	listenStr := flag.String("listen", "", "listen ip/port")
+	certFile := flag.String("tls-cert-file", "", "TLS certificate file")
+	keyFile := flag.String("tls-key-file", "", "TLS key file")
+	showVersion := flag.Bool("version", false, "show version of build")
+	watchConfig := flag.Bool("watch-config", false, "Automatically reload config file on change")
 
 	flag.Parse() // Parse the command-line flags
 
-	config, err := proxy.LoadConfig(*configPath)
+	if *showVersion {
+		fmt.Printf("version: %s (%s), built at %s\n", version, commit, date)
+		os.Exit(0)
+	}
+
+	conf, err := config.LoadConfig(*configPath)
 	if err != nil {
 		fmt.Printf("Error loading config: %v\n", err)
 		os.Exit(1)
 	}
 
-	proxyManager := proxy.New(config)
+	if len(conf.Profiles) > 0 {
-	http.HandleFunc("/", proxyManager.HandleFunc)
+		fmt.Println("WARNING: Profile functionality has been removed in favor of Groups. See the README for more information.")
+	}
 
-	fmt.Println("llamagate listening on " + *listenStr)
+	if mode := os.Getenv("GIN_MODE"); mode != "" {
-	if err := http.ListenAndServe(*listenStr, nil); err != nil {
+		gin.SetMode(mode)
-		fmt.Printf("Error starting server: %v\n", err)
+	} else {
+		gin.SetMode(gin.ReleaseMode)
+	}
+
+	// Validate TLS flags.
+	var useTLS = (*certFile != "" && *keyFile != "")
+	if (*certFile != "" && *keyFile == "") ||
+		(*certFile == "" && *keyFile != "") {
+		fmt.Println("Error: Both --tls-cert-file and --tls-key-file must be provided for TLS.")
 		os.Exit(1)
 	}
+
+	// Set default ports.
+	if *listenStr == "" {
+		defaultPort := ":8080"
+		if useTLS {
+			defaultPort = ":8443"
+		}
+		listenStr = &defaultPort
+	}
+
+	// Setup channels for server management
+	exitChan := make(chan struct{})
+	sigChan := make(chan os.Signal, 1)
+	signal.Notify(sigChan, syscall.SIGINT, syscall.SIGTERM)
+
+	// Create server with initial handler
+	srv := &http.Server{
+		Addr: *listenStr,
+	}
+
+	// Support for watching config and reloading when it changes
+	reloadProxyManager := func() {
+		if currentPM, ok := srv.Handler.(*proxy.ProxyManager); ok {
+			conf, err = config.LoadConfig(*configPath)
+			if err != nil {
+				fmt.Printf("Warning, unable to reload configuration: %v\n", err)
+				return
+			}
+
+			fmt.Println("Configuration Changed")
+			currentPM.Shutdown()
+			srv.Handler = proxy.New(conf)
+			fmt.Println("Configuration Reloaded")
+
+			// wait a few seconds and tell any UI to reload
+			time.AfterFunc(3*time.Second, func() {
+				event.Emit(proxy.ConfigFileChangedEvent{
+					ReloadingState: proxy.ReloadingStateEnd,
+				})
+			})
+		} else {
+			conf, err = config.LoadConfig(*configPath)
+			if err != nil {
+				fmt.Printf("Error, unable to load configuration: %v\n", err)
+				os.Exit(1)
+			}
+			srv.Handler = proxy.New(conf)
+		}
+	}
+
+	// load the initial proxy manager
+	reloadProxyManager()
+	debouncedReload := debounce(time.Second, reloadProxyManager)
+	if *watchConfig {
+		defer event.On(func(e proxy.ConfigFileChangedEvent) {
+			if e.ReloadingState == proxy.ReloadingStateStart {
+				debouncedReload()
+			}
+		})()
+
+		fmt.Println("Watching Configuration for changes")
+		go func() {
+			absConfigPath, err := filepath.Abs(*configPath)
+			if err != nil {
+				fmt.Printf("Error getting absolute path for watching config file: %v\n", err)
+				return
+			}
+			watcher, err := fsnotify.NewWatcher()
+			if err != nil {
+				fmt.Printf("Error creating file watcher: %v. File watching disabled.\n", err)
+				return
+			}
+
+			configDir := filepath.Dir(absConfigPath)
+			err = watcher.Add(configDir)
+			if err != nil {
+				fmt.Printf("Error adding config path directory (%s) to watcher: %v. File watching disabled.", configDir, err)
+				return
+			}
+
+			defer watcher.Close()
+			for {
+				select {
+				case changeEvent := <-watcher.Events:
+					if changeEvent.Name == absConfigPath && (changeEvent.Has(fsnotify.Write) || changeEvent.Has(fsnotify.Create) || changeEvent.Has(fsnotify.Remove)) {
+						event.Emit(proxy.ConfigFileChangedEvent{
+							ReloadingState: proxy.ReloadingStateStart,
+						})
+					} else if changeEvent.Name == filepath.Join(configDir, "..data") && changeEvent.Has(fsnotify.Create) {
+						// the change for k8s configmap
+						event.Emit(proxy.ConfigFileChangedEvent{
+							ReloadingState: proxy.ReloadingStateStart,
+						})
+					}
+
+				case err := <-watcher.Errors:
+					log.Printf("File watcher error: %v", err)
+				}
+			}
+		}()
+	}
+
+	// shutdown on signal
+	go func() {
+		sig := <-sigChan
+		fmt.Printf("Received signal %v, shutting down...\n", sig)
+		ctx, cancel := context.WithTimeout(context.Background(), time.Second*5)
+		defer cancel()
+
+		if pm, ok := srv.Handler.(*proxy.ProxyManager); ok {
+			pm.Shutdown()
+		} else {
+			fmt.Println("srv.Handler is not of type *proxy.ProxyManager")
+		}
+
+		if err := srv.Shutdown(ctx); err != nil {
+			fmt.Printf("Server shutdown error: %v\n", err)
+		}
+		close(exitChan)
+	}()
+
+	// Start server
+	go func() {
+		var err error
+		if useTLS {
+			fmt.Printf("llama-swap listening with TLS on https://%s\n", *listenStr)
+			err = srv.ListenAndServeTLS(*certFile, *keyFile)
+		} else {
+			fmt.Printf("llama-swap listening on http://%s\n", *listenStr)
+			err = srv.ListenAndServe()
+		}
+		if err != nil && err != http.ErrServerClosed {
+			log.Fatalf("Fatal server error: %v\n", err)
+		}
+	}()
+
+	// Wait for exit signal
+	<-exitChan
+}
+
+func debounce(interval time.Duration, f func()) func() {
+	var timer *time.Timer
+	return func() {
+		if timer != nil {
+			timer.Stop()
+		}
+		timer = time.AfterFunc(interval, f)
+	}
 }

misc/benchmark-chatcompletion/main.go

@@ -0,0 +1,159 @@
+package main
+
+// created for issue: #252 https://github.com/mostlygeek/llama-swap/issues/252
+// this simple benchmark tool sends a lot of small chat completion requests to llama-swap
+// to make sure all the requests are accounted for.
+//
+// requests can be sent in parallel, and the tool will report the results.
+// usage: go run main.go -baseurl http://localhost:8080/v1 -model llama3 -requests 1000 -par 5
+
+import (
+	"bytes"
+	"flag"
+	"fmt"
+	"io"
+	"log"
+	"net/http"
+	"os"
+	"sync"
+	"time"
+)
+
+func main() {
+	// ----- CLI arguments ----------------------------------------------------
+	var (
+		baseurl         string
+		modelName       string
+		totalRequests   int
+		parallelization int
+	)
+
+	flag.StringVar(&baseurl, "baseurl", "http://localhost:8080/v1", "Base URL of the API (e.g., https://api.example.com)")
+	flag.StringVar(&modelName, "model", "", "Model name to use")
+	flag.IntVar(&totalRequests, "requests", 1, "Total number of requests to send")
+	flag.IntVar(&parallelization, "par", 1, "Maximum number of concurrent requests")
+	flag.Parse()
+
+	if baseurl == "" || modelName == "" {
+		fmt.Println("Error: both -baseurl and -model are required.")
+		flag.Usage()
+		os.Exit(1)
+	}
+	if totalRequests <= 0 {
+		fmt.Println("Error: -requests must be greater than 0.")
+		os.Exit(1)
+	}
+	if parallelization <= 0 {
+		fmt.Println("Error: -parallelization must be greater than 0.")
+		os.Exit(1)
+	}
+
+	// ----- HTTP client -------------------------------------------------------
+	client := &http.Client{
+		Timeout: 30 * time.Second,
+	}
+
+	// ----- Tracking response codes -------------------------------------------
+	statusCounts := make(map[int]int) // map[statusCode]count
+	var mu sync.Mutex                 // protects statusCounts
+
+	// ----- Request queue (buffered channel) ----------------------------------
+	requests := make(chan int, 10) // Buffered channel with capacity 10
+
+	// Goroutine to fill the request queue
+	go func() {
+		for i := 0; i < totalRequests; i++ {
+			requests <- i + 1
+		}
+		close(requests)
+	}()
+
+	// ----- Worker pool -------------------------------------------------------
+	var wg sync.WaitGroup
+	for i := 0; i < parallelization; i++ {
+		wg.Add(1)
+		go func(workerID int) {
+			defer wg.Done()
+
+			for reqID := range requests {
+				// Build request payload as a single line JSON string
+				payload := `{"model":"` + modelName + `","max_tokens":100,"stream":false,"messages":[{"role":"user","content":"write a snake game in python"}]}`
+
+				// Send POST request
+				req, err := http.NewRequest(http.MethodPost,
+					fmt.Sprintf("%s/chat/completions", baseurl),
+					bytes.NewReader([]byte(payload)))
+				if err != nil {
+					log.Printf("[worker %d][req %d] request creation error: %v", workerID, reqID, err)
+					mu.Lock()
+					statusCounts[-1]++
+					mu.Unlock()
+					continue
+				}
+				req.Header.Set("Content-Type", "application/json")
+
+				resp, err := client.Do(req)
+				if err != nil {
+					log.Printf("[worker %d][req %d] HTTP request error: %v", workerID, reqID, err)
+					mu.Lock()
+					statusCounts[-1]++
+					mu.Unlock()
+					continue
+				}
+				io.Copy(io.Discard, resp.Body)
+				resp.Body.Close()
+
+				// Record status code
+				mu.Lock()
+				statusCounts[resp.StatusCode]++
+				mu.Unlock()
+			}
+		}(i + 1)
+	}
+
+	// ----- Status ticker (prints every second) -------------------------------
+	done := make(chan struct{})
+	tickerDone := make(chan struct{})
+	go func() {
+		ticker := time.NewTicker(1 * time.Second)
+		startTime := time.Now()
+		for {
+			select {
+			case <-ticker.C:
+				mu.Lock()
+				// Compute how many requests have completed so far
+				completed := 0
+				for _, cnt := range statusCounts {
+					completed += cnt
+				}
+				// Calculate duration and progress
+				duration := time.Since(startTime)
+				progress := completed * 100 / totalRequests
+				fmt.Printf("Duration: %v, Completed: %d%% requests\n", duration, progress)
+				mu.Unlock()
+			case <-done:
+				duration := time.Since(startTime)
+				fmt.Printf("Duration: %v, Completed: %d%% requests\n", duration, 100)
+				close(tickerDone)
+				return
+			}
+		}
+	}()
+
+	// Wait for all workers to finish
+	wg.Wait()
+	close(done)  // stops the status-update goroutine
+	<-tickerDone // give ticker time to finish / print
+
+	// ----- Summary ------------------------------------------------------------
+	fmt.Println("\n\n=== HTTP response code summary ===")
+	mu.Lock()
+	for code, cnt := range statusCounts {
+		if code == -1 {
+			fmt.Printf("Client-side errors (no HTTP response): %d\n", cnt)
+		} else {
+			fmt.Printf("%d : %d\n", code, cnt)
+		}
+	}
+	mu.Unlock()
+}

misc/process-cmd-test/main.go

@@ -0,0 +1,91 @@
+package main
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"os"
+	"os/exec"
+	"os/signal"
+	"syscall"
+	"time"
+)
+
+/*
+**
+Test how exec.Cmd.CommandContext behaves under certain conditions:*
+
+  - process is killed externally, what happens with cmd.Wait() *
+    ✔︎ it returns. catches crashes.*
+
+  - process ignores SIGTERM*
+    ✔︎ `kill()` is called after cmd.WaitDelay*
+
+  - this process exits, what happens with children (kill -9 <this process' pid>)*
+    x they stick around. have to be manually killed.*
+
+  - .WithTimeout()'s cancel is called *
+    ✔︎ process is killed after it ignores sigterm, cmd.Wait() catches it.*
+
+  - parent receives SIGINT/SIGTERM, what happens
+    ✔︎ waits for child process to exit, then exits gracefully.
+*/
+func main() {
+
+	// swap between these to use kill -9 <pid> on the cli to sim external crash
+	ctx, cancel := context.WithCancel(context.Background())
+	//ctx, cancel := context.WithTimeout(context.Background(), 1000*time.Millisecond)
+	defer cancel()
+
+	//cmd := exec.CommandContext(ctx, "sleep", "1")
+	cmd := exec.CommandContext(ctx,
+		"../../build/simple-responder_darwin_arm64",
+		//"-ignore-sig-term", /* so it doesn't exit on receiving SIGTERM, test cmd.WaitTimeout */
+	)
+	cmd.Stdin = os.Stdin
+	cmd.Stdout = os.Stdout
+	cmd.Stderr = os.Stderr
+
+	// set a wait delay before signing sig kill
+	cmd.WaitDelay = 500 * time.Millisecond
+	cmd.Cancel = func() error {
+		fmt.Println("✔︎ Cancel() called, sending SIGTERM")
+		cmd.Process.Signal(syscall.SIGTERM)
+
+		//return nil
+
+		// this error is returned by cmd.Wait(), and can be used to
+		// single an error when the process couldn't be normally terminated
+		// but since a SIGTERM is sent, it's probably ok to return a nil
+		// as WaitDelay timing out will override the any error set here.
+		//
+		// test by enabling/disabling -ignore-sig-term on the process
+		// with -ignore-sig-term enabled, cmd.Wait() will have "signal: killed"
+		// without it, it will show the "new error from cancel"
+		return errors.New("error from cmd.Cancel()") // sets error returned by cmd.Wait()
+	}
+
+	if err := cmd.Start(); err != nil {
+		fmt.Println("Error starting process:", err)
+		return
+	}
+
+	// catch signals. Calls cancel() which will cause cmd.Wait() to return and
+	// this program to eventually exit gracefully.
+	sigChan := make(chan os.Signal, 1)
+	signal.Notify(sigChan, syscall.SIGINT, syscall.SIGTERM)
+	go func() {
+		signal := <-sigChan
+		fmt.Printf("✔︎ Received signal: %d, Killing process... with cancel before exiting\n", signal)
+		cancel()
+	}()
+
+	fmt.Printf("✔︎ Parent Pid: %d, Process Pid: %d\n", os.Getpid(), cmd.Process.Pid)
+	fmt.Println("✔︎ Process started, cmd.Wait() ... ")
+	if err := cmd.Wait(); err != nil {
+		fmt.Println("✔︎ cmd.Wait returned, Error:", err)
+	} else {
+		fmt.Println("✔︎ cmd.Wait returned, Process exited on its own")
+	}
+	fmt.Println("✔︎ Child process exited, Done.")
+}

misc/simple-responder/simple-responder.go

@@ -3,47 +3,330 @@ package main
 import (
 	"flag"
 	"fmt"
+	"io"
+	"log"
 	"net/http"
 	"os"
+	"os/signal"
+	"syscall"
+	"time"
+
+	"github.com/gin-gonic/gin"
+	"github.com/tidwall/gjson"
 )
 
 func main() {
+	gin.SetMode(gin.TestMode)
 	// Define a command-line flag for the port
 	port := flag.String("port", "8080", "port to listen on")
+	expectedModel := flag.String("model", "TheExpectedModel", "model name to expect")
 
 	// Define a command-line flag for the response message
 	responseMessage := flag.String("respond", "hi", "message to respond with")
 
+	silent := flag.Bool("silent", false, "disable all logging")
+
+	ignoreSigTerm := flag.Bool("ignore-sig-term", false, "ignore SIGTERM signal")
+
 	flag.Parse() // Parse the command-line flags
 
-	// Set up the handler function using the provided response message
+	// Create a new Gin router
-	http.HandleFunc("/", func(w http.ResponseWriter, r *http.Request) {
+	r := gin.New()
-		// Set the header to text/plain
-		w.Header().Set("Content-Type", "text/plain")
 
-		fmt.Fprintln(w, *responseMessage)
+	// Set up the handler function using the provided response message
+	r.POST("/v1/chat/completions", func(c *gin.Context) {
+		bodyBytes, _ := io.ReadAll(c.Request.Body)
+
+		// Check if streaming is requested
+		// Query is checked instead of JSON body since that event stream conflicts with other tests
+		isStreaming := c.Query("stream") == "true"
+
+		if isStreaming {
+			// Set headers for streaming
+			c.Header("Content-Type", "text/event-stream")
+			c.Header("Cache-Control", "no-cache")
+			c.Header("Connection", "keep-alive")
+			c.Header("Transfer-Encoding", "chunked")
+
+			// add a wait to simulate a slow query
+			if wait, err := time.ParseDuration(c.Query("wait")); err == nil {
+				time.Sleep(wait)
+			}
+
+			// Send 10 "asdf" tokens
+			for i := 0; i < 10; i++ {
+				data := gin.H{
+					"created": time.Now().Unix(),
+					"choices": []gin.H{
+						{
+							"index": 0,
+							"delta": gin.H{
+								"content": "asdf",
+							},
+							"finish_reason": nil,
+						},
+					},
+				}
+				c.SSEvent("message", data)
+				c.Writer.Flush()
+			}
+
+			// Send final data with usage info
+			finalData := gin.H{
+				"usage": gin.H{
+					"completion_tokens": 10,
+					"prompt_tokens":     25,
+					"total_tokens":      35,
+				},
+				// add timings to simulate llama.cpp
+				"timings": gin.H{
+					"prompt_n":             25,
+					"prompt_ms":            13,
+					"predicted_n":          10,
+					"predicted_ms":         17,
+					"predicted_per_second": 10,
+				},
+			}
+			c.SSEvent("message", finalData)
+			c.Writer.Flush()
+
+			// Send [DONE]
+			c.SSEvent("message", "[DONE]")
+			c.Writer.Flush()
+		} else {
+			c.Header("Content-Type", "application/json")
+
+			// add a wait to simulate a slow query
+			if wait, err := time.ParseDuration(c.Query("wait")); err == nil {
+				time.Sleep(wait)
+			}
+
+			c.JSON(http.StatusOK, gin.H{
+				"responseMessage":  *responseMessage,
+				"h_content_length": c.Request.Header.Get("Content-Length"),
+				"request_body":     string(bodyBytes),
+				"usage": gin.H{
+					"completion_tokens": 10,
+					"prompt_tokens":     25,
+					"total_tokens":      35,
+				},
+				"timings": gin.H{
+					"prompt_n":             25,
+					"prompt_ms":            13,
+					"predicted_n":          10,
+					"predicted_ms":         17,
+					"predicted_per_second": 10,
+				},
+			})
+		}
+	})
+
+	// for issue #62 to check model name strips profile slug
+	// has to be one of the openAI API endpoints that llama-swap proxies
+	// curl http://localhost:8080/v1/audio/speech -d '{"model":"profile:TheExpectedModel"}'
+	r.POST("/v1/audio/speech", func(c *gin.Context) {
+		body, err := io.ReadAll(c.Request.Body)
+		if err != nil {
+			c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to read request body"})
+			return
+		}
+		defer c.Request.Body.Close()
+		modelName := gjson.GetBytes(body, "model").String()
+		if modelName != *expectedModel {
+			c.JSON(http.StatusBadRequest, gin.H{"error": fmt.Sprintf("Invalid model: %s, expected: %s", modelName, *expectedModel)})
+			return
+		} else {
+			c.JSON(http.StatusOK, gin.H{"message": "ok"})
+		}
+	})
+
+	r.POST("/v1/completions", func(c *gin.Context) {
+		c.Header("Content-Type", "application/json")
+		c.JSON(http.StatusOK, gin.H{
+			"responseMessage": *responseMessage,
+			"usage": gin.H{
+				"completion_tokens": 10,
+				"prompt_tokens":     25,
+				"total_tokens":      35,
+			},
+		})
+
+	})
+
+	// llama-server compatibility: /completion
+	r.POST("/completion", func(c *gin.Context) {
+		c.Header("Content-Type", "application/json")
+		c.JSON(http.StatusOK, gin.H{
+			"responseMessage": *responseMessage,
+			"usage": gin.H{
+				"completion_tokens": 10,
+				"prompt_tokens":     25,
+				"total_tokens":      35,
+			},
+		})
+	})
+
+	// issue #41
+	r.POST("/v1/audio/transcriptions", func(c *gin.Context) {
+		// Parse the multipart form
+		if err := c.Request.ParseMultipartForm(10 << 20); err != nil { // 10 MB max memory
+			c.JSON(http.StatusBadRequest, gin.H{"error": fmt.Sprintf("Error parsing multipart form: %s", err)})
+			return
+		}
+
+		// Get the model from the form values
+		model := c.Request.FormValue("model")
+
+		if model == "" {
+			c.JSON(http.StatusBadRequest, gin.H{"error": "Missing model parameter"})
+			return
+		}
+
+		// Get the file from the form
+		file, _, err := c.Request.FormFile("file")
+		if err != nil {
+			c.JSON(http.StatusBadRequest, gin.H{"error": fmt.Sprintf("Error getting file: %s", err)})
+			return
+		}
+		defer file.Close()
+
+		// Read the file content to get its size
+		fileBytes, err := io.ReadAll(file)
+		if err != nil {
+			c.JSON(http.StatusInternalServerError, gin.H{"error": fmt.Sprintf("Error reading file: %s", err)})
+			return
+		}
+
+		fileSize := len(fileBytes)
+
+		// Return a JSON response with the model and transcription text including file size
+		c.JSON(http.StatusOK, gin.H{
+			"text":  fmt.Sprintf("The length of the file is %d bytes", fileSize),
+			"model": model,
+
+			// expose some header values for testing
+			"h_content_type":   c.GetHeader("Content-Type"),
+			"h_content_length": c.GetHeader("Content-Length"),
+		})
+	})
+
+	r.GET("/slow-respond", func(c *gin.Context) {
+		echo := c.Query("echo")
+		delay := c.Query("delay")
+
+		if echo == "" {
+			echo = "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789"
+		}
+
+		// Parse the duration
+		if delay == "" {
+			delay = "100ms"
+		}
+
+		t, err := time.ParseDuration(delay)
+		if err != nil {
+			c.Header("Content-Type", "text/plain")
+			c.String(http.StatusBadRequest, fmt.Sprintf("Invalid duration: %s", err))
+			return
+		}
+
+		c.Header("Content-Type", "text/plain")
+		for _, char := range echo {
+			c.Writer.Write([]byte(string(char)))
+			c.Writer.Flush()
+
+			// wait
+			<-time.After(t)
+		}
+	})
+
+	r.GET("/test", func(c *gin.Context) {
+		c.Header("Content-Type", "text/plain")
+		c.String(200, *responseMessage)
+	})
+
+	r.GET("/env", func(c *gin.Context) {
+		c.Header("Content-Type", "text/plain")
+		c.String(200, *responseMessage)
 
 		// Get environment variables
 		envVars := os.Environ()
 
 		// Write each environment variable to the response
 		for _, envVar := range envVars {
-			fmt.Fprintln(w, envVar)
+			c.String(200, envVar)
 		}
 	})
 
 	// Set up the /health endpoint handler function
-	http.HandleFunc("/health", func(w http.ResponseWriter, r *http.Request) {
+	r.GET("/health", func(c *gin.Context) {
-		w.Header().Set("Content-Type", "application/json")
+		c.Header("Content-Type", "application/json")
-		response := `{"status": "ok"}`
+		c.JSON(200, gin.H{"status": "ok"})
-		w.Write([]byte(response))
 	})
 
-	address := ":" + *port // Address with the specified port
+	r.GET("/", func(c *gin.Context) {
-	fmt.Printf("Server is listening on port %s\n", *port)
+		c.Header("Content-Type", "text/plain")
+		c.String(200, fmt.Sprintf("%s %s", c.Request.Method, c.Request.URL.Path))
+	})
 
-	// Start the server and log any error if it occurs
+	address := "127.0.0.1:" + *port // Address with the specified port
-	if err := http.ListenAndServe(address, nil); err != nil {
+
-		fmt.Printf("Error starting server: %s\n", err)
+	srv := &http.Server{
+		Addr:    address,
+		Handler: r.Handler(),
 	}
+
+	// Disable logging if the --silent flag is set
+	if *silent {
+		gin.SetMode(gin.ReleaseMode)
+		gin.DefaultWriter = io.Discard
+		log.SetOutput(io.Discard)
+	}
+
+	if !*silent {
+		fmt.Printf("My PID: %d\n", os.Getpid())
+	}
+
+	go func() {
+		log.Printf("simple-responder listening on %s\n", address)
+		// service connections
+		if err := srv.ListenAndServe(); err != nil && err != http.ErrServerClosed {
+			log.Fatalf("simple-responder err: %s\n", err)
+		}
+	}()
+
+	// Wait for interrupt signal to gracefully shutdown the server with
+	// a timeout of 5 seconds.
+	sigChan := make(chan os.Signal, 1)
+	// kill (no param) default send syscall.SIGTERM
+	// kill -2 is syscall.SIGINT
+	// kill -9 is syscall.SIGKILL but can't be catch, so don't need add it
+	signal.Notify(sigChan, syscall.SIGINT, syscall.SIGTERM)
+
+	countSigInt := 0
+
+runloop:
+	for {
+		signal := <-sigChan
+		switch signal {
+		case syscall.SIGINT:
+			countSigInt++
+			if countSigInt > 1 {
+				break runloop
+			} else {
+				log.Println("Received SIGINT, send another SIGINT to shutdown")
+			}
+		case syscall.SIGTERM:
+			if *ignoreSigTerm {
+				log.Println("Ignoring SIGTERM")
+			} else {
+				log.Println("Received SIGTERM, shutting down")
+				break runloop
+			}
+		default:
+			break runloop
+		}
+	}
+
+	log.Println("simple-responder shutting down")
 }

misc/test-rerank/README.md

@@ -0,0 +1,4 @@
+The rerank-test.json data is from https://github.com/ggerganov/llama.cpp/pull/9510
+
+To run it:
+> curl http://127.0.0.1:8080/v1/rerank -H "Content-Type: application/json" -d @reranker-test.json  -v | jq .

misc/test-rerank/reranker-test.json

@@ -0,0 +1,17 @@
+{
+  "model": "bge-reranker",
+  "query": "Organic skincare products for sensitive skin",
+  "top_n": 3,
+  "documents": [
+    "Organic skincare for sensitive skin with aloe vera and chamomile: Imagine the soothing embrace of nature with our organic skincare range, crafted specifically for sensitive skin. Infused with the calming properties of aloe vera and chamomile, each product provides gentle nourishment and protection. Say goodbye to irritation and hello to a glowing, healthy complexion.",
+    "New makeup trends focus on bold colors and innovative techniques: Step into the world of cutting-edge beauty with this seasons makeup trends. Bold, vibrant colors and groundbreaking techniques are redefining the art of makeup. From neon eyeliners to holographic highlighters, unleash your creativity and make a statement with every look.",
+    "Bio-Hautpflege für empfindliche Haut mit Aloe Vera und Kamille: Erleben Sie die wohltuende Wirkung unserer Bio-Hautpflege, speziell für empfindliche Haut entwickelt. Mit den beruhigenden Eigenschaften von Aloe Vera und Kamille pflegen und schützen unsere Produkte Ihre Haut auf natürliche Weise. Verabschieden Sie sich von Hautirritationen und genießen Sie einen strahlenden Teint.",
+    "Neue Make-up-Trends setzen auf kräftige Farben und innovative Techniken: Tauchen Sie ein in die Welt der modernen Schönheit mit den neuesten Make-up-Trends. Kräftige, lebendige Farben und innovative Techniken setzen neue Maßstäbe. Von auffälligen Eyelinern bis hin zu holografischen Highlightern – lassen Sie Ihrer Kreativität freien Lauf und setzen Sie jedes Mal ein Statement.",
+    "Cuidado de la piel orgánico para piel sensible con aloe vera y manzanilla: Descubre el poder de la naturaleza con nuestra línea de cuidado de la piel orgánico, diseñada especialmente para pieles sensibles. Enriquecidos con aloe vera y manzanilla, estos productos ofrecen una hidratación y protección suave. Despídete de las irritaciones y saluda a una piel radiante y saludable.",
+    "Las nuevas tendencias de maquillaje se centran en colores vivos y técnicas innovadoras: Entra en el fascinante mundo del maquillaje con las tendencias más actuales. Colores vivos y técnicas innovadoras están revolucionando el arte del maquillaje. Desde delineadores neón hasta iluminadores holográficos, desata tu creatividad y destaca en cada look.",
+    "针对敏感肌专门设计的天然有机护肤产品：体验由芦荟和洋甘菊提取物带来的自然呵护。我们的护肤产品特别为敏感肌设计，温和滋润，保护您的肌肤不受刺激。让您的肌肤告别不适，迎来健康光彩。",
+    "新的化妆趋势注重鲜艳的颜色和创新的技巧：进入化妆艺术的新纪元，本季的化妆趋势以大胆的颜色和创新的技巧为主。无论是霓虹眼线还是全息高光，每一款妆容都能让您脱颖而出，展现独特魅力。",
+    "敏感肌のために特別に設計された天然有機スキンケア製品: アロエベラとカモミールのやさしい力で、自然の抱擁を感じてください。敏感肌用に特別に設計された私たちのスキンケア製品は、肌に優しく栄養を与え、保護します。肌トラブルにさようなら、輝く健康な肌にこんにちは。",
+    "新しいメイクのトレンドは鮮やかな色と革新的な技術に焦点を当てています: 今シーズンのメイクアップトレンドは、大胆な色彩と革新的な技術に注目しています。ネオンアイライナーからホログラフィックハイライターまで、クリエイティビティを解き放ち、毎回ユニークなルックを演出しましょう。"
+  ]
+}

proxy/.gitignore

@@ -0,0 +1 @@
+ui_dist/*

proxy/config.go

@@ -1,56 +0,0 @@
-package proxy
-
-import (
-	"os"
-
-	"gopkg.in/yaml.v3"
-)
-
-type ModelConfig struct {
-	Cmd     string   `yaml:"cmd"`
-	Proxy   string   `yaml:"proxy"`
-	Aliases []string `yaml:"aliases"`
-	Env     []string `yaml:"env"`
-}
-
-type Config struct {
-	Models             map[string]ModelConfig `yaml:"models"`
-	HealthCheckTimeout int                    `yaml:"healthCheckTimeout"`
-}
-
-func (c *Config) FindConfig(modelName string) (ModelConfig, bool) {
-	modelConfig, found := c.Models[modelName]
-	if found {
-		return modelConfig, true
-	}
-
-	// Search through aliases to find the right config
-	for _, config := range c.Models {
-		for _, alias := range config.Aliases {
-			if alias == modelName {
-				return config, true
-			}
-		}
-	}
-
-	return ModelConfig{}, false
-}
-
-func LoadConfig(path string) (*Config, error) {
-	data, err := os.ReadFile(path)
-	if err != nil {
-		return nil, err
-	}
-
-	var config Config
-	err = yaml.Unmarshal(data, &config)
-	if err != nil {
-		return nil, err
-	}
-
-	if config.HealthCheckTimeout < 15 {
-		config.HealthCheckTimeout = 15
-	}
-
-	return &config, nil
-}

proxy/config/config.go

@@ -0,0 +1,601 @@
+package config
+
+import (
+	"fmt"
+	"io"
+	"net/url"
+	"os"
+	"regexp"
+	"runtime"
+	"sort"
+	"strings"
+
+	"github.com/billziss-gh/golib/shlex"
+	"gopkg.in/yaml.v3"
+)
+
+const DEFAULT_GROUP_ID = "(default)"
+
+type MacroEntry struct {
+	Name  string
+	Value any
+}
+
+type MacroList []MacroEntry
+
+// UnmarshalYAML implements custom YAML unmarshaling that preserves macro definition order
+func (ml *MacroList) UnmarshalYAML(value *yaml.Node) error {
+	if value.Kind != yaml.MappingNode {
+		return fmt.Errorf("macros must be a mapping")
+	}
+
+	// yaml.Node.Content for a mapping contains alternating key/value nodes
+	entries := make([]MacroEntry, 0, len(value.Content)/2)
+	for i := 0; i < len(value.Content); i += 2 {
+		keyNode := value.Content[i]
+		valueNode := value.Content[i+1]
+
+		var name string
+		if err := keyNode.Decode(&name); err != nil {
+			return fmt.Errorf("failed to decode macro name: %w", err)
+		}
+
+		var val any
+		if err := valueNode.Decode(&val); err != nil {
+			return fmt.Errorf("failed to decode macro value for '%s': %w", name, err)
+		}
+
+		entries = append(entries, MacroEntry{Name: name, Value: val})
+	}
+
+	*ml = entries
+	return nil
+}
+
+// Get retrieves a macro value by name
+func (ml MacroList) Get(name string) (any, bool) {
+	for _, entry := range ml {
+		if entry.Name == name {
+			return entry.Value, true
+		}
+	}
+	return nil, false
+}
+
+// ToMap converts MacroList to a map (for backward compatibility if needed)
+func (ml MacroList) ToMap() map[string]any {
+	result := make(map[string]any, len(ml))
+	for _, entry := range ml {
+		result[entry.Name] = entry.Value
+	}
+	return result
+}
+
+type GroupConfig struct {
+	Swap       bool     `yaml:"swap"`
+	Exclusive  bool     `yaml:"exclusive"`
+	Persistent bool     `yaml:"persistent"`
+	Members    []string `yaml:"members"`
+}
+
+var (
+	macroNameRegex    = regexp.MustCompile(`^[a-zA-Z0-9_-]+$`)
+	macroPatternRegex = regexp.MustCompile(`\$\{([a-zA-Z0-9_-]+)\}`)
+)
+
+// set default values for GroupConfig
+func (c *GroupConfig) UnmarshalYAML(unmarshal func(interface{}) error) error {
+	type rawGroupConfig GroupConfig
+	defaults := rawGroupConfig{
+		Swap:       true,
+		Exclusive:  true,
+		Persistent: false,
+		Members:    []string{},
+	}
+
+	if err := unmarshal(&defaults); err != nil {
+		return err
+	}
+
+	*c = GroupConfig(defaults)
+	return nil
+}
+
+type HooksConfig struct {
+	OnStartup HookOnStartup `yaml:"on_startup"`
+}
+
+type HookOnStartup struct {
+	Preload []string `yaml:"preload"`
+}
+
+type Config struct {
+	HealthCheckTimeout int                    `yaml:"healthCheckTimeout"`
+	LogRequests        bool                   `yaml:"logRequests"`
+	LogLevel           string                 `yaml:"logLevel"`
+	MetricsMaxInMemory int                    `yaml:"metricsMaxInMemory"`
+	Models             map[string]ModelConfig `yaml:"models"` /* key is model ID */
+	Profiles           map[string][]string    `yaml:"profiles"`
+	Groups             map[string]GroupConfig `yaml:"groups"` /* key is group ID */
+
+	// for key/value replacements in model's cmd, cmdStop, proxy, checkEndPoint
+	Macros MacroList `yaml:"macros"`
+
+	// map aliases to actual model IDs
+	aliases map[string]string
+
+	// automatic port assignments
+	StartPort int `yaml:"startPort"`
+
+	// hooks, see: #209
+	Hooks HooksConfig `yaml:"hooks"`
+}
+
+func (c *Config) RealModelName(search string) (string, bool) {
+	if _, found := c.Models[search]; found {
+		return search, true
+	} else if name, found := c.aliases[search]; found {
+		return name, found
+	} else {
+		return "", false
+	}
+}
+
+func (c *Config) FindConfig(modelName string) (ModelConfig, string, bool) {
+	if realName, found := c.RealModelName(modelName); !found {
+		return ModelConfig{}, "", false
+	} else {
+		return c.Models[realName], realName, true
+	}
+}
+
+func LoadConfig(path string) (Config, error) {
+	file, err := os.Open(path)
+	if err != nil {
+		return Config{}, err
+	}
+	defer file.Close()
+	return LoadConfigFromReader(file)
+}
+
+func LoadConfigFromReader(r io.Reader) (Config, error) {
+	data, err := io.ReadAll(r)
+	if err != nil {
+		return Config{}, err
+	}
+
+	// default configuration values
+	config := Config{
+		HealthCheckTimeout: 120,
+		StartPort:          5800,
+		LogLevel:           "info",
+		MetricsMaxInMemory: 1000,
+	}
+	err = yaml.Unmarshal(data, &config)
+	if err != nil {
+		return Config{}, err
+	}
+
+	if config.HealthCheckTimeout < 15 {
+		// set a minimum of 15 seconds
+		config.HealthCheckTimeout = 15
+	}
+
+	if config.StartPort < 1 {
+		return Config{}, fmt.Errorf("startPort must be greater than 1")
+	}
+
+	// Populate the aliases map
+	config.aliases = make(map[string]string)
+	for modelName, modelConfig := range config.Models {
+		for _, alias := range modelConfig.Aliases {
+			if _, found := config.aliases[alias]; found {
+				return Config{}, fmt.Errorf("duplicate alias %s found in model: %s", alias, modelName)
+			}
+			config.aliases[alias] = modelName
+		}
+	}
+
+	/* check macro constraint rules:
+
+	- name must fit the regex ^[a-zA-Z0-9_-]+$
+	- names must be less than 64 characters (no reason, just cause)
+	- name can not be any reserved macros: PORT, MODEL_ID
+	- macro values must be less than 1024 characters
+	*/
+	for _, macro := range config.Macros {
+		if err = validateMacro(macro.Name, macro.Value); err != nil {
+			return Config{}, err
+		}
+	}
+
+	// Get and sort all model IDs first, makes testing more consistent
+	modelIds := make([]string, 0, len(config.Models))
+	for modelId := range config.Models {
+		modelIds = append(modelIds, modelId)
+	}
+	sort.Strings(modelIds) // This guarantees stable iteration order
+
+	nextPort := config.StartPort
+	for _, modelId := range modelIds {
+		modelConfig := config.Models[modelId]
+
+		// Strip comments from command fields before macro expansion
+		modelConfig.Cmd = StripComments(modelConfig.Cmd)
+		modelConfig.CmdStop = StripComments(modelConfig.CmdStop)
+
+		// validate model macros
+		for _, macro := range modelConfig.Macros {
+			if err = validateMacro(macro.Name, macro.Value); err != nil {
+				return Config{}, fmt.Errorf("model %s: %s", modelId, err.Error())
+			}
+		}
+
+		// Merge global config and model macros. Model macros take precedence
+		mergedMacros := make(MacroList, 0, len(config.Macros)+len(modelConfig.Macros))
+		mergedMacros = append(mergedMacros, MacroEntry{Name: "MODEL_ID", Value: modelId})
+
+		// Add global macros first
+		mergedMacros = append(mergedMacros, config.Macros...)
+
+		// Add model macros (can override global)
+		for _, entry := range modelConfig.Macros {
+			// Remove any existing global macro with same name
+			found := false
+			for i, existing := range mergedMacros {
+				if existing.Name == entry.Name {
+					mergedMacros[i] = entry // Override
+					found = true
+					break
+				}
+			}
+			if !found {
+				mergedMacros = append(mergedMacros, entry)
+			}
+		}
+
+		// First pass: Substitute user-defined macros in reverse order (LIFO - last defined first)
+		// This allows later macros to reference earlier ones
+		for i := len(mergedMacros) - 1; i >= 0; i-- {
+			entry := mergedMacros[i]
+			macroSlug := fmt.Sprintf("${%s}", entry.Name)
+			macroStr := fmt.Sprintf("%v", entry.Value)
+
+			// Substitute in command fields
+			modelConfig.Cmd = strings.ReplaceAll(modelConfig.Cmd, macroSlug, macroStr)
+			modelConfig.CmdStop = strings.ReplaceAll(modelConfig.CmdStop, macroSlug, macroStr)
+			modelConfig.Proxy = strings.ReplaceAll(modelConfig.Proxy, macroSlug, macroStr)
+			modelConfig.CheckEndpoint = strings.ReplaceAll(modelConfig.CheckEndpoint, macroSlug, macroStr)
+			modelConfig.Filters.StripParams = strings.ReplaceAll(modelConfig.Filters.StripParams, macroSlug, macroStr)
+
+			// Substitute in metadata (recursive)
+			if len(modelConfig.Metadata) > 0 {
+				var err error
+				result, err := substituteMacroInValue(modelConfig.Metadata, entry.Name, entry.Value)
+				if err != nil {
+					return Config{}, fmt.Errorf("model %s metadata: %s", modelId, err.Error())
+				}
+				modelConfig.Metadata = result.(map[string]any)
+			}
+		}
+
+		// Final pass: check if PORT macro is needed after macro expansion
+		// ${PORT} is a resource on the local machine so a new port is only allocated
+		// if it is required in either cmd or proxy keys
+		cmdHasPort := strings.Contains(modelConfig.Cmd, "${PORT}")
+		proxyHasPort := strings.Contains(modelConfig.Proxy, "${PORT}")
+		if cmdHasPort || proxyHasPort { // either has it
+			if !cmdHasPort && proxyHasPort { // but both don't have it
+				return Config{}, fmt.Errorf("model %s: proxy uses ${PORT} but cmd does not - ${PORT} is only available when used in cmd", modelId)
+			}
+
+			// Add PORT macro and substitute it
+			portEntry := MacroEntry{Name: "PORT", Value: nextPort}
+			macroSlug := "${PORT}"
+			macroStr := fmt.Sprintf("%v", nextPort)
+
+			modelConfig.Cmd = strings.ReplaceAll(modelConfig.Cmd, macroSlug, macroStr)
+			modelConfig.CmdStop = strings.ReplaceAll(modelConfig.CmdStop, macroSlug, macroStr)
+			modelConfig.Proxy = strings.ReplaceAll(modelConfig.Proxy, macroSlug, macroStr)
+
+			// Substitute PORT in metadata
+			if len(modelConfig.Metadata) > 0 {
+				var err error
+				result, err := substituteMacroInValue(modelConfig.Metadata, portEntry.Name, portEntry.Value)
+				if err != nil {
+					return Config{}, fmt.Errorf("model %s metadata: %s", modelId, err.Error())
+				}
+				modelConfig.Metadata = result.(map[string]any)
+			}
+
+			nextPort++
+		}
+
+		// make sure there are no unknown macros that have not been replaced
+		fieldMap := map[string]string{
+			"cmd":                 modelConfig.Cmd,
+			"cmdStop":             modelConfig.CmdStop,
+			"proxy":               modelConfig.Proxy,
+			"checkEndpoint":       modelConfig.CheckEndpoint,
+			"filters.stripParams": modelConfig.Filters.StripParams,
+		}
+
+		for fieldName, fieldValue := range fieldMap {
+			matches := macroPatternRegex.FindAllStringSubmatch(fieldValue, -1)
+			for _, match := range matches {
+				macroName := match[1]
+				if macroName == "PID" && fieldName == "cmdStop" {
+					continue // this is ok, has to be replaced by process later
+				}
+				// Reserved macros are always valid (they should have been substituted already)
+				if macroName == "PORT" || macroName == "MODEL_ID" {
+					return Config{}, fmt.Errorf("macro '${%s}' should have been substituted in %s.%s", macroName, modelId, fieldName)
+				}
+				// Any other macro is unknown
+				return Config{}, fmt.Errorf("unknown macro '${%s}' found in %s.%s", macroName, modelId, fieldName)
+			}
+		}
+
+		// Check for unknown macros in metadata
+		if len(modelConfig.Metadata) > 0 {
+			if err := validateMetadataForUnknownMacros(modelConfig.Metadata, modelId); err != nil {
+				return Config{}, err
+			}
+		}
+
+		// Validate the proxy URL.
+		if _, err := url.Parse(modelConfig.Proxy); err != nil {
+			return Config{}, fmt.Errorf(
+				"model %s: invalid proxy URL: %w", modelId, err,
+			)
+		}
+
+		config.Models[modelId] = modelConfig
+	}
+
+	config = AddDefaultGroupToConfig(config)
+	// check that members are all unique in the groups
+	memberUsage := make(map[string]string) // maps member to group it appears in
+	for groupID, groupConfig := range config.Groups {
+		prevSet := make(map[string]bool)
+		for _, member := range groupConfig.Members {
+			// Check for duplicates within this group
+			if _, found := prevSet[member]; found {
+				return Config{}, fmt.Errorf("duplicate model member %s found in group: %s", member, groupID)
+			}
+			prevSet[member] = true
+
+			// Check if member is used in another group
+			if existingGroup, exists := memberUsage[member]; exists {
+				return Config{}, fmt.Errorf("model member %s is used in multiple groups: %s and %s", member, existingGroup, groupID)
+			}
+			memberUsage[member] = groupID
+		}
+	}
+
+	// clean up hooks preload
+	if len(config.Hooks.OnStartup.Preload) > 0 {
+		var toPreload []string
+		for _, modelID := range config.Hooks.OnStartup.Preload {
+			modelID = strings.TrimSpace(modelID)
+			if modelID == "" {
+				continue
+			}
+			if real, found := config.RealModelName(modelID); found {
+				toPreload = append(toPreload, real)
+			}
+		}
+
+		config.Hooks.OnStartup.Preload = toPreload
+	}
+
+	return config, nil
+}
+
+// rewrites the yaml to include a default group with any orphaned models
+func AddDefaultGroupToConfig(config Config) Config {
+
+	if config.Groups == nil {
+		config.Groups = make(map[string]GroupConfig)
+	}
+
+	defaultGroup := GroupConfig{
+		Swap:      true,
+		Exclusive: true,
+		Members:   []string{},
+	}
+	// if groups is empty, create a default group and put
+	// all models into it
+	if len(config.Groups) == 0 {
+		for modelName := range config.Models {
+			defaultGroup.Members = append(defaultGroup.Members, modelName)
+		}
+	} else {
+		// iterate over existing group members and add non-grouped models into the default group
+		for modelName := range config.Models {
+			foundModel := false
+		found:
+			// search for the model in existing groups
+			for _, groupConfig := range config.Groups {
+				for _, member := range groupConfig.Members {
+					if member == modelName {
+						foundModel = true
+						break found
+					}
+				}
+			}
+
+			if !foundModel {
+				defaultGroup.Members = append(defaultGroup.Members, modelName)
+			}
+		}
+	}
+
+	sort.Strings(defaultGroup.Members) // make consistent ordering for testing
+	config.Groups[DEFAULT_GROUP_ID] = defaultGroup
+
+	return config
+}
+
+func SanitizeCommand(cmdStr string) ([]string, error) {
+	var cleanedLines []string
+	for _, line := range strings.Split(cmdStr, "\n") {
+		trimmed := strings.TrimSpace(line)
+		// Skip comment lines
+		if strings.HasPrefix(trimmed, "#") {
+			continue
+		}
+		// Handle trailing backslashes by replacing with space
+		if strings.HasSuffix(trimmed, "\\") {
+			cleanedLines = append(cleanedLines, strings.TrimSuffix(trimmed, "\\")+" ")
+		} else {
+			cleanedLines = append(cleanedLines, line)
+		}
+	}
+
+	// put it back together
+	cmdStr = strings.Join(cleanedLines, "\n")
+
+	// Split the command into arguments
+	var args []string
+	if runtime.GOOS == "windows" {
+		args = shlex.Windows.Split(cmdStr)
+	} else {
+		args = shlex.Posix.Split(cmdStr)
+	}
+
+	// Ensure the command is not empty
+	if len(args) == 0 {
+		return nil, fmt.Errorf("empty command")
+	}
+
+	return args, nil
+}
+
+func StripComments(cmdStr string) string {
+	var cleanedLines []string
+	for _, line := range strings.Split(cmdStr, "\n") {
+		trimmed := strings.TrimSpace(line)
+		// Skip comment lines
+		if strings.HasPrefix(trimmed, "#") {
+			continue
+		}
+		cleanedLines = append(cleanedLines, line)
+	}
+	return strings.Join(cleanedLines, "\n")
+}
+
+// validateMacro validates macro name and value constraints
+func validateMacro(name string, value any) error {
+	if len(name) >= 64 {
+		return fmt.Errorf("macro name '%s' exceeds maximum length of 63 characters", name)
+	}
+	if !macroNameRegex.MatchString(name) {
+		return fmt.Errorf("macro name '%s' contains invalid characters, must match pattern ^[a-zA-Z0-9_-]+$", name)
+	}
+
+	// Validate that value is a scalar type
+	switch v := value.(type) {
+	case string:
+		if len(v) >= 1024 {
+			return fmt.Errorf("macro value for '%s' exceeds maximum length of 1024 characters", name)
+		}
+		// Check for self-reference
+		macroSlug := fmt.Sprintf("${%s}", name)
+		if strings.Contains(v, macroSlug) {
+			return fmt.Errorf("macro '%s' contains self-reference", name)
+		}
+	case int, int8, int16, int32, int64, uint, uint8, uint16, uint32, uint64, float32, float64, bool:
+		// These types are allowed
+	default:
+		return fmt.Errorf("macro '%s' has invalid type %T, must be a scalar type (string, int, float, or bool)", name, value)
+	}
+
+	switch name {
+	case "PORT", "MODEL_ID":
+		return fmt.Errorf("macro name '%s' is reserved", name)
+	}
+
+	return nil
+}
+
+// validateMetadataForUnknownMacros recursively checks for any remaining macro references in metadata
+func validateMetadataForUnknownMacros(value any, modelId string) error {
+	switch v := value.(type) {
+	case string:
+		matches := macroPatternRegex.FindAllStringSubmatch(v, -1)
+		for _, match := range matches {
+			macroName := match[1]
+			return fmt.Errorf("model %s metadata: unknown macro '${%s}'", modelId, macroName)
+		}
+		return nil
+
+	case map[string]any:
+		for _, val := range v {
+			if err := validateMetadataForUnknownMacros(val, modelId); err != nil {
+				return err
+			}
+		}
+		return nil
+
+	case []any:
+		for _, val := range v {
+			if err := validateMetadataForUnknownMacros(val, modelId); err != nil {
+				return err
+			}
+		}
+		return nil
+
+	default:
+		// Scalar types don't contain macros
+		return nil
+	}
+}
+
+// substituteMacroInValue recursively substitutes a single macro in a value structure
+// This is called once per macro, allowing LIFO substitution order
+func substituteMacroInValue(value any, macroName string, macroValue any) (any, error) {
+	macroSlug := fmt.Sprintf("${%s}", macroName)
+	macroStr := fmt.Sprintf("%v", macroValue)
+
+	switch v := value.(type) {
+	case string:
+		// Check if this is a direct macro substitution
+		if v == macroSlug {
+			return macroValue, nil
+		}
+		// Handle string interpolation
+		if strings.Contains(v, macroSlug) {
+			return strings.ReplaceAll(v, macroSlug, macroStr), nil
+		}
+		return v, nil
+
+	case map[string]any:
+		// Recursively process map values
+		newMap := make(map[string]any)
+		for key, val := range v {
+			newVal, err := substituteMacroInValue(val, macroName, macroValue)
+			if err != nil {
+				return nil, err
+			}
+			newMap[key] = newVal
+		}
+		return newMap, nil
+
+	case []any:
+		// Recursively process slice elements
+		newSlice := make([]any, len(v))
+		for i, val := range v {
+			newVal, err := substituteMacroInValue(val, macroName, macroValue)
+			if err != nil {
+				return nil, err
+			}
+			newSlice[i] = newVal
+		}
+		return newSlice, nil
+
+	default:
+		// Return scalar types as-is
+		return value, nil
+	}
+}

proxy/config/config_posix_test.go

@@ -0,0 +1,242 @@
+//go:build !windows
+
+package config
+
+import (
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+)
+
+func TestConfig_SanitizeCommand(t *testing.T) {
+	// Test a command with spaces and newlines
+	args, err := SanitizeCommand(`python model1.py \
+		-a "double quotes" \
+		--arg2 'single quotes'
+		-s
+		# comment 1
+		--arg3 123 \
+
+		  # comment 2
+		--arg4 '"string in string"'
+
+
+		# this will get stripped out as well as the white space above
+		-c "'single quoted'"
+		`)
+	assert.NoError(t, err)
+	assert.Equal(t, []string{
+		"python", "model1.py",
+		"-a", "double quotes",
+		"--arg2", "single quotes",
+		"-s",
+		"--arg3", "123",
+		"--arg4", `"string in string"`,
+		"-c", `'single quoted'`,
+	}, args)
+
+	// Test an empty command
+	args, err = SanitizeCommand("")
+	assert.Error(t, err)
+	assert.Nil(t, args)
+}
+
+// Test the default values are automatically set for global, model and group configurations
+// after loading the configuration
+func TestConfig_DefaultValuesPosix(t *testing.T) {
+	content := `
+models:
+  model1:
+    cmd: path/to/cmd --port ${PORT}
+`
+
+	config, err := LoadConfigFromReader(strings.NewReader(content))
+	assert.NoError(t, err)
+	assert.Equal(t, 120, config.HealthCheckTimeout)
+	assert.Equal(t, 5800, config.StartPort)
+	assert.Equal(t, "info", config.LogLevel)
+
+	// Test default group exists
+	defaultGroup, exists := config.Groups["(default)"]
+	assert.True(t, exists, "default group should exist")
+	if assert.NotNil(t, defaultGroup, "default group should not be nil") {
+		assert.Equal(t, true, defaultGroup.Swap)
+		assert.Equal(t, true, defaultGroup.Exclusive)
+		assert.Equal(t, false, defaultGroup.Persistent)
+		assert.Equal(t, []string{"model1"}, defaultGroup.Members)
+	}
+
+	model1, exists := config.Models["model1"]
+	assert.True(t, exists, "model1 should exist")
+	if assert.NotNil(t, model1, "model1 should not be nil") {
+		assert.Equal(t, "path/to/cmd --port 5800", model1.Cmd) // has the port replaced
+		assert.Equal(t, "", model1.CmdStop)
+		assert.Equal(t, "http://localhost:5800", model1.Proxy)
+		assert.Equal(t, "/health", model1.CheckEndpoint)
+		assert.Equal(t, []string{}, model1.Aliases)
+		assert.Equal(t, []string{}, model1.Env)
+		assert.Equal(t, 0, model1.UnloadAfter)
+		assert.Equal(t, false, model1.Unlisted)
+		assert.Equal(t, "", model1.UseModelName)
+		assert.Equal(t, 0, model1.ConcurrencyLimit)
+	}
+
+	// default empty filter exists
+	assert.Equal(t, "", model1.Filters.StripParams)
+}
+
+func TestConfig_LoadPosix(t *testing.T) {
+	// Create a temporary YAML file for testing
+	tempDir, err := os.MkdirTemp("", "test-config")
+	if err != nil {
+		t.Fatalf("Failed to create temporary directory: %v", err)
+	}
+	defer os.RemoveAll(tempDir)
+
+	tempFile := filepath.Join(tempDir, "config.yaml")
+	content := `
+macros:
+  svr-path: "path/to/server"
+hooks:
+  on_startup:
+    preload: ["model1", "model2"]
+models:
+  model1:
+    cmd: path/to/cmd --arg1 one
+    proxy: "http://localhost:8080"
+    name: "Model 1"
+    description: "This is model 1"
+    aliases:
+      - "m1"
+      - "model-one"
+    env:
+      - "VAR1=value1"
+      - "VAR2=value2"
+    checkEndpoint: "/health"
+  model2:
+    cmd: ${svr-path} --arg1 one
+    proxy: "http://localhost:8081"
+    aliases:
+      - "m2"
+    checkEndpoint: "/"
+  model3:
+    cmd: path/to/cmd --arg1 one
+    proxy: "http://localhost:8081"
+    aliases:
+      - "mthree"
+    checkEndpoint: "/"
+  model4:
+    cmd: path/to/cmd --arg1 one
+    proxy: "http://localhost:8082"
+    checkEndpoint: "/"
+
+healthCheckTimeout: 15
+profiles:
+  test:
+    - model1
+    - model2
+groups:
+  group1:
+    swap: true
+    exclusive: false
+    members: ["model2"]
+  forever:
+    exclusive: false
+    persistent: true
+    members:
+      - "model4"
+`
+
+	if err := os.WriteFile(tempFile, []byte(content), 0644); err != nil {
+		t.Fatalf("Failed to write temporary file: %v", err)
+	}
+
+	// Load the config and verify
+	config, err := LoadConfig(tempFile)
+	if err != nil {
+		t.Fatalf("Failed to load config: %v", err)
+	}
+
+	expected := Config{
+		LogLevel:  "info",
+		StartPort: 5800,
+		Macros: MacroList{
+			{"svr-path", "path/to/server"},
+		},
+		Hooks: HooksConfig{
+			OnStartup: HookOnStartup{
+				Preload: []string{"model1", "model2"},
+			},
+		},
+		Models: map[string]ModelConfig{
+			"model1": {
+				Cmd:           "path/to/cmd --arg1 one",
+				Proxy:         "http://localhost:8080",
+				Aliases:       []string{"m1", "model-one"},
+				Env:           []string{"VAR1=value1", "VAR2=value2"},
+				CheckEndpoint: "/health",
+				Name:          "Model 1",
+				Description:   "This is model 1",
+			},
+			"model2": {
+				Cmd:           "path/to/server --arg1 one",
+				Proxy:         "http://localhost:8081",
+				Aliases:       []string{"m2"},
+				Env:           []string{},
+				CheckEndpoint: "/",
+			},
+			"model3": {
+				Cmd:           "path/to/cmd --arg1 one",
+				Proxy:         "http://localhost:8081",
+				Aliases:       []string{"mthree"},
+				Env:           []string{},
+				CheckEndpoint: "/",
+			},
+			"model4": {
+				Cmd:           "path/to/cmd --arg1 one",
+				Proxy:         "http://localhost:8082",
+				CheckEndpoint: "/",
+				Aliases:       []string{},
+				Env:           []string{},
+			},
+		},
+		HealthCheckTimeout: 15,
+		MetricsMaxInMemory: 1000,
+		Profiles: map[string][]string{
+			"test": {"model1", "model2"},
+		},
+		aliases: map[string]string{
+			"m1":        "model1",
+			"model-one": "model1",
+			"m2":        "model2",
+			"mthree":    "model3",
+		},
+		Groups: map[string]GroupConfig{
+			DEFAULT_GROUP_ID: {
+				Swap:      true,
+				Exclusive: true,
+				Members:   []string{"model1", "model3"},
+			},
+			"group1": {
+				Swap:      true,
+				Exclusive: false,
+				Members:   []string{"model2"},
+			},
+			"forever": {
+				Swap:       true,
+				Exclusive:  false,
+				Persistent: true,
+				Members:    []string{"model4"},
+			},
+		},
+	}
+
+	assert.Equal(t, expected, config)
+
+	realname, found := config.RealModelName("m1")
+	assert.True(t, found)
+	assert.Equal(t, "model1", realname)
+}

proxy/config/config_test.go

@@ -0,0 +1,763 @@
+package config
+
+import (
+	"slices"
+	"strings"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+)
+
+func TestConfig_GroupMemberIsUnique(t *testing.T) {
+	content := `
+models:
+  model1:
+    cmd: path/to/cmd --arg1 one
+    proxy: "http://localhost:8080"
+  model2:
+    cmd: path/to/cmd --arg1 one
+    proxy: "http://localhost:8081"
+    checkEndpoint: "/"
+  model3:
+    cmd: path/to/cmd --arg1 one
+    proxy: "http://localhost:8081"
+    checkEndpoint: "/"
+
+healthCheckTimeout: 15
+groups:
+  group1:
+    swap: true
+    exclusive: false
+    members: ["model2"]
+  group2:
+    swap: true
+    exclusive: false
+    members: ["model2"]
+`
+	// Load the config and verify
+	_, err := LoadConfigFromReader(strings.NewReader(content))
+
+	// a Contains as order of the map is not guaranteed
+	assert.Contains(t, err.Error(), "model member model2 is used in multiple groups:")
+}
+
+func TestConfig_ModelAliasesAreUnique(t *testing.T) {
+	content := `
+models:
+  model1:
+    cmd: path/to/cmd --arg1 one
+    proxy: "http://localhost:8080"
+    aliases:
+      - m1
+  model2:
+    cmd: path/to/cmd --arg1 one
+    proxy: "http://localhost:8081"
+    checkEndpoint: "/"
+    aliases:
+      - m1
+      - m2
+`
+	// Load the config and verify
+	_, err := LoadConfigFromReader(strings.NewReader(content))
+
+	// this is a contains because it could be `model1` or `model2` depending on the order
+	// go decided on the order of the map
+	assert.Contains(t, err.Error(), "duplicate alias m1 found in model: model")
+}
+
+func TestConfig_FindConfig(t *testing.T) {
+
+	// TODO?
+	// make make this shared between the different tests
+	config := &Config{
+		Models: map[string]ModelConfig{
+			"model1": {
+				Cmd:           "python model1.py",
+				Proxy:         "http://localhost:8080",
+				Aliases:       []string{"m1", "model-one"},
+				Env:           []string{"VAR1=value1", "VAR2=value2"},
+				CheckEndpoint: "/health",
+			},
+			"model2": {
+				Cmd:           "python model2.py",
+				Proxy:         "http://localhost:8081",
+				Aliases:       []string{"m2", "model-two"},
+				Env:           []string{"VAR3=value3", "VAR4=value4"},
+				CheckEndpoint: "/status",
+			},
+		},
+		HealthCheckTimeout: 10,
+		aliases: map[string]string{
+			"m1":        "model1",
+			"model-one": "model1",
+			"m2":        "model2",
+		},
+	}
+
+	// Test finding a model by its name
+	modelConfig, modelId, found := config.FindConfig("model1")
+	assert.True(t, found)
+	assert.Equal(t, "model1", modelId)
+	assert.Equal(t, config.Models["model1"], modelConfig)
+
+	// Test finding a model by its alias
+	modelConfig, modelId, found = config.FindConfig("m1")
+	assert.True(t, found)
+	assert.Equal(t, "model1", modelId)
+	assert.Equal(t, config.Models["model1"], modelConfig)
+
+	// Test finding a model that does not exist
+	modelConfig, modelId, found = config.FindConfig("model3")
+	assert.False(t, found)
+	assert.Equal(t, "", modelId)
+	assert.Equal(t, ModelConfig{}, modelConfig)
+}
+
+func TestConfig_AutomaticPortAssignments(t *testing.T) {
+
+	t.Run("Default Port Ranges", func(t *testing.T) {
+		content := ``
+		config, err := LoadConfigFromReader(strings.NewReader(content))
+		if !assert.NoError(t, err) {
+			t.Fatalf("Failed to load config: %v", err)
+		}
+
+		assert.Equal(t, 5800, config.StartPort)
+	})
+	t.Run("User specific port ranges", func(t *testing.T) {
+		content := `startPort: 1000`
+		config, err := LoadConfigFromReader(strings.NewReader(content))
+		if !assert.NoError(t, err) {
+			t.Fatalf("Failed to load config: %v", err)
+		}
+
+		assert.Equal(t, 1000, config.StartPort)
+	})
+
+	t.Run("Invalid start port", func(t *testing.T) {
+		content := `startPort: abcd`
+		_, err := LoadConfigFromReader(strings.NewReader(content))
+		assert.NotNil(t, err)
+	})
+
+	t.Run("start port must be greater than 1", func(t *testing.T) {
+		content := `startPort: -99`
+		_, err := LoadConfigFromReader(strings.NewReader(content))
+		assert.NotNil(t, err)
+	})
+
+	t.Run("Automatic port assignments", func(t *testing.T) {
+		content := `
+startPort: 5800
+models:
+  model1:
+    cmd: svr --port ${PORT}
+  model2:
+    cmd: svr --port ${PORT}
+    proxy: "http://172.11.22.33:${PORT}"
+  model3:
+    cmd: svr --port 1999
+    proxy: "http://1.2.3.4:1999"
+`
+		config, err := LoadConfigFromReader(strings.NewReader(content))
+		if !assert.NoError(t, err) {
+			t.Fatalf("Failed to load config: %v", err)
+		}
+
+		assert.Equal(t, 5800, config.StartPort)
+		assert.Equal(t, "svr --port 5800", config.Models["model1"].Cmd)
+		assert.Equal(t, "http://localhost:5800", config.Models["model1"].Proxy)
+
+		assert.Equal(t, "svr --port 5801", config.Models["model2"].Cmd)
+		assert.Equal(t, "http://172.11.22.33:5801", config.Models["model2"].Proxy)
+
+		assert.Equal(t, "svr --port 1999", config.Models["model3"].Cmd)
+		assert.Equal(t, "http://1.2.3.4:1999", config.Models["model3"].Proxy)
+
+	})
+
+	t.Run("Proxy value required if no ${PORT} in cmd", func(t *testing.T) {
+		content := `
+models:
+  model1:
+    cmd: svr --port 111
+`
+		_, err := LoadConfigFromReader(strings.NewReader(content))
+		assert.Equal(t, "model model1: proxy uses ${PORT} but cmd does not - ${PORT} is only available when used in cmd", err.Error())
+	})
+}
+
+func TestConfig_MacroReplacement(t *testing.T) {
+	content := `
+startPort: 9990
+macros:
+  svr-path: "path/to/server"
+  argOne: "--arg1"
+  argTwo: "--arg2"
+  autoPort: "--port ${PORT}"
+  overriddenByModelMacro: failed
+
+models:
+  model1:
+    macros:
+      overriddenByModelMacro: success
+    cmd: |
+      ${svr-path} ${argTwo}
+      # the automatic ${PORT} is replaced
+      ${autoPort}
+      ${argOne}
+      --arg3 three
+      --overridden ${overriddenByModelMacro}
+    cmdStop: |
+      /path/to/stop.sh --port ${PORT} ${argTwo}
+`
+
+	config, err := LoadConfigFromReader(strings.NewReader(content))
+	if !assert.NoError(t, err) {
+		t.FailNow()
+	}
+	sanitizedCmd, err := SanitizeCommand(config.Models["model1"].Cmd)
+	assert.NoError(t, err)
+	assert.Equal(t, "path/to/server --arg2 --port 9990 --arg1 --arg3 three --overridden success", strings.Join(sanitizedCmd, " "))
+
+	sanitizedCmdStop, err := SanitizeCommand(config.Models["model1"].CmdStop)
+	assert.NoError(t, err)
+	assert.Equal(t, "/path/to/stop.sh --port 9990 --arg2", strings.Join(sanitizedCmdStop, " "))
+}
+
+func TestConfig_MacroReservedNames(t *testing.T) {
+
+	tests := []struct {
+		name          string
+		config        string
+		expectedError string
+	}{
+		{
+			name: "global macro named PORT",
+			config: `
+macros:
+  PORT: "1111"
+`,
+			expectedError: "macro name 'PORT' is reserved",
+		},
+		{
+			name: "global macro named MODEL_ID",
+			config: `
+macros:
+  MODEL_ID: model1
+`,
+			expectedError: "macro name 'MODEL_ID' is reserved",
+		},
+		{
+			name: "model macro named PORT",
+			config: `
+models:
+  model1:
+    macros:
+      PORT: 1111
+`,
+			expectedError: "model model1: macro name 'PORT' is reserved",
+		},
+
+		{
+			name: "model macro named MODEL_ID",
+			config: `
+models:
+  model1:
+    macros:
+      MODEL_ID: model1
+`,
+			expectedError: "model model1: macro name 'MODEL_ID' is reserved",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			_, err := LoadConfigFromReader(strings.NewReader(tt.config))
+			assert.NotNil(t, err)
+			assert.Equal(t, tt.expectedError, err.Error())
+		})
+	}
+}
+
+func TestConfig_MacroErrorOnUnknownMacros(t *testing.T) {
+	tests := []struct {
+		name    string
+		field   string
+		content string
+	}{
+		{
+			name:  "unknown macro in cmd",
+			field: "cmd",
+			content: `
+startPort: 9990
+macros:
+  svr-path: "path/to/server"
+models:
+  model1:
+    cmd: |
+      ${svr-path} --port ${PORT}
+      ${unknownMacro}
+`,
+		},
+		{
+			name:  "unknown macro in cmdStop",
+			field: "cmdStop",
+			content: `
+startPort: 9990
+macros:
+  svr-path: "path/to/server"
+models:
+  model1:
+    cmd: "${svr-path} --port ${PORT}"
+    cmdStop: "kill ${unknownMacro}"
+`,
+		},
+		{
+			name:  "unknown macro in proxy",
+			field: "proxy",
+			content: `
+startPort: 9990
+macros:
+  svr-path: "path/to/server"
+models:
+  model1:
+    cmd: "${svr-path} --port ${PORT}"
+    proxy: "http://${unknownMacro}:${PORT}"
+`,
+		},
+		{
+			name:  "unknown macro in checkEndpoint",
+			field: "checkEndpoint",
+			content: `
+startPort: 9990
+macros:
+  svr-path: "path/to/server"
+models:
+  model1:
+    cmd: "${svr-path} --port ${PORT}"
+    checkEndpoint: "http://localhost:${unknownMacro}/health"
+`,
+		},
+		{
+			name:  "unknown macro in filters.stripParams",
+			field: "filters.stripParams",
+			content: `
+startPort: 9990
+macros:
+  svr-path: "path/to/server"
+models:
+  model1:
+    cmd: "${svr-path} --port ${PORT}"
+    filters:
+      stripParams: "model,${unknownMacro}"
+`,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			_, err := LoadConfigFromReader(strings.NewReader(tt.content))
+			if assert.Error(t, err) {
+				assert.Contains(t, err.Error(), "unknown macro '${unknownMacro}' found in model1."+tt.field)
+			}
+			//t.Log(err)
+		})
+	}
+}
+func TestStripComments(t *testing.T) {
+	tests := []struct {
+		name     string
+		input    string
+		expected string
+	}{
+		{
+			name:     "no comments",
+			input:    "echo hello\necho world",
+			expected: "echo hello\necho world",
+		},
+		{
+			name:     "single comment line",
+			input:    "# this is a comment\necho hello",
+			expected: "echo hello",
+		},
+		{
+			name:     "multiple comment lines",
+			input:    "# comment 1\necho hello\n# comment 2\necho world",
+			expected: "echo hello\necho world",
+		},
+		{
+			name:     "comment with spaces",
+			input:    "   # indented comment\necho hello",
+			expected: "echo hello",
+		},
+		{
+			name:     "empty lines preserved",
+			input:    "echo hello\n\necho world",
+			expected: "echo hello\n\necho world",
+		},
+		{
+			name:     "only comments",
+			input:    "# comment 1\n# comment 2",
+			expected: "",
+		},
+		{
+			name:     "empty string",
+			input:    "",
+			expected: "",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := StripComments(tt.input)
+			if result != tt.expected {
+				t.Errorf("StripComments() = %q, expected %q", result, tt.expected)
+			}
+		})
+	}
+}
+
+func TestConfig_MacroInCommentStrippedBeforeExpansion(t *testing.T) {
+	// Test case that reproduces the original bug where a macro in a comment
+	// would get expanded and cause the comment text to be included in the command
+	content := `
+startPort: 9990
+macros:
+  "latest-llama": >
+    /user/llama.cpp/build/bin/llama-server
+    --port ${PORT}
+
+models:
+  "test-model":
+    cmd: |
+      # ${latest-llama} is a macro that is defined above
+      ${latest-llama}
+      --model /path/to/model.gguf
+      -ngl 99
+`
+
+	config, err := LoadConfigFromReader(strings.NewReader(content))
+	assert.NoError(t, err)
+
+	// Get the sanitized command
+	sanitizedCmd, err := SanitizeCommand(config.Models["test-model"].Cmd)
+	assert.NoError(t, err)
+
+	// Join the command for easier inspection
+	cmdStr := strings.Join(sanitizedCmd, " ")
+
+	// Verify that comment text is NOT present in the final command as separate arguments
+	commentWords := []string{"is", "macro", "that", "defined", "above"}
+	for _, word := range commentWords {
+		found := slices.Contains(sanitizedCmd, word)
+		assert.False(t, found, "Comment text '%s' should not be present as a separate argument in final command", word)
+	}
+
+	// Verify that the actual command components ARE present
+	expectedParts := []string{
+		"/user/llama.cpp/build/bin/llama-server",
+		"--port",
+		"9990",
+		"--model",
+		"/path/to/model.gguf",
+		"-ngl",
+		"99",
+	}
+
+	for _, part := range expectedParts {
+		assert.Contains(t, cmdStr, part, "Expected command part '%s' not found in final command", part)
+	}
+
+	// Verify the server path appears exactly once (not duplicated due to macro expansion)
+	serverPath := "/user/llama.cpp/build/bin/llama-server"
+	count := strings.Count(cmdStr, serverPath)
+	assert.Equal(t, 1, count, "Expected exactly 1 occurrence of server path, found %d", count)
+
+	// Verify the expected final command structure
+	expectedCmd := "/user/llama.cpp/build/bin/llama-server --port 9990 --model /path/to/model.gguf -ngl 99"
+	assert.Equal(t, expectedCmd, cmdStr, "Final command does not match expected structure")
+}
+
+func TestConfig_MacroModelId(t *testing.T) {
+	content := `
+startPort: 9000
+macros:
+  "docker-llama": docker run --name ${MODEL_ID} -p ${PORT}:8080 docker_img
+  "docker-stop": docker stop ${MODEL_ID}
+
+models:
+  model1:
+    cmd: /path/to/server -p ${PORT} -hf ${MODEL_ID}
+
+  model2:
+    cmd: ${docker-llama}
+    cmdStop: ${docker-stop}
+
+  author/model:F16:
+    cmd: /path/to/server -p ${PORT} -hf ${MODEL_ID}
+    cmdStop: stop
+`
+
+	config, err := LoadConfigFromReader(strings.NewReader(content))
+	assert.NoError(t, err)
+	sanitizedCmd, err := SanitizeCommand(config.Models["model1"].Cmd)
+	assert.NoError(t, err)
+	assert.Equal(t, "/path/to/server -p 9001 -hf model1", strings.Join(sanitizedCmd, " "))
+
+	dockerStopMacro, found := config.Macros.Get("docker-stop")
+	assert.True(t, found)
+	assert.Equal(t, "docker stop ${MODEL_ID}", dockerStopMacro)
+
+	sanitizedCmd2, err := SanitizeCommand(config.Models["model2"].Cmd)
+	assert.NoError(t, err)
+	assert.Equal(t, "docker run --name model2 -p 9002:8080 docker_img", strings.Join(sanitizedCmd2, " "))
+
+	sanitizedCmdStop, err := SanitizeCommand(config.Models["model2"].CmdStop)
+	assert.NoError(t, err)
+	assert.Equal(t, "docker stop model2", strings.Join(sanitizedCmdStop, " "))
+
+	sanitizedCmd3, err := SanitizeCommand(config.Models["author/model:F16"].Cmd)
+	assert.NoError(t, err)
+	assert.Equal(t, "/path/to/server -p 9000 -hf author/model:F16", strings.Join(sanitizedCmd3, " "))
+}
+
+func TestConfig_TypedMacrosInMetadata(t *testing.T) {
+	content := `
+startPort: 10000
+macros:
+  PORT_NUM: 10001
+  TEMP: 0.7
+  ENABLED: true
+  NAME: "llama model"
+  CTX: 16384
+
+models:
+  test-model:
+    cmd: /path/to/server -p ${PORT}
+    metadata:
+      port: ${PORT_NUM}
+      temperature: ${TEMP}
+      enabled: ${ENABLED}
+      model_name: ${NAME}
+      context: ${CTX}
+      note: "Running on port ${PORT_NUM} with temp ${TEMP} and context ${CTX}"
+`
+
+	config, err := LoadConfigFromReader(strings.NewReader(content))
+	assert.NoError(t, err)
+
+	meta := config.Models["test-model"].Metadata
+	assert.NotNil(t, meta)
+
+	// Verify direct substitution preserves types
+	assert.Equal(t, 10001, meta["port"])
+	assert.Equal(t, 0.7, meta["temperature"])
+	assert.Equal(t, true, meta["enabled"])
+	assert.Equal(t, "llama model", meta["model_name"])
+	assert.Equal(t, 16384, meta["context"])
+
+	// Verify string interpolation converts to string
+	assert.Equal(t, "Running on port 10001 with temp 0.7 and context 16384", meta["note"])
+}
+
+func TestConfig_NestedStructuresInMetadata(t *testing.T) {
+	content := `
+startPort: 10000
+macros:
+  TEMP: 0.7
+
+models:
+  test-model:
+    cmd: /path/to/server -p ${PORT}
+    metadata:
+      config:
+        port: ${PORT}
+        temperature: ${TEMP}
+      tags: ["model:${MODEL_ID}", "port:${PORT}"]
+      nested:
+        deep:
+          value: ${TEMP}
+`
+
+	config, err := LoadConfigFromReader(strings.NewReader(content))
+	assert.NoError(t, err)
+
+	meta := config.Models["test-model"].Metadata
+	assert.NotNil(t, meta)
+
+	// Verify nested objects
+	configMap := meta["config"].(map[string]any)
+	assert.Equal(t, 10000, configMap["port"])
+	assert.Equal(t, 0.7, configMap["temperature"])
+
+	// Verify arrays
+	tags := meta["tags"].([]any)
+	assert.Equal(t, "model:test-model", tags[0])
+	assert.Equal(t, "port:10000", tags[1])
+
+	// Verify deeply nested structures
+	nested := meta["nested"].(map[string]any)
+	deep := nested["deep"].(map[string]any)
+	assert.Equal(t, 0.7, deep["value"])
+}
+
+func TestConfig_ModelLevelMacroPrecedenceInMetadata(t *testing.T) {
+	content := `
+startPort: 10000
+macros:
+  TEMP: 0.5
+  GLOBAL_VAL: "global"
+
+models:
+  test-model:
+    cmd: /path/to/server -p ${PORT}
+    macros:
+      TEMP: 0.9
+      LOCAL_VAL: "local"
+    metadata:
+      temperature: ${TEMP}
+      global: ${GLOBAL_VAL}
+      local: ${LOCAL_VAL}
+`
+
+	config, err := LoadConfigFromReader(strings.NewReader(content))
+	assert.NoError(t, err)
+
+	meta := config.Models["test-model"].Metadata
+	assert.NotNil(t, meta)
+
+	// Model-level macro should override global
+	assert.Equal(t, 0.9, meta["temperature"])
+	// Global macro should be accessible
+	assert.Equal(t, "global", meta["global"])
+	// Model-level macro should be accessible
+	assert.Equal(t, "local", meta["local"])
+}
+
+func TestConfig_UnknownMacroInMetadata(t *testing.T) {
+	content := `
+startPort: 10000
+models:
+  test-model:
+    cmd: /path/to/server -p ${PORT}
+    metadata:
+      value: ${UNKNOWN_MACRO}
+`
+
+	_, err := LoadConfigFromReader(strings.NewReader(content))
+	assert.Error(t, err)
+	assert.Contains(t, err.Error(), "test-model")
+	assert.Contains(t, err.Error(), "UNKNOWN_MACRO")
+}
+
+func TestConfig_InvalidMacroType(t *testing.T) {
+	content := `
+startPort: 10000
+macros:
+  INVALID:
+    nested: value
+
+models:
+  test-model:
+    cmd: /path/to/server -p ${PORT}
+`
+
+	_, err := LoadConfigFromReader(strings.NewReader(content))
+	assert.Error(t, err)
+	assert.Contains(t, err.Error(), "INVALID")
+	assert.Contains(t, err.Error(), "must be a scalar type")
+}
+
+func TestConfig_MacroTypeValidation(t *testing.T) {
+	tests := []struct {
+		name      string
+		yaml      string
+		shouldErr bool
+	}{
+		{
+			name: "string macro",
+			yaml: `
+startPort: 10000
+macros:
+  STR: "test"
+models:
+  test-model:
+    cmd: /path/to/server -p ${PORT}
+`,
+			shouldErr: false,
+		},
+		{
+			name: "int macro",
+			yaml: `
+startPort: 10000
+macros:
+  NUM: 42
+models:
+  test-model:
+    cmd: /path/to/server -p ${PORT}
+`,
+			shouldErr: false,
+		},
+		{
+			name: "float macro",
+			yaml: `
+startPort: 10000
+macros:
+  FLOAT: 3.14
+models:
+  test-model:
+    cmd: /path/to/server -p ${PORT}
+`,
+			shouldErr: false,
+		},
+		{
+			name: "bool macro",
+			yaml: `
+startPort: 10000
+macros:
+  BOOL: true
+models:
+  test-model:
+    cmd: /path/to/server -p ${PORT}
+`,
+			shouldErr: false,
+		},
+		{
+			name: "array macro (invalid)",
+			yaml: `
+startPort: 10000
+macros:
+  ARR: [1, 2, 3]
+models:
+  test-model:
+    cmd: /path/to/server -p ${PORT}
+`,
+			shouldErr: true,
+		},
+		{
+			name: "map macro (invalid)",
+			yaml: `
+startPort: 10000
+macros:
+  MAP:
+    key: value
+models:
+  test-model:
+    cmd: /path/to/server -p ${PORT}
+`,
+			shouldErr: true,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			_, err := LoadConfigFromReader(strings.NewReader(tt.yaml))
+			if tt.shouldErr {
+				assert.Error(t, err)
+			} else {
+				assert.NoError(t, err)
+			}
+		})
+	}
+}

proxy/config/config_windows_test.go

@@ -0,0 +1,231 @@
+//go:build windows
+
+package config
+
+import (
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+)
+
+func TestConfig_SanitizeCommand(t *testing.T) {
+	// does not support single quoted strings like in config_posix_test.go
+	args, err := SanitizeCommand(`python model1.py \
+
+	-a "double quotes" \
+	-s
+	--arg3 123 \
+
+	   # comment 2
+	--arg4 '"string in string"'
+
+
+
+	# this will get stripped out as well as the white space above
+	-c "'single quoted'"
+	`)
+	assert.NoError(t, err)
+	assert.Equal(t, []string{
+		"python", "model1.py",
+		"-a", "double quotes",
+		"-s",
+		"--arg3", "123",
+		"--arg4", "'string in string'", // this is a little weird but the lexer says so...?
+		"-c", `'single quoted'`,
+	}, args)
+
+	// Test an empty command
+	args, err = SanitizeCommand("")
+	assert.Error(t, err)
+	assert.Nil(t, args)
+}
+
+func TestConfig_DefaultValuesWindows(t *testing.T) {
+	content := `
+models:
+  model1:
+    cmd: path/to/cmd --port ${PORT}
+`
+
+	config, err := LoadConfigFromReader(strings.NewReader(content))
+	assert.NoError(t, err)
+	assert.Equal(t, 120, config.HealthCheckTimeout)
+	assert.Equal(t, 5800, config.StartPort)
+	assert.Equal(t, "info", config.LogLevel)
+
+	// Test default group exists
+	defaultGroup, exists := config.Groups["(default)"]
+	assert.True(t, exists, "default group should exist")
+	if assert.NotNil(t, defaultGroup, "default group should not be nil") {
+		assert.Equal(t, true, defaultGroup.Swap)
+		assert.Equal(t, true, defaultGroup.Exclusive)
+		assert.Equal(t, false, defaultGroup.Persistent)
+		assert.Equal(t, []string{"model1"}, defaultGroup.Members)
+	}
+
+	model1, exists := config.Models["model1"]
+	assert.True(t, exists, "model1 should exist")
+	if assert.NotNil(t, model1, "model1 should not be nil") {
+		assert.Equal(t, "path/to/cmd --port 5800", model1.Cmd) // has the port replaced
+		assert.Equal(t, "taskkill /f /t /pid ${PID}", model1.CmdStop)
+		assert.Equal(t, "http://localhost:5800", model1.Proxy)
+		assert.Equal(t, "/health", model1.CheckEndpoint)
+		assert.Equal(t, []string{}, model1.Aliases)
+		assert.Equal(t, []string{}, model1.Env)
+		assert.Equal(t, 0, model1.UnloadAfter)
+		assert.Equal(t, false, model1.Unlisted)
+		assert.Equal(t, "", model1.UseModelName)
+		assert.Equal(t, 0, model1.ConcurrencyLimit)
+	}
+
+	// default empty filter exists
+	assert.Equal(t, "", model1.Filters.StripParams)
+}
+
+func TestConfig_LoadWindows(t *testing.T) {
+	// Create a temporary YAML file for testing
+	tempDir, err := os.MkdirTemp("", "test-config")
+	if err != nil {
+		t.Fatalf("Failed to create temporary directory: %v", err)
+	}
+	defer os.RemoveAll(tempDir)
+
+	tempFile := filepath.Join(tempDir, "config.yaml")
+	content := `
+macros:
+  svr-path: "path/to/server"
+models:
+  model1:
+    cmd: path/to/cmd --arg1 one
+    proxy: "http://localhost:8080"
+    aliases:
+      - "m1"
+      - "model-one"
+    env:
+      - "VAR1=value1"
+      - "VAR2=value2"
+    checkEndpoint: "/health"
+  model2:
+    cmd: ${svr-path} --arg1 one
+    proxy: "http://localhost:8081"
+    aliases:
+      - "m2"
+    checkEndpoint: "/"
+  model3:
+    cmd: path/to/cmd --arg1 one
+    proxy: "http://localhost:8081"
+    aliases:
+      - "mthree"
+    checkEndpoint: "/"
+  model4:
+    cmd: path/to/cmd --arg1 one
+    proxy: "http://localhost:8082"
+    checkEndpoint: "/"
+
+healthCheckTimeout: 15
+profiles:
+  test:
+    - model1
+    - model2
+groups:
+  group1:
+    swap: true
+    exclusive: false
+    members: ["model2"]
+  forever:
+    exclusive: false
+    persistent: true
+    members:
+      - "model4"
+`
+
+	if err := os.WriteFile(tempFile, []byte(content), 0644); err != nil {
+		t.Fatalf("Failed to write temporary file: %v", err)
+	}
+
+	// Load the config and verify
+	config, err := LoadConfig(tempFile)
+	if err != nil {
+		t.Fatalf("Failed to load config: %v", err)
+	}
+
+	expected := Config{
+		LogLevel:  "info",
+		StartPort: 5800,
+		Macros: MacroList{
+			{"svr-path", "path/to/server"},
+		},
+		Models: map[string]ModelConfig{
+			"model1": {
+				Cmd:           "path/to/cmd --arg1 one",
+				CmdStop:       "taskkill /f /t /pid ${PID}",
+				Proxy:         "http://localhost:8080",
+				Aliases:       []string{"m1", "model-one"},
+				Env:           []string{"VAR1=value1", "VAR2=value2"},
+				CheckEndpoint: "/health",
+			},
+			"model2": {
+				Cmd:           "path/to/server --arg1 one",
+				CmdStop:       "taskkill /f /t /pid ${PID}",
+				Proxy:         "http://localhost:8081",
+				Aliases:       []string{"m2"},
+				Env:           []string{},
+				CheckEndpoint: "/",
+			},
+			"model3": {
+				Cmd:           "path/to/cmd --arg1 one",
+				CmdStop:       "taskkill /f /t /pid ${PID}",
+				Proxy:         "http://localhost:8081",
+				Aliases:       []string{"mthree"},
+				Env:           []string{},
+				CheckEndpoint: "/",
+			},
+			"model4": {
+				Cmd:           "path/to/cmd --arg1 one",
+				CmdStop:       "taskkill /f /t /pid ${PID}",
+				Proxy:         "http://localhost:8082",
+				CheckEndpoint: "/",
+				Aliases:       []string{},
+				Env:           []string{},
+			},
+		},
+		HealthCheckTimeout: 15,
+		MetricsMaxInMemory: 1000,
+		Profiles: map[string][]string{
+			"test": {"model1", "model2"},
+		},
+		aliases: map[string]string{
+			"m1":        "model1",
+			"model-one": "model1",
+			"m2":        "model2",
+			"mthree":    "model3",
+		},
+		Groups: map[string]GroupConfig{
+			DEFAULT_GROUP_ID: {
+				Swap:      true,
+				Exclusive: true,
+				Members:   []string{"model1", "model3"},
+			},
+			"group1": {
+				Swap:      true,
+				Exclusive: false,
+				Members:   []string{"model2"},
+			},
+			"forever": {
+				Swap:       true,
+				Exclusive:  false,
+				Persistent: true,
+				Members:    []string{"model4"},
+			},
+		},
+	}
+
+	assert.Equal(t, expected, config)
+
+	realname, found := config.RealModelName("m1")
+	assert.True(t, found)
+	assert.Equal(t, "model1", realname)
+}

proxy/config/macro_in_macro_test.go

@@ -0,0 +1,123 @@
+package config
+
+import (
+	"strings"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+)
+
+// Test macro-in-macro basic substitution
+func TestConfig_MacroInMacroBasic(t *testing.T) {
+	content := `
+startPort: 10000
+macros:
+  "A": "value-A"
+  "B": "prefix-${A}-suffix"
+
+models:
+  test:
+    cmd: echo ${B}
+    proxy: http://localhost:8080
+`
+
+	config, err := LoadConfigFromReader(strings.NewReader(content))
+	assert.NoError(t, err)
+	assert.Equal(t, "echo prefix-value-A-suffix", config.Models["test"].Cmd)
+}
+
+// Test LIFO substitution order with 3+ macro levels
+func TestConfig_MacroInMacroLIFOOrder(t *testing.T) {
+	content := `
+startPort: 10000
+macros:
+  "base": "/models"
+  "path": "${base}/llama"
+  "full": "${path}/model.gguf"
+
+models:
+  test:
+    cmd: load ${full}
+    proxy: http://localhost:8080
+`
+
+	config, err := LoadConfigFromReader(strings.NewReader(content))
+	assert.NoError(t, err)
+	assert.Equal(t, "load /models/llama/model.gguf", config.Models["test"].Cmd)
+}
+
+// Test MODEL_ID in global macro used by model
+func TestConfig_ModelIdInGlobalMacro(t *testing.T) {
+	content := `
+startPort: 10000
+macros:
+  "podman-llama": "podman run --name ${MODEL_ID} ghcr.io/ggml-org/llama.cpp:server-cuda"
+
+models:
+  my-model:
+    cmd: ${podman-llama} -m model.gguf
+    proxy: http://localhost:8080
+`
+
+	config, err := LoadConfigFromReader(strings.NewReader(content))
+	assert.NoError(t, err)
+	assert.Equal(t, "podman run --name my-model ghcr.io/ggml-org/llama.cpp:server-cuda -m model.gguf", config.Models["my-model"].Cmd)
+}
+
+// Test model macro overrides global macro in substitution
+func TestConfig_ModelMacroOverridesGlobal(t *testing.T) {
+	content := `
+startPort: 10000
+macros:
+  "tag": "global"
+  "msg": "value-${tag}"
+
+models:
+  test:
+    macros:
+      "tag": "model-level"
+    cmd: echo ${msg}
+    proxy: http://localhost:8080
+`
+
+	config, err := LoadConfigFromReader(strings.NewReader(content))
+	assert.NoError(t, err)
+	assert.Equal(t, "echo value-model-level", config.Models["test"].Cmd)
+}
+
+// Test self-reference detection error
+func TestConfig_SelfReferenceDetection(t *testing.T) {
+	content := `
+startPort: 10000
+macros:
+  "recursive": "value-${recursive}"
+
+models:
+  test:
+    cmd: echo ${recursive}
+    proxy: http://localhost:8080
+`
+
+	_, err := LoadConfigFromReader(strings.NewReader(content))
+	assert.Error(t, err)
+	assert.Contains(t, err.Error(), "recursive")
+	assert.Contains(t, err.Error(), "self-reference")
+}
+
+// Test undefined macro reference error
+func TestConfig_UndefinedMacroReference(t *testing.T) {
+	content := `
+startPort: 10000
+macros:
+  "A": "value-${UNDEFINED}"
+
+models:
+  test:
+    cmd: echo ${A}
+    proxy: http://localhost:8080
+`
+
+	_, err := LoadConfigFromReader(strings.NewReader(content))
+	assert.Error(t, err)
+	assert.Contains(t, err.Error(), "UNDEFINED")
+}

proxy/config/model_config.go

@@ -0,0 +1,125 @@
+package config
+
+import (
+	"errors"
+	"runtime"
+	"slices"
+	"strings"
+)
+
+type ModelConfig struct {
+	Cmd           string   `yaml:"cmd"`
+	CmdStop       string   `yaml:"cmdStop"`
+	Proxy         string   `yaml:"proxy"`
+	Aliases       []string `yaml:"aliases"`
+	Env           []string `yaml:"env"`
+	CheckEndpoint string   `yaml:"checkEndpoint"`
+	UnloadAfter   int      `yaml:"ttl"`
+	Unlisted      bool     `yaml:"unlisted"`
+	UseModelName  string   `yaml:"useModelName"`
+
+	// #179 for /v1/models
+	Name        string `yaml:"name"`
+	Description string `yaml:"description"`
+
+	// Limit concurrency of HTTP requests to process
+	ConcurrencyLimit int `yaml:"concurrencyLimit"`
+
+	// Model filters see issue #174
+	Filters ModelFilters `yaml:"filters"`
+
+	// Macros: see #264
+	// Model level macros take precedence over the global macros
+	Macros MacroList `yaml:"macros"`
+
+	// Metadata: see #264
+	// Arbitrary metadata that can be exposed through the API
+	Metadata map[string]any `yaml:"metadata"`
+}
+
+func (m *ModelConfig) UnmarshalYAML(unmarshal func(interface{}) error) error {
+	type rawModelConfig ModelConfig
+	defaults := rawModelConfig{
+		Cmd:              "",
+		CmdStop:          "",
+		Proxy:            "http://localhost:${PORT}",
+		Aliases:          []string{},
+		Env:              []string{},
+		CheckEndpoint:    "/health",
+		UnloadAfter:      0,
+		Unlisted:         false,
+		UseModelName:     "",
+		ConcurrencyLimit: 0,
+		Name:             "",
+		Description:      "",
+	}
+
+	// the default cmdStop to taskkill /f /t /pid ${PID}
+	if runtime.GOOS == "windows" {
+		defaults.CmdStop = "taskkill /f /t /pid ${PID}"
+	}
+
+	if err := unmarshal(&defaults); err != nil {
+		return err
+	}
+
+	*m = ModelConfig(defaults)
+	return nil
+}
+
+func (m *ModelConfig) SanitizedCommand() ([]string, error) {
+	return SanitizeCommand(m.Cmd)
+}
+
+// ModelFilters see issue #174
+type ModelFilters struct {
+	StripParams string `yaml:"stripParams"`
+}
+
+func (m *ModelFilters) UnmarshalYAML(unmarshal func(interface{}) error) error {
+	type rawModelFilters ModelFilters
+	defaults := rawModelFilters{
+		StripParams: "",
+	}
+
+	if err := unmarshal(&defaults); err != nil {
+		return err
+	}
+
+	// Try to unmarshal with the old field name for backwards compatibility
+	if defaults.StripParams == "" {
+		var legacy struct {
+			StripParams string `yaml:"strip_params"`
+		}
+		if legacyErr := unmarshal(&legacy); legacyErr != nil {
+			return errors.New("failed to unmarshal legacy filters.strip_params: " + legacyErr.Error())
+		}
+		defaults.StripParams = legacy.StripParams
+	}
+
+	*m = ModelFilters(defaults)
+	return nil
+}
+
+func (f ModelFilters) SanitizedStripParams() ([]string, error) {
+	if f.StripParams == "" {
+		return nil, nil
+	}
+
+	params := strings.Split(f.StripParams, ",")
+	cleaned := make([]string, 0, len(params))
+	seen := make(map[string]bool)
+
+	for _, param := range params {
+		trimmed := strings.TrimSpace(param)
+		if trimmed == "model" || trimmed == "" || seen[trimmed] {
+			continue
+		}
+		seen[trimmed] = true
+		cleaned = append(cleaned, trimmed)
+	}
+
+	// sort cleaned
+	slices.Sort(cleaned)
+	return cleaned, nil
+}

proxy/config/model_config_test.go

@@ -0,0 +1,54 @@
+package config
+
+import (
+	"fmt"
+	"strings"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+)
+
+func TestConfig_ModelConfigSanitizedCommand(t *testing.T) {
+	config := &ModelConfig{
+		Cmd: `python model1.py \
+    --arg1 value1 \
+    --arg2 value2`,
+	}
+
+	args, err := config.SanitizedCommand()
+	assert.NoError(t, err)
+	assert.Equal(t, []string{"python", "model1.py", "--arg1", "value1", "--arg2", "value2"}, args)
+}
+
+func TestConfig_ModelFilters(t *testing.T) {
+	content := `
+macros:
+  default_strip: "temperature, top_p"
+models:
+  model1:
+    cmd: path/to/cmd --port ${PORT}
+    filters:
+      # macros inserted and list is cleaned of duplicates and empty strings
+      stripParams: "model, top_k, top_k, temperature, ${default_strip}, , ,"
+  # check for strip_params (legacy field name) compatibility
+  legacy:
+    cmd: path/to/cmd --port ${PORT}
+    filters:
+      strip_params: "model, top_k, top_k, temperature, ${default_strip}, , ,"
+`
+	config, err := LoadConfigFromReader(strings.NewReader(content))
+	assert.NoError(t, err)
+	for modelId, modelConfig := range config.Models {
+		t.Run(fmt.Sprintf("Testing macros in filters for model %s", modelId), func(t *testing.T) {
+			assert.Equal(t, "model, top_k, top_k, temperature, temperature, top_p, , ,", modelConfig.Filters.StripParams)
+			sanitized, err := modelConfig.Filters.SanitizedStripParams()
+			if assert.NoError(t, err) {
+				// model has been removed
+				// empty strings have been removed
+				// duplicates have been removed
+				assert.Equal(t, []string{"temperature", "top_k", "top_p"}, sanitized)
+			}
+		})
+	}
+
+}

proxy/discardWriter.go

@@ -0,0 +1,27 @@
+package proxy
+
+import "net/http"
+
+// Custom discard writer that implements http.ResponseWriter but just discards everything
+type DiscardWriter struct {
+	header http.Header
+	status int
+}
+
+func (w *DiscardWriter) Header() http.Header {
+	if w.header == nil {
+		w.header = make(http.Header)
+	}
+	return w.header
+}
+
+func (w *DiscardWriter) Write(data []byte) (int, error) {
+	return len(data), nil
+}
+
+func (w *DiscardWriter) WriteHeader(code int) {
+	w.status = code
+}
+
+// Satisfy the http.Flusher interface for streaming responses
+func (w *DiscardWriter) Flush() {}

proxy/events.go

@@ -0,0 +1,60 @@
+package proxy
+
+// package level registry of the different event types
+
+const ProcessStateChangeEventID = 0x01
+const ChatCompletionStatsEventID = 0x02
+const ConfigFileChangedEventID = 0x03
+const LogDataEventID = 0x04
+const TokenMetricsEventID = 0x05
+const ModelPreloadedEventID = 0x06
+
+type ProcessStateChangeEvent struct {
+	ProcessName string
+	NewState    ProcessState
+	OldState    ProcessState
+}
+
+func (e ProcessStateChangeEvent) Type() uint32 {
+	return ProcessStateChangeEventID
+}
+
+type ChatCompletionStats struct {
+	TokensGenerated int
+}
+
+func (e ChatCompletionStats) Type() uint32 {
+	return ChatCompletionStatsEventID
+}
+
+type ReloadingState int
+
+const (
+	ReloadingStateStart ReloadingState = iota
+	ReloadingStateEnd
+)
+
+type ConfigFileChangedEvent struct {
+	ReloadingState ReloadingState
+}
+
+func (e ConfigFileChangedEvent) Type() uint32 {
+	return ConfigFileChangedEventID
+}
+
+type LogDataEvent struct {
+	Data []byte
+}
+
+func (e LogDataEvent) Type() uint32 {
+	return LogDataEventID
+}
+
+type ModelPreloadedEvent struct {
+	ModelName string
+	Success   bool
+}
+
+func (e ModelPreloadedEvent) Type() uint32 {
+	return ModelPreloadedEventID
+}

proxy/helpers_test.go

@@ -0,0 +1,86 @@
+package proxy
+
+import (
+	"fmt"
+	"os"
+	"path/filepath"
+	"runtime"
+	"sync"
+	"testing"
+
+	"github.com/gin-gonic/gin"
+	"github.com/mostlygeek/llama-swap/proxy/config"
+	"gopkg.in/yaml.v3"
+)
+
+var (
+	nextTestPort        int = 12000
+	portMutex           sync.Mutex
+	testLogger          = NewLogMonitorWriter(os.Stdout)
+	simpleResponderPath = getSimpleResponderPath()
+)
+
+// Check if the binary exists
+func TestMain(m *testing.M) {
+	binaryPath := getSimpleResponderPath()
+	if _, err := os.Stat(binaryPath); os.IsNotExist(err) {
+		fmt.Printf("simple-responder not found at %s, did you `make simple-responder`?\n", binaryPath)
+		os.Exit(1)
+	}
+
+	gin.SetMode(gin.TestMode)
+
+	switch os.Getenv("LOG_LEVEL") {
+	case "debug":
+		testLogger.SetLogLevel(LevelDebug)
+	case "warn":
+		testLogger.SetLogLevel(LevelWarn)
+	case "info":
+		testLogger.SetLogLevel(LevelInfo)
+	default:
+		testLogger.SetLogLevel(LevelWarn)
+	}
+
+	m.Run()
+}
+
+// Helper function to get the binary path
+func getSimpleResponderPath() string {
+	goos := runtime.GOOS
+	goarch := runtime.GOARCH
+
+	if goos == "windows" {
+		return filepath.Join("..", "build", "simple-responder.exe")
+	} else {
+		return filepath.Join("..", "build", fmt.Sprintf("simple-responder_%s_%s", goos, goarch))
+	}
+}
+
+func getTestPort() int {
+	portMutex.Lock()
+	defer portMutex.Unlock()
+
+	port := nextTestPort
+	nextTestPort++
+
+	return port
+}
+
+func getTestSimpleResponderConfig(expectedMessage string) config.ModelConfig {
+	return getTestSimpleResponderConfigPort(expectedMessage, getTestPort())
+}
+
+func getTestSimpleResponderConfigPort(expectedMessage string, port int) config.ModelConfig {
+	// Create a YAML string with just the values we want to set
+	yamlStr := fmt.Sprintf(`
+cmd: '%s --port %d --silent --respond %s'
+proxy: "http://127.0.0.1:%d"
+`, simpleResponderPath, port, expectedMessage, port)
+
+	var cfg config.ModelConfig
+	if err := yaml.Unmarshal([]byte(yamlStr), &cfg); err != nil {
+		panic(fmt.Sprintf("failed to unmarshal test config: %v in [%s]", err, yamlStr))
+	}
+
+	return cfg
+}

proxy/logMonitor.go

@@ -0,0 +1,169 @@
+package proxy
+
+import (
+	"container/ring"
+	"context"
+	"fmt"
+	"io"
+	"os"
+	"sync"
+
+	"github.com/mostlygeek/llama-swap/event"
+)
+
+type LogLevel int
+
+const (
+	LevelDebug LogLevel = iota
+	LevelInfo
+	LevelWarn
+	LevelError
+)
+
+type LogMonitor struct {
+	eventbus *event.Dispatcher
+	mu       sync.RWMutex
+	buffer   *ring.Ring
+	bufferMu sync.RWMutex
+
+	// typically this can be os.Stdout
+	stdout io.Writer
+
+	// logging levels
+	level  LogLevel
+	prefix string
+}
+
+func NewLogMonitor() *LogMonitor {
+	return NewLogMonitorWriter(os.Stdout)
+}
+
+func NewLogMonitorWriter(stdout io.Writer) *LogMonitor {
+	return &LogMonitor{
+		eventbus: event.NewDispatcherConfig(1000),
+		buffer:   ring.New(10 * 1024), // keep 10KB of buffered logs
+		stdout:   stdout,
+		level:    LevelInfo,
+		prefix:   "",
+	}
+}
+
+func (w *LogMonitor) Write(p []byte) (n int, err error) {
+	if len(p) == 0 {
+		return 0, nil
+	}
+
+	n, err = w.stdout.Write(p)
+	if err != nil {
+		return n, err
+	}
+
+	w.bufferMu.Lock()
+	bufferCopy := make([]byte, len(p))
+	copy(bufferCopy, p)
+	w.buffer.Value = bufferCopy
+	w.buffer = w.buffer.Next()
+	w.bufferMu.Unlock()
+
+	w.broadcast(bufferCopy)
+	return n, nil
+}
+
+func (w *LogMonitor) GetHistory() []byte {
+	w.bufferMu.RLock()
+	defer w.bufferMu.RUnlock()
+
+	var history []byte
+	w.buffer.Do(func(p any) {
+		if p != nil {
+			if content, ok := p.([]byte); ok {
+				history = append(history, content...)
+			}
+		}
+	})
+	return history
+}
+
+func (w *LogMonitor) OnLogData(callback func(data []byte)) context.CancelFunc {
+	return event.Subscribe(w.eventbus, func(e LogDataEvent) {
+		callback(e.Data)
+	})
+}
+
+func (w *LogMonitor) broadcast(msg []byte) {
+	event.Publish(w.eventbus, LogDataEvent{Data: msg})
+}
+
+func (w *LogMonitor) SetPrefix(prefix string) {
+	w.mu.Lock()
+	defer w.mu.Unlock()
+	w.prefix = prefix
+}
+
+func (w *LogMonitor) SetLogLevel(level LogLevel) {
+	w.mu.Lock()
+	defer w.mu.Unlock()
+	w.level = level
+}
+
+func (w *LogMonitor) formatMessage(level string, msg string) []byte {
+	prefix := ""
+	if w.prefix != "" {
+		prefix = fmt.Sprintf("[%s] ", w.prefix)
+	}
+	return []byte(fmt.Sprintf("%s[%s] %s\n", prefix, level, msg))
+}
+
+func (w *LogMonitor) log(level LogLevel, msg string) {
+	if level < w.level {
+		return
+	}
+	w.Write(w.formatMessage(level.String(), msg))
+}
+
+func (w *LogMonitor) Debug(msg string) {
+	w.log(LevelDebug, msg)
+}
+
+func (w *LogMonitor) Info(msg string) {
+	w.log(LevelInfo, msg)
+}
+
+func (w *LogMonitor) Warn(msg string) {
+	w.log(LevelWarn, msg)
+}
+
+func (w *LogMonitor) Error(msg string) {
+	w.log(LevelError, msg)
+}
+
+func (w *LogMonitor) Debugf(format string, args ...interface{}) {
+	w.log(LevelDebug, fmt.Sprintf(format, args...))
+}
+
+func (w *LogMonitor) Infof(format string, args ...interface{}) {
+	w.log(LevelInfo, fmt.Sprintf(format, args...))
+}
+
+func (w *LogMonitor) Warnf(format string, args ...interface{}) {
+	w.log(LevelWarn, fmt.Sprintf(format, args...))
+}
+
+func (w *LogMonitor) Errorf(format string, args ...interface{}) {
+	w.log(LevelError, fmt.Sprintf(format, args...))
+}
+
+func (l LogLevel) String() string {
+	switch l {
+	case LevelDebug:
+		return "DEBUG"
+	case LevelInfo:
+		return "INFO"
+	case LevelWarn:
+		return "WARN"
+	case LevelError:
+		return "ERROR"
+	default:
+		return "UNKNOWN"
+	}
+}

proxy/logMonitor_test.go

@@ -0,0 +1,86 @@
+package proxy
+
+import (
+	"bytes"
+	"io"
+	"sync"
+	"testing"
+)
+
+func TestLogMonitor(t *testing.T) {
+	logMonitor := NewLogMonitorWriter(io.Discard)
+
+	// A WaitGroup is used to wait for all the expected writes to complete
+	var wg sync.WaitGroup
+
+	client1Messages := make([]byte, 0)
+	client2Messages := make([]byte, 0)
+
+	defer logMonitor.OnLogData(func(data []byte) {
+		client1Messages = append(client1Messages, data...)
+		wg.Done()
+	})()
+
+	defer logMonitor.OnLogData(func(data []byte) {
+		client2Messages = append(client2Messages, data...)
+		wg.Done()
+	})()
+
+	wg.Add(6) // 2 x 3 writes
+
+	logMonitor.Write([]byte("1"))
+	logMonitor.Write([]byte("2"))
+	logMonitor.Write([]byte("3"))
+
+	// wait for all writes to complete
+	wg.Wait()
+
+	// Check the buffer
+	expectedHistory := "123"
+	history := string(logMonitor.GetHistory())
+
+	if history != expectedHistory {
+		t.Errorf("Expected history: %s, got: %s", expectedHistory, history)
+	}
+
+	c1Data := string(client1Messages)
+	if c1Data != expectedHistory {
+		t.Errorf("Client1 expected %s, got: %s", expectedHistory, c1Data)
+	}
+
+	c2Data := string(client2Messages)
+	if c2Data != expectedHistory {
+		t.Errorf("Client2 expected %s, got: %s", expectedHistory, c2Data)
+	}
+}
+
+func TestWrite_ImmutableBuffer(t *testing.T) {
+	// Create a new LogMonitor instance
+	lm := NewLogMonitorWriter(io.Discard)
+
+	// Prepare a message to write
+	msg := []byte("Hello, World!")
+	lenmsg := len(msg)
+
+	// Write the message to the LogMonitor
+	n, err := lm.Write(msg)
+	if err != nil {
+		t.Fatalf("Write failed: %v", err)
+	}
+
+	if n != lenmsg {
+		t.Errorf("Expected %d bytes written but got %d", lenmsg, n)
+	}
+
+	// Change the original message
+	msg[0] = 'B' // This should not affect the buffer
+
+	// Get the history from the LogMonitor
+	history := lm.GetHistory()
+
+	// Check that the history contains the original message, not the modified one
+	expected := []byte("Hello, World!")
+	if !bytes.Equal(history, expected) {
+		t.Errorf("Expected history to be %q, got %q", expected, history)
+	}
+}

proxy/manager.go

@@ -1,211 +0,0 @@
-package proxy
-
-import (
-	"bytes"
-	"context"
-	"encoding/json"
-	"fmt"
-	"io"
-	"net/http"
-	"os"
-	"os/exec"
-	"strings"
-	"sync"
-	"syscall"
-	"time"
-)
-
-type ProxyManager struct {
-	sync.Mutex
-
-	config        *Config
-	currentCmd    *exec.Cmd
-	currentConfig ModelConfig
-}
-
-func New(config *Config) *ProxyManager {
-	return &ProxyManager{config: config}
-}
-
-func (pm *ProxyManager) HandleFunc(w http.ResponseWriter, r *http.Request) {
-
-	// https://github.com/ggerganov/llama.cpp/blob/master/examples/server/README.md#api-endpoints
-
-	if r.URL.Path == "/v1/chat/completions" {
-		// extracts the `model` from json body
-		pm.proxyChatRequest(w, r)
-	} else {
-		pm.proxyRequest(w, r)
-	}
-}
-
-func (pm *ProxyManager) swapModel(requestedModel string) error {
-	pm.Lock()
-	defer pm.Unlock()
-
-	// find the model configuration matching requestedModel
-	modelConfig, found := pm.config.FindConfig(requestedModel)
-	if !found {
-		return fmt.Errorf("could not find configuration for %s", requestedModel)
-	}
-
-	// no need to swap llama.cpp instances
-	if pm.currentConfig.Cmd == modelConfig.Cmd {
-		return nil
-	}
-
-	// kill the current running one to swap it
-	if pm.currentCmd != nil {
-		pm.currentCmd.Process.Signal(syscall.SIGTERM)
-
-		// wait for it to end
-		pm.currentCmd.Process.Wait()
-	}
-
-	pm.currentConfig = modelConfig
-
-	args := strings.Fields(modelConfig.Cmd)
-	cmd := exec.Command(args[0], args[1:]...)
-	cmd.Stdout = os.Stdout
-	cmd.Stderr = os.Stderr
-	cmd.Env = modelConfig.Env
-
-	err := cmd.Start()
-	if err != nil {
-		return err
-	}
-	pm.currentCmd = cmd
-
-	if err := pm.checkHealthEndpoint(); err != nil {
-		return err
-	}
-
-	return nil
-}
-
-func (pm *ProxyManager) checkHealthEndpoint() error {
-
-	if pm.currentConfig.Proxy == "" {
-		return fmt.Errorf("no upstream available to check /health")
-	}
-
-	proxyTo := pm.currentConfig.Proxy
-
-	maxDuration := time.Second * time.Duration(pm.config.HealthCheckTimeout)
-
-	healthURL := proxyTo + "/health"
-	client := &http.Client{}
-	startTime := time.Now()
-
-	for {
-		req, err := http.NewRequest("GET", healthURL, nil)
-		if err != nil {
-			return err
-		}
-		ctx, cancel := context.WithTimeout(req.Context(), 250*time.Millisecond)
-		defer cancel()
-		req = req.WithContext(ctx)
-		resp, err := client.Do(req)
-		if err != nil {
-			if strings.Contains(err.Error(), "connection refused") {
-
-				// if TCP dial can't connect any HTTP response after 5 seconds
-				// exit quickly.
-				if time.Since(startTime) > 5*time.Second {
-					return fmt.Errorf("/healthy endpoint took more than 5 seconds to respond")
-				}
-			}
-
-			if time.Since(startTime) >= maxDuration {
-				return fmt.Errorf("failed to check /healthy from: %s", healthURL)
-			}
-			time.Sleep(time.Second)
-			continue
-		}
-		defer resp.Body.Close()
-		if resp.StatusCode == http.StatusOK {
-			return nil
-		}
-		if time.Since(startTime) >= maxDuration {
-			return fmt.Errorf("failed to check /healthy from: %s", healthURL)
-		}
-		time.Sleep(time.Second)
-	}
-}
-
-func (pm *ProxyManager) proxyChatRequest(w http.ResponseWriter, r *http.Request) {
-	bodyBytes, err := io.ReadAll(r.Body)
-	if err != nil {
-		http.Error(w, "Invalid JSON", http.StatusBadRequest)
-		return
-	}
-	var requestBody map[string]interface{}
-	if err := json.Unmarshal(bodyBytes, &requestBody); err != nil {
-		http.Error(w, "Invalid JSON", http.StatusBadRequest)
-		return
-	}
-	model, ok := requestBody["model"].(string)
-	if !ok {
-		http.Error(w, "Missing or invalid 'model' key", http.StatusBadRequest)
-		return
-	}
-
-	if err := pm.swapModel(model); err != nil {
-		http.Error(w, fmt.Sprintf("unable to swap to model: %s", err.Error()), http.StatusNotFound)
-		return
-	}
-
-	r.Body = io.NopCloser(bytes.NewBuffer(bodyBytes))
-	pm.proxyRequest(w, r)
-}
-
-func (pm *ProxyManager) proxyRequest(w http.ResponseWriter, r *http.Request) {
-	if pm.currentConfig.Proxy == "" {
-		http.Error(w, "No upstream proxy", http.StatusInternalServerError)
-		return
-	}
-
-	proxyTo := pm.currentConfig.Proxy
-
-	client := &http.Client{}
-	req, err := http.NewRequest(r.Method, proxyTo+r.URL.String(), r.Body)
-	if err != nil {
-		http.Error(w, err.Error(), http.StatusInternalServerError)
-		return
-	}
-	req.Header = r.Header
-	resp, err := client.Do(req)
-	if err != nil {
-		http.Error(w, err.Error(), http.StatusBadGateway)
-		return
-	}
-	defer resp.Body.Close()
-	for k, vv := range resp.Header {
-		for _, v := range vv {
-			w.Header().Add(k, v)
-		}
-	}
-	w.WriteHeader(resp.StatusCode)
-
-	// faster than io.Copy when streaming
-	buf := make([]byte, 32*1024)
-	for {
-		n, err := resp.Body.Read(buf)
-		if n > 0 {
-			if _, writeErr := w.Write(buf[:n]); writeErr != nil {
-				http.Error(w, writeErr.Error(), http.StatusInternalServerError)
-				return
-			}
-			if flusher, ok := w.(http.Flusher); ok {
-				flusher.Flush()
-			}
-		}
-		if err == io.EOF {
-			break
-		}
-		if err != nil {
-			http.Error(w, err.Error(), http.StatusBadGateway)
-			return
-		}
-	}
-}

proxy/metrics_middleware.go

@@ -0,0 +1,184 @@
+package proxy
+
+import (
+	"bytes"
+	"fmt"
+	"io"
+	"net/http"
+	"strings"
+	"time"
+
+	"github.com/gin-gonic/gin"
+	"github.com/tidwall/gjson"
+)
+
+type MetricsRecorder struct {
+	metricsMonitor *MetricsMonitor
+	realModelName  string
+	//	isStreaming    bool
+	startTime time.Time
+}
+
+// MetricsMiddleware sets up the MetricsResponseWriter for capturing upstream requests
+func MetricsMiddleware(pm *ProxyManager) gin.HandlerFunc {
+	return func(c *gin.Context) {
+		bodyBytes, err := io.ReadAll(c.Request.Body)
+		if err != nil {
+			pm.sendErrorResponse(c, http.StatusBadRequest, "could not ready request body")
+			c.Abort()
+			return
+		}
+		c.Request.Body = io.NopCloser(bytes.NewBuffer(bodyBytes))
+
+		requestedModel := gjson.GetBytes(bodyBytes, "model").String()
+		if requestedModel == "" {
+			pm.sendErrorResponse(c, http.StatusBadRequest, "missing or invalid 'model' key")
+			c.Abort()
+			return
+		}
+
+		realModelName, found := pm.config.RealModelName(requestedModel)
+		if !found {
+			pm.sendErrorResponse(c, http.StatusBadRequest, fmt.Sprintf("could not find real modelID for %s", requestedModel))
+			c.Abort()
+			return
+		}
+
+		writer := &MetricsResponseWriter{
+			ResponseWriter: c.Writer,
+			metricsRecorder: &MetricsRecorder{
+				metricsMonitor: pm.metricsMonitor,
+				realModelName:  realModelName,
+				startTime:      time.Now(),
+			},
+		}
+		c.Writer = writer
+		c.Next()
+
+		// check for streaming response
+		if strings.Contains(c.Writer.Header().Get("Content-Type"), "text/event-stream") {
+			writer.metricsRecorder.processStreamingResponse(writer.body)
+		} else {
+			writer.metricsRecorder.processNonStreamingResponse(writer.body)
+		}
+	}
+}
+
+func (rec *MetricsRecorder) parseAndRecordMetrics(jsonData gjson.Result) bool {
+	usage := jsonData.Get("usage")
+	timings := jsonData.Get("timings")
+	if !usage.Exists() && !timings.Exists() {
+		return false
+	}
+
+	// default values
+	cachedTokens := -1 // unknown or missing data
+	outputTokens := 0
+	inputTokens := 0
+
+	// timings data
+	tokensPerSecond := -1.0
+	promptPerSecond := -1.0
+	durationMs := int(time.Since(rec.startTime).Milliseconds())
+
+	if usage.Exists() {
+		outputTokens = int(jsonData.Get("usage.completion_tokens").Int())
+		inputTokens = int(jsonData.Get("usage.prompt_tokens").Int())
+	}
+
+	// use llama-server's timing data for tok/sec and duration as it is more accurate
+	if timings.Exists() {
+		inputTokens = int(jsonData.Get("timings.prompt_n").Int())
+		outputTokens = int(jsonData.Get("timings.predicted_n").Int())
+		promptPerSecond = jsonData.Get("timings.prompt_per_second").Float()
+		tokensPerSecond = jsonData.Get("timings.predicted_per_second").Float()
+		durationMs = int(jsonData.Get("timings.prompt_ms").Float() + jsonData.Get("timings.predicted_ms").Float())
+
+		if cachedValue := jsonData.Get("timings.cache_n"); cachedValue.Exists() {
+			cachedTokens = int(cachedValue.Int())
+		}
+	}
+
+	rec.metricsMonitor.addMetrics(TokenMetrics{
+		Timestamp:       time.Now(),
+		Model:           rec.realModelName,
+		CachedTokens:    cachedTokens,
+		InputTokens:     inputTokens,
+		OutputTokens:    outputTokens,
+		PromptPerSecond: promptPerSecond,
+		TokensPerSecond: tokensPerSecond,
+		DurationMs:      durationMs,
+	})
+
+	return true
+}
+
+func (rec *MetricsRecorder) processStreamingResponse(body []byte) {
+	// Iterate **backwards** through the lines looking for the data payload with
+	// usage data
+	lines := bytes.Split(body, []byte("\n"))
+
+	for i := len(lines) - 1; i >= 0; i-- {
+		line := bytes.TrimSpace(lines[i])
+		if len(line) == 0 {
+			continue
+		}
+
+		// SSE payload always follows "data:"
+		prefix := []byte("data:")
+		if !bytes.HasPrefix(line, prefix) {
+			continue
+		}
+		data := bytes.TrimSpace(line[len(prefix):])
+
+		if len(data) == 0 {
+			continue
+		}
+
+		if bytes.Equal(data, []byte("[DONE]")) {
+			// [DONE] line itself contains nothing of interest.
+			continue
+		}
+
+		if gjson.ValidBytes(data) {
+			if rec.parseAndRecordMetrics(gjson.ParseBytes(data)) {
+				return // short circuit if a metric was recorded
+			}
+		}
+	}
+}
+
+func (rec *MetricsRecorder) processNonStreamingResponse(body []byte) {
+	if len(body) == 0 {
+		return
+	}
+
+	// Parse JSON to extract usage information
+	if gjson.ValidBytes(body) {
+		rec.parseAndRecordMetrics(gjson.ParseBytes(body))
+	}
+}
+
+// MetricsResponseWriter captures the entire response for non-streaming
+type MetricsResponseWriter struct {
+	gin.ResponseWriter
+	body            []byte
+	metricsRecorder *MetricsRecorder
+}
+
+func (w *MetricsResponseWriter) Write(b []byte) (int, error) {
+	n, err := w.ResponseWriter.Write(b)
+	if err != nil {
+		return n, err
+	}
+	w.body = append(w.body, b...)
+	return n, nil
+}
+
+func (w *MetricsResponseWriter) WriteHeader(statusCode int) {
+	w.ResponseWriter.WriteHeader(statusCode)
+}
+
+func (w *MetricsResponseWriter) Header() http.Header {
+	return w.ResponseWriter.Header()
+}

proxy/metrics_monitor.go

@@ -0,0 +1,84 @@
+package proxy
+
+import (
+	"encoding/json"
+	"sync"
+	"time"
+
+	"github.com/mostlygeek/llama-swap/event"
+	"github.com/mostlygeek/llama-swap/proxy/config"
+)
+
+// TokenMetrics represents parsed token statistics from llama-server logs
+type TokenMetrics struct {
+	ID              int       `json:"id"`
+	Timestamp       time.Time `json:"timestamp"`
+	Model           string    `json:"model"`
+	CachedTokens    int       `json:"cache_tokens"`
+	InputTokens     int       `json:"input_tokens"`
+	OutputTokens    int       `json:"output_tokens"`
+	PromptPerSecond float64   `json:"prompt_per_second"`
+	TokensPerSecond float64   `json:"tokens_per_second"`
+	DurationMs      int       `json:"duration_ms"`
+}
+
+// TokenMetricsEvent represents a token metrics event
+type TokenMetricsEvent struct {
+	Metrics TokenMetrics
+}
+
+func (e TokenMetricsEvent) Type() uint32 {
+	return TokenMetricsEventID // defined in events.go
+}
+
+// MetricsMonitor parses llama-server output for token statistics
+type MetricsMonitor struct {
+	mu         sync.RWMutex
+	metrics    []TokenMetrics
+	maxMetrics int
+	nextID     int
+}
+
+func NewMetricsMonitor(config *config.Config) *MetricsMonitor {
+	maxMetrics := config.MetricsMaxInMemory
+	if maxMetrics <= 0 {
+		maxMetrics = 1000 // Default fallback
+	}
+
+	mp := &MetricsMonitor{
+		maxMetrics: maxMetrics,
+	}
+
+	return mp
+}
+
+// addMetrics adds a new metric to the collection and publishes an event
+func (mp *MetricsMonitor) addMetrics(metric TokenMetrics) {
+	mp.mu.Lock()
+	defer mp.mu.Unlock()
+
+	metric.ID = mp.nextID
+	mp.nextID++
+	mp.metrics = append(mp.metrics, metric)
+	if len(mp.metrics) > mp.maxMetrics {
+		mp.metrics = mp.metrics[len(mp.metrics)-mp.maxMetrics:]
+	}
+	event.Emit(TokenMetricsEvent{Metrics: metric})
+}
+
+// GetMetrics returns a copy of the current metrics
+func (mp *MetricsMonitor) GetMetrics() []TokenMetrics {
+	mp.mu.RLock()
+	defer mp.mu.RUnlock()
+
+	result := make([]TokenMetrics, len(mp.metrics))
+	copy(result, mp.metrics)
+	return result
+}
+
+// GetMetricsJSON returns metrics as JSON
+func (mp *MetricsMonitor) GetMetricsJSON() ([]byte, error) {
+	mp.mu.RLock()
+	defer mp.mu.RUnlock()
+	return json.Marshal(mp.metrics)
+}

proxy/process.go

@@ -0,0 +1,590 @@
+package proxy
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"net"
+	"net/http"
+	"net/http/httputil"
+	"net/url"
+	"os/exec"
+	"strings"
+	"sync"
+	"sync/atomic"
+	"syscall"
+	"time"
+
+	"github.com/mostlygeek/llama-swap/event"
+	"github.com/mostlygeek/llama-swap/proxy/config"
+)
+
+type ProcessState string
+
+const (
+	StateStopped  ProcessState = ProcessState("stopped")
+	StateStarting ProcessState = ProcessState("starting")
+	StateReady    ProcessState = ProcessState("ready")
+	StateStopping ProcessState = ProcessState("stopping")
+
+	// process is shutdown and will not be restarted
+	StateShutdown ProcessState = ProcessState("shutdown")
+)
+
+type StopStrategy int
+
+const (
+	StopImmediately StopStrategy = iota
+	StopWaitForInflightRequest
+)
+
+type Process struct {
+	ID           string
+	config       config.ModelConfig
+	cmd          *exec.Cmd
+	reverseProxy *httputil.ReverseProxy
+
+	// PR #155 called to cancel the upstream process
+	cmdMutex       sync.RWMutex
+	cancelUpstream context.CancelFunc
+
+	// closed when command exits
+	cmdWaitChan chan struct{}
+
+	processLogger *LogMonitor
+	proxyLogger   *LogMonitor
+
+	healthCheckTimeout      int
+	healthCheckLoopInterval time.Duration
+
+	lastRequestHandledMutex sync.RWMutex
+	lastRequestHandled      time.Time
+
+	stateMutex sync.RWMutex
+	state      ProcessState
+
+	inFlightRequests      sync.WaitGroup
+	inFlightRequestsCount atomic.Int32
+
+	// used to block on multiple start() calls
+	waitStarting sync.WaitGroup
+
+	// for managing concurrency limits
+	concurrencyLimitSemaphore chan struct{}
+
+	// used for testing to override the default value
+	gracefulStopTimeout time.Duration
+
+	// track the number of failed starts
+	failedStartCount int
+}
+
+func NewProcess(ID string, healthCheckTimeout int, config config.ModelConfig, processLogger *LogMonitor, proxyLogger *LogMonitor) *Process {
+	concurrentLimit := 10
+	if config.ConcurrencyLimit > 0 {
+		concurrentLimit = config.ConcurrencyLimit
+	}
+
+	// Setup the reverse proxy.
+	proxyURL, err := url.Parse(config.Proxy)
+	if err != nil {
+		proxyLogger.Errorf("<%s> invalid proxy URL %q: %v", ID, config.Proxy, err)
+	}
+
+	var reverseProxy *httputil.ReverseProxy
+	if proxyURL != nil {
+		reverseProxy = httputil.NewSingleHostReverseProxy(proxyURL)
+		reverseProxy.ModifyResponse = func(resp *http.Response) error {
+			// prevent nginx from buffering streaming responses (e.g., SSE)
+			if strings.Contains(strings.ToLower(resp.Header.Get("Content-Type")), "text/event-stream") {
+				resp.Header.Set("X-Accel-Buffering", "no")
+			}
+			return nil
+		}
+	}
+
+	return &Process{
+		ID:                      ID,
+		config:                  config,
+		cmd:                     nil,
+		reverseProxy:            reverseProxy,
+		cancelUpstream:          nil,
+		processLogger:           processLogger,
+		proxyLogger:             proxyLogger,
+		healthCheckTimeout:      healthCheckTimeout,
+		healthCheckLoopInterval: 5 * time.Second, /* default, can not be set by user - used for testing */
+		state:                   StateStopped,
+
+		// concurrency limit
+		concurrencyLimitSemaphore: make(chan struct{}, concurrentLimit),
+
+		// To be removed when migration over exec.CommandContext is complete
+		// stop timeout
+		gracefulStopTimeout: 10 * time.Second,
+		cmdWaitChan:         make(chan struct{}),
+	}
+}
+
+// LogMonitor returns the log monitor associated with the process.
+func (p *Process) LogMonitor() *LogMonitor {
+	return p.processLogger
+}
+
+// setLastRequestHandled sets the last request handled time in a thread-safe manner.
+func (p *Process) setLastRequestHandled(t time.Time) {
+	p.lastRequestHandledMutex.Lock()
+	defer p.lastRequestHandledMutex.Unlock()
+	p.lastRequestHandled = t
+}
+
+// getLastRequestHandled gets the last request handled time in a thread-safe manner.
+func (p *Process) getLastRequestHandled() time.Time {
+	p.lastRequestHandledMutex.RLock()
+	defer p.lastRequestHandledMutex.RUnlock()
+	return p.lastRequestHandled
+}
+
+// custom error types for swapping state
+var (
+	ErrExpectedStateMismatch  = errors.New("expected state mismatch")
+	ErrInvalidStateTransition = errors.New("invalid state transition")
+)
+
+// swapState performs a compare and swap of the state atomically. It returns the current state
+// and an error if the swap failed.
+func (p *Process) swapState(expectedState, newState ProcessState) (ProcessState, error) {
+	p.stateMutex.Lock()
+	defer p.stateMutex.Unlock()
+
+	if p.state != expectedState {
+		p.proxyLogger.Warnf("<%s> swapState() Unexpected current state %s, expected %s", p.ID, p.state, expectedState)
+		return p.state, ErrExpectedStateMismatch
+	}
+
+	if !isValidTransition(p.state, newState) {
+		p.proxyLogger.Warnf("<%s> swapState() Invalid state transition from %s to %s", p.ID, p.state, newState)
+		return p.state, ErrInvalidStateTransition
+	}
+
+	p.state = newState
+
+	// Atomically increment waitStarting when entering StateStarting
+	// This ensures any thread that sees StateStarting will also see the WaitGroup counter incremented
+	if newState == StateStarting {
+		p.waitStarting.Add(1)
+	}
+
+	p.proxyLogger.Debugf("<%s> swapState() State transitioned from %s to %s", p.ID, expectedState, newState)
+	event.Emit(ProcessStateChangeEvent{ProcessName: p.ID, NewState: newState, OldState: expectedState})
+	return p.state, nil
+}
+
+// Helper function to encapsulate transition rules
+func isValidTransition(from, to ProcessState) bool {
+	switch from {
+	case StateStopped:
+		return to == StateStarting
+	case StateStarting:
+		return to == StateReady || to == StateStopping || to == StateStopped
+	case StateReady:
+		return to == StateStopping
+	case StateStopping:
+		return to == StateStopped || to == StateShutdown
+	case StateShutdown:
+		return false // No transitions allowed from these states
+	}
+	return false
+}
+
+func (p *Process) CurrentState() ProcessState {
+	p.stateMutex.RLock()
+	defer p.stateMutex.RUnlock()
+	return p.state
+}
+
+// forceState forces the process state to the new state with mutex protection.
+// This should only be used in exceptional cases where the normal state transition
+// validation via swapState() cannot be used.
+func (p *Process) forceState(newState ProcessState) {
+	p.stateMutex.Lock()
+	defer p.stateMutex.Unlock()
+	p.state = newState
+}
+
+// start starts the upstream command, checks the health endpoint, and sets the state to Ready
+// it is a private method because starting is automatic but stopping can be called
+// at any time.
+func (p *Process) start() error {
+
+	if p.config.Proxy == "" {
+		return fmt.Errorf("can not start(), upstream proxy missing")
+	}
+
+	args, err := p.config.SanitizedCommand()
+	if err != nil {
+		return fmt.Errorf("unable to get sanitized command: %v", err)
+	}
+
+	if curState, err := p.swapState(StateStopped, StateStarting); err != nil {
+		if err == ErrExpectedStateMismatch {
+			// already starting, just wait for it to complete and expect
+			// it to be be in the Ready start after. If not, return an error
+			if curState == StateStarting {
+				p.waitStarting.Wait()
+				if state := p.CurrentState(); state == StateReady {
+					return nil
+				} else {
+					return fmt.Errorf("process was already starting but wound up in state %v", state)
+				}
+			} else {
+				return fmt.Errorf("processes was in state %v when start() was called", curState)
+			}
+		} else {
+			return fmt.Errorf("failed to set Process state to starting: current state: %v, error: %v", curState, err)
+		}
+	}
+
+	// waitStarting.Add(1) is now called atomically in swapState() when transitioning to StateStarting
+	defer p.waitStarting.Done()
+	cmdContext, ctxCancelUpstream := context.WithCancel(context.Background())
+
+	p.cmd = exec.CommandContext(cmdContext, args[0], args[1:]...)
+	p.cmd.Stdout = p.processLogger
+	p.cmd.Stderr = p.processLogger
+	p.cmd.Env = append(p.cmd.Environ(), p.config.Env...)
+	p.cmd.Cancel = p.cmdStopUpstreamProcess
+	p.cmd.WaitDelay = p.gracefulStopTimeout
+
+	p.cmdMutex.Lock()
+	p.cancelUpstream = ctxCancelUpstream
+	p.cmdWaitChan = make(chan struct{})
+	p.cmdMutex.Unlock()
+
+	p.failedStartCount++ // this will be reset to zero when the process has successfully started
+
+	p.proxyLogger.Debugf("<%s> Executing start command: %s, env: %s", p.ID, strings.Join(args, " "), strings.Join(p.config.Env, ", "))
+	err = p.cmd.Start()
+
+	// Set process state to failed
+	if err != nil {
+		if curState, swapErr := p.swapState(StateStarting, StateStopped); swapErr != nil {
+			p.forceState(StateStopped) // force it into a stopped state
+			return fmt.Errorf(
+				"failed to start command '%s' and state swap failed. command error: %v, current state: %v, state swap error: %v",
+				strings.Join(args, " "), err, curState, swapErr,
+			)
+		}
+		return fmt.Errorf("start() failed for command '%s': %v", strings.Join(args, " "), err)
+	}
+
+	// Capture the exit error for later signalling
+	go p.waitForCmd()
+
+	// One of three things can happen at this stage:
+	// 1. The command exits unexpectedly
+	// 2. The health check fails
+	// 3. The health check passes
+	//
+	// only in the third case will the process be considered Ready to accept
+	<-time.After(250 * time.Millisecond) // give process a bit of time to start
+
+	checkStartTime := time.Now()
+	maxDuration := time.Second * time.Duration(p.healthCheckTimeout)
+	checkEndpoint := strings.TrimSpace(p.config.CheckEndpoint)
+
+	// a "none" means don't check for health ... I could have picked a better word :facepalm:
+	if checkEndpoint != "none" {
+		proxyTo := p.config.Proxy
+		healthURL, err := url.JoinPath(proxyTo, checkEndpoint)
+		if err != nil {
+			return fmt.Errorf("failed to create health check URL proxy=%s and checkEndpoint=%s", proxyTo, checkEndpoint)
+		}
+
+		// Ready Check loop
+		for {
+			currentState := p.CurrentState()
+			if currentState != StateStarting {
+				if currentState == StateStopped {
+					return fmt.Errorf("upstream command exited prematurely but successfully")
+				}
+				return errors.New("health check interrupted due to shutdown")
+			}
+
+			if time.Since(checkStartTime) > maxDuration {
+				p.stopCommand()
+				return fmt.Errorf("health check timed out after %vs", maxDuration.Seconds())
+			}
+
+			if err := p.checkHealthEndpoint(healthURL); err == nil {
+				p.proxyLogger.Infof("<%s> Health check passed on %s", p.ID, healthURL)
+				break
+			} else {
+				if strings.Contains(err.Error(), "connection refused") {
+					ttl := time.Until(checkStartTime.Add(maxDuration))
+					p.proxyLogger.Debugf("<%s> Connection refused on %s, giving up in %.0fs (normal during startup)", p.ID, healthURL, ttl.Seconds())
+				} else {
+					p.proxyLogger.Debugf("<%s> Health check error on %s, %v (normal during startup)", p.ID, healthURL, err)
+				}
+			}
+			<-time.After(p.healthCheckLoopInterval)
+		}
+	}
+
+	if p.config.UnloadAfter > 0 {
+		// start a goroutine to check every second if
+		// the process should be stopped
+		go func() {
+			maxDuration := time.Duration(p.config.UnloadAfter) * time.Second
+
+			for range time.Tick(time.Second) {
+				if p.CurrentState() != StateReady {
+					return
+				}
+
+				// skip the TTL check if there are inflight requests
+				if p.inFlightRequestsCount.Load() != 0 {
+					continue
+				}
+
+				if time.Since(p.getLastRequestHandled()) > maxDuration {
+					p.proxyLogger.Infof("<%s> Unloading model, TTL of %ds reached", p.ID, p.config.UnloadAfter)
+					p.Stop()
+					return
+				}
+			}
+		}()
+	}
+
+	if curState, err := p.swapState(StateStarting, StateReady); err != nil {
+		return fmt.Errorf("failed to set Process state to ready: current state: %v, error: %v", curState, err)
+	} else {
+		p.failedStartCount = 0
+		return nil
+	}
+}
+
+// Stop will wait for inflight requests to complete before stopping the process.
+func (p *Process) Stop() {
+	if !isValidTransition(p.CurrentState(), StateStopping) {
+		return
+	}
+
+	// wait for any inflight requests before proceeding
+	p.proxyLogger.Debugf("<%s> Stop(): Waiting for inflight requests to complete", p.ID)
+	p.inFlightRequests.Wait()
+	p.StopImmediately()
+}
+
+// StopImmediately will transition the process to the stopping state and stop the process with a SIGTERM.
+// If the process does not stop within the specified timeout, it will be forcefully stopped with a SIGKILL.
+func (p *Process) StopImmediately() {
+	if !isValidTransition(p.CurrentState(), StateStopping) {
+		return
+	}
+
+	p.proxyLogger.Debugf("<%s> Stopping process, current state: %s", p.ID, p.CurrentState())
+	if curState, err := p.swapState(StateReady, StateStopping); err != nil {
+		p.proxyLogger.Infof("<%s> Stop() Ready -> StateStopping err: %v, current state: %v", p.ID, err, curState)
+		return
+	}
+
+	p.stopCommand()
+}
+
+// Shutdown is called when llama-swap is shutting down. It will give a little bit
+// of time for any inflight requests to complete before shutting down. If the Process
+// is in the state of starting, it will cancel it and shut it down. Once a process is in
+// the StateShutdown state, it can not be started again.
+func (p *Process) Shutdown() {
+	if !isValidTransition(p.CurrentState(), StateStopping) {
+		return
+	}
+
+	p.stopCommand()
+	// just force it to this state since there is no recovery from shutdown
+	p.forceState(StateShutdown)
+}
+
+// stopCommand will send a SIGTERM to the process and wait for it to exit.
+// If it does not exit within 5 seconds, it will send a SIGKILL.
+func (p *Process) stopCommand() {
+	stopStartTime := time.Now()
+	defer func() {
+		p.proxyLogger.Debugf("<%s> stopCommand took %v", p.ID, time.Since(stopStartTime))
+	}()
+
+	p.cmdMutex.RLock()
+	cancelUpstream := p.cancelUpstream
+	cmdWaitChan := p.cmdWaitChan
+	p.cmdMutex.RUnlock()
+
+	if cancelUpstream == nil {
+		p.proxyLogger.Errorf("<%s> stopCommand has a nil p.cancelUpstream()", p.ID)
+		return
+	}
+
+	cancelUpstream()
+	<-cmdWaitChan
+}
+
+func (p *Process) checkHealthEndpoint(healthURL string) error {
+
+	client := &http.Client{
+		// wait a short time for a tcp connection to be established
+		Transport: &http.Transport{
+			DialContext: (&net.Dialer{
+				Timeout: 500 * time.Millisecond,
+			}).DialContext,
+		},
+
+		// give a long time to respond to the health check endpoint
+		// after the connection is established. See issue: 276
+		Timeout: 5000 * time.Millisecond,
+	}
+
+	req, err := http.NewRequest("GET", healthURL, nil)
+	if err != nil {
+		return err
+	}
+
+	resp, err := client.Do(req)
+	if err != nil {
+		return err
+	}
+	defer resp.Body.Close()
+
+	// got a response but it was not an OK
+	if resp.StatusCode != http.StatusOK {
+		return fmt.Errorf("status code: %d", resp.StatusCode)
+	}
+
+	return nil
+}
+
+func (p *Process) ProxyRequest(w http.ResponseWriter, r *http.Request) {
+	requestBeginTime := time.Now()
+	var startDuration time.Duration
+
+	// prevent new requests from being made while stopping or irrecoverable
+	currentState := p.CurrentState()
+	if currentState == StateShutdown || currentState == StateStopping {
+		http.Error(w, fmt.Sprintf("Process can not ProxyRequest, state is %s", currentState), http.StatusServiceUnavailable)
+		return
+	}
+
+	select {
+	case p.concurrencyLimitSemaphore <- struct{}{}:
+		defer func() { <-p.concurrencyLimitSemaphore }()
+	default:
+		http.Error(w, "Too many requests", http.StatusTooManyRequests)
+		return
+	}
+
+	p.inFlightRequests.Add(1)
+	p.inFlightRequestsCount.Add(1)
+	defer func() {
+		p.setLastRequestHandled(time.Now())
+		p.inFlightRequestsCount.Add(-1)
+		p.inFlightRequests.Done()
+	}()
+
+	// start the process on demand
+	if p.CurrentState() != StateReady {
+		beginStartTime := time.Now()
+		if err := p.start(); err != nil {
+			errstr := fmt.Sprintf("unable to start process: %s", err)
+			http.Error(w, errstr, http.StatusBadGateway)
+			return
+		}
+		startDuration = time.Since(beginStartTime)
+	}
+
+	if p.reverseProxy != nil {
+		p.reverseProxy.ServeHTTP(w, r)
+	} else {
+		http.Error(w, fmt.Sprintf("No reverse proxy available for %s", p.ID), http.StatusInternalServerError)
+	}
+
+	totalTime := time.Since(requestBeginTime)
+	p.proxyLogger.Debugf("<%s> request %s - start: %v, total: %v",
+		p.ID, r.RequestURI, startDuration, totalTime)
+}
+
+// waitForCmd waits for the command to exit and handles exit conditions depending on current state
+func (p *Process) waitForCmd() {
+	exitErr := p.cmd.Wait()
+	p.proxyLogger.Debugf("<%s> cmd.Wait() returned error: %v", p.ID, exitErr)
+
+	if exitErr != nil {
+		if errno, ok := exitErr.(syscall.Errno); ok {
+			p.proxyLogger.Errorf("<%s> errno >> %v", p.ID, errno)
+		} else if exitError, ok := exitErr.(*exec.ExitError); ok {
+			if strings.Contains(exitError.String(), "signal: terminated") {
+				p.proxyLogger.Debugf("<%s> Process stopped OK", p.ID)
+			} else if strings.Contains(exitError.String(), "signal: interrupt") {
+				p.proxyLogger.Debugf("<%s> Process interrupted OK", p.ID)
+			} else {
+				p.proxyLogger.Warnf("<%s> ExitError >> %v, exit code: %d", p.ID, exitError, exitError.ExitCode())
+			}
+		} else {
+			if exitErr.Error() != "context canceled" /* this is normal */ {
+				p.proxyLogger.Errorf("<%s> Process exited >> %v", p.ID, exitErr)
+			}
+		}
+	}
+
+	currentState := p.CurrentState()
+	switch currentState {
+	case StateStopping:
+		if curState, err := p.swapState(StateStopping, StateStopped); err != nil {
+			p.proxyLogger.Errorf("<%s> Process exited but could not swap to StateStopped. curState=%s, err: %v", p.ID, curState, err)
+			p.forceState(StateStopped)
+		}
+	default:
+		p.proxyLogger.Infof("<%s> process exited but not StateStopping, current state: %s", p.ID, currentState)
+		p.forceState(StateStopped) // force it to be in this state
+	}
+
+	p.cmdMutex.Lock()
+	close(p.cmdWaitChan)
+	p.cmdMutex.Unlock()
+}
+
+// cmdStopUpstreamProcess attemps to stop the upstream process gracefully
+func (p *Process) cmdStopUpstreamProcess() error {
+	p.processLogger.Debugf("<%s> cmdStopUpstreamProcess() initiating graceful stop of upstream process", p.ID)
+
+	// this should never happen ...
+	if p.cmd == nil || p.cmd.Process == nil {
+		p.proxyLogger.Debugf("<%s> cmd or cmd.Process is nil (normal during config reload)", p.ID)
+		return fmt.Errorf("<%s> process is nil or cmd is nil, skipping graceful stop", p.ID)
+	}
+
+	if p.config.CmdStop != "" {
+		// replace ${PID} with the pid of the process
+		stopArgs, err := config.SanitizeCommand(strings.ReplaceAll(p.config.CmdStop, "${PID}", fmt.Sprintf("%d", p.cmd.Process.Pid)))
+		if err != nil {
+			p.proxyLogger.Errorf("<%s> Failed to sanitize stop command: %v", p.ID, err)
+			return err
+		}
+
+		p.proxyLogger.Debugf("<%s> Executing stop command: %s", p.ID, strings.Join(stopArgs, " "))
+
+		stopCmd := exec.Command(stopArgs[0], stopArgs[1:]...)
+		stopCmd.Stdout = p.processLogger
+		stopCmd.Stderr = p.processLogger
+		stopCmd.Env = p.cmd.Env
+
+		if err := stopCmd.Run(); err != nil {
+			p.proxyLogger.Errorf("<%s> Failed to exec stop command: %v", p.ID, err)
+			return err
+		}
+	} else {
+		if err := p.cmd.Process.Signal(syscall.SIGTERM); err != nil {
+			p.proxyLogger.Errorf("<%s> Failed to send SIGTERM to process: %v", p.ID, err)
+			return err
+		}
+	}
+
+	return nil
+}

proxy/process_test.go

@@ -0,0 +1,496 @@
+package proxy
+
+import (
+	"fmt"
+	"net/http"
+	"net/http/httptest"
+	"os"
+	"runtime"
+	"sync"
+	"testing"
+	"time"
+
+	"github.com/mostlygeek/llama-swap/proxy/config"
+	"github.com/stretchr/testify/assert"
+)
+
+var (
+	debugLogger = NewLogMonitorWriter(os.Stdout)
+)
+
+func init() {
+	// flip to help with debugging tests
+	if false {
+		debugLogger.SetLogLevel(LevelDebug)
+	} else {
+		debugLogger.SetLogLevel(LevelError)
+	}
+}
+
+func TestProcess_AutomaticallyStartsUpstream(t *testing.T) {
+
+	expectedMessage := "testing91931"
+	config := getTestSimpleResponderConfig(expectedMessage)
+
+	// Create a process
+	process := NewProcess("test-process", 5, config, debugLogger, debugLogger)
+	defer process.Stop()
+
+	req := httptest.NewRequest("GET", "/test", nil)
+	w := httptest.NewRecorder()
+
+	// process is automatically started
+	assert.Equal(t, StateStopped, process.CurrentState())
+	process.ProxyRequest(w, req)
+	assert.Equal(t, StateReady, process.CurrentState())
+
+	assert.Equal(t, http.StatusOK, w.Code, "Expected status code %d, got %d", http.StatusOK, w.Code)
+	assert.Contains(t, w.Body.String(), expectedMessage)
+
+	// Stop the process
+	process.Stop()
+
+	req = httptest.NewRequest("GET", "/", nil)
+	w = httptest.NewRecorder()
+
+	// Proxy the request
+	process.ProxyRequest(w, req)
+
+	// should have automatically started the process again
+	if w.Code != http.StatusOK {
+		t.Errorf("Expected status code %d, got %d", http.StatusOK, w.Code)
+	}
+}
+
+// TestProcess_WaitOnMultipleStarts tests that multiple concurrent requests
+// are all handled successfully, even though they all may ask for the process to .start()
+func TestProcess_WaitOnMultipleStarts(t *testing.T) {
+
+	expectedMessage := "testing91931"
+	config := getTestSimpleResponderConfig(expectedMessage)
+
+	process := NewProcess("test-process", 5, config, debugLogger, debugLogger)
+	defer process.Stop()
+
+	var wg sync.WaitGroup
+	for i := 0; i < 5; i++ {
+		wg.Add(1)
+		go func(reqID int) {
+			defer wg.Done()
+			req := httptest.NewRequest("GET", "/test", nil)
+			w := httptest.NewRecorder()
+			process.ProxyRequest(w, req)
+			assert.Equal(t, http.StatusOK, w.Code, "Worker %d got wrong HTTP code", reqID)
+			assert.Contains(t, w.Body.String(), expectedMessage, "Worker %d got wrong message", reqID)
+		}(i)
+	}
+	wg.Wait()
+	assert.Equal(t, StateReady, process.CurrentState())
+}
+
+// test that the automatic start returns the expected error type
+func TestProcess_BrokenModelConfig(t *testing.T) {
+	// Create a process configuration
+	config := config.ModelConfig{
+		Cmd:           "nonexistent-command",
+		Proxy:         "http://127.0.0.1:9913",
+		CheckEndpoint: "/health",
+	}
+
+	process := NewProcess("broken", 1, config, debugLogger, debugLogger)
+
+	req := httptest.NewRequest("GET", "/", nil)
+	w := httptest.NewRecorder()
+	process.ProxyRequest(w, req)
+	assert.Equal(t, http.StatusBadGateway, w.Code)
+	assert.Contains(t, w.Body.String(), "unable to start process")
+
+	w = httptest.NewRecorder()
+	process.ProxyRequest(w, req)
+	assert.Equal(t, http.StatusBadGateway, w.Code)
+	assert.Contains(t, w.Body.String(), "start() failed for command 'nonexistent-command':")
+}
+
+func TestProcess_UnloadAfterTTL(t *testing.T) {
+	if testing.Short() {
+		t.Skip("skipping long auto unload TTL test")
+	}
+
+	expectedMessage := "I_sense_imminent_danger"
+	config := getTestSimpleResponderConfig(expectedMessage)
+	assert.Equal(t, 0, config.UnloadAfter)
+	config.UnloadAfter = 3 // seconds
+	assert.Equal(t, 3, config.UnloadAfter)
+
+	process := NewProcess("ttl_test", 2, config, debugLogger, debugLogger)
+	defer process.Stop()
+
+	// this should take 4 seconds
+	req1 := httptest.NewRequest("GET", "/slow-respond?echo=1234&delay=1000ms", nil)
+	req2 := httptest.NewRequest("GET", "/test", nil)
+
+	w := httptest.NewRecorder()
+
+	// Proxy the request (auto start) with a slow response that takes longer than config.UnloadAfter
+	process.ProxyRequest(w, req1)
+
+	t.Log("sending slow first request (4 seconds)")
+	assert.Equal(t, http.StatusOK, w.Code, "Expected status code %d, got %d", http.StatusOK, w.Code)
+	assert.Contains(t, w.Body.String(), "1234")
+	assert.Equal(t, StateReady, process.CurrentState())
+
+	// ensure the TTL timeout does not race slow requests (see issue #25)
+	t.Log("sending second request (1 second)")
+	time.Sleep(time.Second)
+	w = httptest.NewRecorder()
+	process.ProxyRequest(w, req2)
+	assert.Equal(t, http.StatusOK, w.Code, "Expected status code %d, got %d", http.StatusOK, w.Code)
+	assert.Contains(t, w.Body.String(), expectedMessage)
+	assert.Equal(t, StateReady, process.CurrentState())
+
+	// wait 5 seconds
+	t.Log("sleep 5 seconds and check if unloaded")
+	time.Sleep(5 * time.Second)
+	assert.Equal(t, StateStopped, process.CurrentState())
+}
+
+func TestProcess_LowTTLValue(t *testing.T) {
+	if true { // change this code to run this ...
+		t.Skip("skipping test, edit process_test.go to run it ")
+	}
+
+	config := getTestSimpleResponderConfig("fast_ttl")
+	assert.Equal(t, 0, config.UnloadAfter)
+	config.UnloadAfter = 1 // second
+	assert.Equal(t, 1, config.UnloadAfter)
+
+	process := NewProcess("ttl", 2, config, debugLogger, debugLogger)
+	defer process.Stop()
+
+	for i := 0; i < 100; i++ {
+		t.Logf("Waiting before sending request %d", i)
+		time.Sleep(1500 * time.Millisecond)
+
+		expected := fmt.Sprintf("echo=test_%d", i)
+		req := httptest.NewRequest("GET", fmt.Sprintf("/slow-respond?echo=%s&delay=50ms", expected), nil)
+		w := httptest.NewRecorder()
+		process.ProxyRequest(w, req)
+		assert.Equal(t, http.StatusOK, w.Code)
+		assert.Contains(t, w.Body.String(), expected)
+	}
+
+}
+
+// issue #19
+// This test makes sure using Process.Stop() does not affect pending HTTP
+// requests. All HTTP requests in this test should complete successfully.
+func TestProcess_HTTPRequestsHaveTimeToFinish(t *testing.T) {
+	if testing.Short() {
+		t.Skip("skipping slow test")
+	}
+
+	expectedMessage := "12345"
+	config := getTestSimpleResponderConfig(expectedMessage)
+	process := NewProcess("t", 10, config, debugLogger, debugLogger)
+	defer process.Stop()
+
+	results := map[string]string{
+		"12345": "",
+		"abcde": "",
+		"fghij": "",
+	}
+
+	var wg sync.WaitGroup
+	var mu sync.Mutex
+
+	for key := range results {
+		wg.Add(1)
+		go func(key string) {
+			defer wg.Done()
+			// send a request where simple-responder is will wait 300ms before responding
+			// this will simulate an in-progress request.
+			req := httptest.NewRequest("GET", fmt.Sprintf("/slow-respond?echo=%s&delay=300ms", key), nil)
+			w := httptest.NewRecorder()
+
+			process.ProxyRequest(w, req)
+
+			if w.Code != http.StatusOK {
+				t.Errorf("Expected status OK, got %d for key %s", w.Code, key)
+			}
+
+			mu.Lock()
+			results[key] = w.Body.String()
+			mu.Unlock()
+
+		}(key)
+	}
+
+	// Stop the process while requests are still being processed
+	go func() {
+		<-time.After(150 * time.Millisecond)
+		process.Stop()
+	}()
+
+	wg.Wait()
+
+	for key, result := range results {
+		assert.Equal(t, key, result)
+	}
+}
+
+func TestProcess_SwapState(t *testing.T) {
+	tests := []struct {
+		name           string
+		currentState   ProcessState
+		expectedState  ProcessState
+		newState       ProcessState
+		expectedError  error
+		expectedResult ProcessState
+	}{
+		{"Stopped to Starting", StateStopped, StateStopped, StateStarting, nil, StateStarting},
+		{"Starting to Ready", StateStarting, StateStarting, StateReady, nil, StateReady},
+		{"Starting to Stopping", StateStarting, StateStarting, StateStopping, nil, StateStopping},
+		{"Starting to Stopped", StateStarting, StateStarting, StateStopped, nil, StateStopped},
+		{"Ready to Stopping", StateReady, StateReady, StateStopping, nil, StateStopping},
+		{"Stopping to Stopped", StateStopping, StateStopping, StateStopped, nil, StateStopped},
+		{"Stopping to Shutdown", StateStopping, StateStopping, StateShutdown, nil, StateShutdown},
+		{"Stopped to Ready", StateStopped, StateStopped, StateReady, ErrInvalidStateTransition, StateStopped},
+		{"Ready to Starting", StateReady, StateReady, StateStarting, ErrInvalidStateTransition, StateReady},
+		{"Stopping to Ready", StateStopping, StateStopping, StateReady, ErrInvalidStateTransition, StateStopping},
+		{"Shutdown to Stopped", StateShutdown, StateShutdown, StateStopped, ErrInvalidStateTransition, StateShutdown},
+		{"Shutdown to Starting", StateShutdown, StateShutdown, StateStarting, ErrInvalidStateTransition, StateShutdown},
+		{"Expected state mismatch", StateStopped, StateStarting, StateStarting, ErrExpectedStateMismatch, StateStopped},
+	}
+
+	for _, test := range tests {
+		t.Run(test.name, func(t *testing.T) {
+			p := NewProcess("test", 10, getTestSimpleResponderConfig("test"), debugLogger, debugLogger)
+			p.state = test.currentState
+
+			resultState, err := p.swapState(test.expectedState, test.newState)
+			if err != nil && test.expectedError == nil {
+				t.Errorf("Unexpected error: %v", err)
+			} else if err == nil && test.expectedError != nil {
+				t.Errorf("Expected error: %v, but got none", test.expectedError)
+			} else if err != nil && test.expectedError != nil {
+				if err.Error() != test.expectedError.Error() {
+					t.Errorf("Expected error: %v, got: %v", test.expectedError, err)
+				}
+			}
+
+			if resultState != test.expectedResult {
+				t.Errorf("Expected state: %v, got: %v", test.expectedResult, resultState)
+			}
+		})
+	}
+}
+
+func TestProcess_ShutdownInterruptsHealthCheck(t *testing.T) {
+	if testing.Short() {
+		t.Skip("skipping long shutdown test")
+	}
+
+	expectedMessage := "testing91931"
+
+	// make a config where the healthcheck will always fail because port is wrong
+	config := getTestSimpleResponderConfigPort(expectedMessage, 9999)
+	config.Proxy = "http://localhost:9998/test"
+
+	healthCheckTTLSeconds := 30
+	process := NewProcess("test-process", healthCheckTTLSeconds, config, debugLogger, debugLogger)
+
+	// make it a lot faster
+	process.healthCheckLoopInterval = time.Second
+
+	// start a goroutine to simulate a shutdown
+	var wg sync.WaitGroup
+	go func() {
+		defer wg.Done()
+		<-time.After(time.Millisecond * 500)
+		process.Shutdown()
+	}()
+	wg.Add(1)
+
+	// start the process, this is a blocking call
+	err := process.start()
+
+	wg.Wait()
+	assert.ErrorContains(t, err, "health check interrupted due to shutdown")
+	assert.Equal(t, StateShutdown, process.CurrentState())
+}
+
+func TestProcess_ExitInterruptsHealthCheck(t *testing.T) {
+	if testing.Short() {
+		t.Skip("skipping Exit Interrupts Health Check test")
+	}
+
+	// should run and exit but interrupt the long checkHealthTimeout
+	checkHealthTimeout := 5
+	config := config.ModelConfig{
+		Cmd:           "sleep 1",
+		Proxy:         "http://127.0.0.1:9913",
+		CheckEndpoint: "/health",
+	}
+
+	process := NewProcess("sleepy", checkHealthTimeout, config, debugLogger, debugLogger)
+	process.healthCheckLoopInterval = time.Second // make it faster
+	err := process.start()
+	assert.Equal(t, "upstream command exited prematurely but successfully", err.Error())
+	assert.Equal(t, process.CurrentState(), StateStopped)
+}
+
+func TestProcess_ConcurrencyLimit(t *testing.T) {
+	if testing.Short() {
+		t.Skip("skipping long concurrency limit test")
+	}
+
+	expectedMessage := "concurrency_limit_test"
+	config := getTestSimpleResponderConfig(expectedMessage)
+
+	// only allow 1 concurrent request at a time
+	config.ConcurrencyLimit = 1
+
+	process := NewProcess("ttl_test", 2, config, debugLogger, debugLogger)
+	assert.Equal(t, 1, cap(process.concurrencyLimitSemaphore))
+	defer process.Stop()
+
+	// launch a goroutine first to take up the semaphore
+	go func() {
+		req1 := httptest.NewRequest("GET", "/slow-respond?echo=12345&delay=75ms", nil)
+		w := httptest.NewRecorder()
+		process.ProxyRequest(w, req1)
+		assert.Equal(t, http.StatusOK, w.Code)
+	}()
+
+	// let the goroutine start
+	<-time.After(time.Millisecond * 25)
+
+	denied := httptest.NewRequest("GET", "/test", nil)
+	w := httptest.NewRecorder()
+	process.ProxyRequest(w, denied)
+	assert.Equal(t, http.StatusTooManyRequests, w.Code)
+}
+
+func TestProcess_StopImmediately(t *testing.T) {
+	expectedMessage := "test_stop_immediate"
+	config := getTestSimpleResponderConfig(expectedMessage)
+
+	process := NewProcess("stop_immediate", 2, config, debugLogger, debugLogger)
+	defer process.Stop()
+
+	err := process.start()
+	assert.Nil(t, err)
+	assert.Equal(t, process.CurrentState(), StateReady)
+	go func() {
+		// slow, but will get killed by StopImmediate
+		req := httptest.NewRequest("GET", "/slow-respond?echo=12345&delay=1s", nil)
+		w := httptest.NewRecorder()
+		process.ProxyRequest(w, req)
+	}()
+	<-time.After(time.Millisecond)
+	process.StopImmediately()
+	assert.Equal(t, process.CurrentState(), StateStopped)
+}
+
+// Test that SIGKILL is sent when gracefulStopTimeout is reached and properly terminates
+// the upstream command
+func TestProcess_ForceStopWithKill(t *testing.T) {
+	if runtime.GOOS == "windows" {
+		t.Skip("skipping SIGTERM test on Windows ")
+	}
+
+	expectedMessage := "test_sigkill"
+	binaryPath := getSimpleResponderPath()
+	port := getTestPort()
+
+	conf := config.ModelConfig{
+		// note --ignore-sig-term which ignores the SIGTERM signal so a SIGKILL must be sent
+		// to force the process to exit
+		Cmd:           fmt.Sprintf("%s --port %d --respond %s --silent --ignore-sig-term", binaryPath, port, expectedMessage),
+		Proxy:         fmt.Sprintf("http://127.0.0.1:%d", port),
+		CheckEndpoint: "/health",
+	}
+
+	process := NewProcess("stop_immediate", 2, conf, debugLogger, debugLogger)
+	defer process.Stop()
+
+	// reduce to make testing go faster
+	process.gracefulStopTimeout = time.Second
+
+	err := process.start()
+	assert.Nil(t, err)
+	assert.Equal(t, process.CurrentState(), StateReady)
+
+	waitChan := make(chan struct{})
+	go func() {
+		// slow, but will get killed by StopImmediate
+		req := httptest.NewRequest("GET", "/slow-respond?echo=12345&delay=2s", nil)
+		w := httptest.NewRecorder()
+		process.ProxyRequest(w, req)
+
+		// StatusOK because that was already sent before the kill
+		assert.Equal(t, http.StatusOK, w.Code)
+
+		// unexpected EOF because the kill happened, the "1" is sent before the kill
+		// then the unexpected EOF is sent after the kill
+		if runtime.GOOS == "windows" {
+			assert.Contains(t, w.Body.String(), "wsarecv: An existing connection was forcibly closed by the remote host")
+		} else {
+			// Upstream may be killed mid-response.
+			// Assert an incomplete or partial response.
+			assert.NotEqual(t, "12345", w.Body.String())
+		}
+
+		close(waitChan)
+	}()
+
+	<-time.After(time.Millisecond)
+	process.StopImmediately()
+	assert.Equal(t, process.CurrentState(), StateStopped)
+
+	// the request should have been interrupted by SIGKILL
+	<-waitChan
+}
+
+func TestProcess_StopCmd(t *testing.T) {
+	conf := getTestSimpleResponderConfig("test_stop_cmd")
+
+	if runtime.GOOS == "windows" {
+		conf.CmdStop = "taskkill /f /t /pid ${PID}"
+	} else {
+		conf.CmdStop = "kill -TERM ${PID}"
+	}
+
+	process := NewProcess("testStopCmd", 2, conf, debugLogger, debugLogger)
+	defer process.Stop()
+
+	err := process.start()
+	assert.Nil(t, err)
+	assert.Equal(t, process.CurrentState(), StateReady)
+	process.StopImmediately()
+	assert.Equal(t, process.CurrentState(), StateStopped)
+}
+
+func TestProcess_EnvironmentSetCorrectly(t *testing.T) {
+	expectedMessage := "test_env_not_emptied"
+	conf := getTestSimpleResponderConfig(expectedMessage)
+
+	// ensure that the the default config does not blank out the inherited environment
+	configWEnv := conf
+
+	// ensure the additiona variables are appended to the process' environment
+	configWEnv.Env = append(configWEnv.Env, "TEST_ENV1=1", "TEST_ENV2=2")
+
+	process1 := NewProcess("env_test", 2, conf, debugLogger, debugLogger)
+	process2 := NewProcess("env_test", 2, configWEnv, debugLogger, debugLogger)
+
+	process1.start()
+	defer process1.Stop()
+	process2.start()
+	defer process2.Stop()
+
+	assert.NotZero(t, len(process1.cmd.Environ()))
+	assert.NotZero(t, len(process2.cmd.Environ()))
+	assert.Equal(t, len(process1.cmd.Environ())+2, len(process2.cmd.Environ()), "process2 should have 2 more environment variables than process1")
+
+}

proxy/processgroup.go

@@ -0,0 +1,149 @@
+package proxy
+
+import (
+	"fmt"
+	"net/http"
+	"slices"
+	"sync"
+
+	"github.com/mostlygeek/llama-swap/proxy/config"
+)
+
+type ProcessGroup struct {
+	sync.Mutex
+
+	config     config.Config
+	id         string
+	swap       bool
+	exclusive  bool
+	persistent bool
+
+	proxyLogger    *LogMonitor
+	upstreamLogger *LogMonitor
+
+	// map of current processes
+	processes       map[string]*Process
+	lastUsedProcess string
+}
+
+func NewProcessGroup(id string, config config.Config, proxyLogger *LogMonitor, upstreamLogger *LogMonitor) *ProcessGroup {
+	groupConfig, ok := config.Groups[id]
+	if !ok {
+		panic("Unable to find configuration for group id: " + id)
+	}
+
+	pg := &ProcessGroup{
+		id:             id,
+		config:         config,
+		swap:           groupConfig.Swap,
+		exclusive:      groupConfig.Exclusive,
+		persistent:     groupConfig.Persistent,
+		proxyLogger:    proxyLogger,
+		upstreamLogger: upstreamLogger,
+		processes:      make(map[string]*Process),
+	}
+
+	// Create a Process for each member in the group
+	for _, modelID := range groupConfig.Members {
+		modelConfig, modelID, _ := pg.config.FindConfig(modelID)
+		process := NewProcess(modelID, pg.config.HealthCheckTimeout, modelConfig, pg.upstreamLogger, pg.proxyLogger)
+		pg.processes[modelID] = process
+	}
+
+	return pg
+}
+
+// ProxyRequest proxies a request to the specified model
+func (pg *ProcessGroup) ProxyRequest(modelID string, writer http.ResponseWriter, request *http.Request) error {
+	if !pg.HasMember(modelID) {
+		return fmt.Errorf("model %s not part of group %s", modelID, pg.id)
+	}
+
+	if pg.swap {
+		pg.Lock()
+		if pg.lastUsedProcess != modelID {
+
+			// is there something already running?
+			if pg.lastUsedProcess != "" {
+				pg.processes[pg.lastUsedProcess].Stop()
+			}
+
+			// wait for the request to the new model to be fully handled
+			// and prevent race conditions see issue #277
+			pg.processes[modelID].ProxyRequest(writer, request)
+			pg.lastUsedProcess = modelID
+
+			// short circuit and exit
+			pg.Unlock()
+			return nil
+		}
+		pg.Unlock()
+	}
+
+	pg.processes[modelID].ProxyRequest(writer, request)
+	return nil
+}
+
+func (pg *ProcessGroup) HasMember(modelName string) bool {
+	return slices.Contains(pg.config.Groups[pg.id].Members, modelName)
+}
+
+func (pg *ProcessGroup) StopProcess(modelID string, strategy StopStrategy) error {
+	pg.Lock()
+
+	process, exists := pg.processes[modelID]
+	if !exists {
+		pg.Unlock()
+		return fmt.Errorf("process not found for %s", modelID)
+	}
+
+	if pg.lastUsedProcess == modelID {
+		pg.lastUsedProcess = ""
+	}
+	pg.Unlock()
+
+	switch strategy {
+	case StopImmediately:
+		process.StopImmediately()
+	default:
+		process.Stop()
+	}
+	return nil
+}
+
+func (pg *ProcessGroup) StopProcesses(strategy StopStrategy) {
+	pg.Lock()
+	defer pg.Unlock()
+
+	if len(pg.processes) == 0 {
+		return
+	}
+
+	// stop Processes in parallel
+	var wg sync.WaitGroup
+	for _, process := range pg.processes {
+		wg.Add(1)
+		go func(process *Process) {
+			defer wg.Done()
+			switch strategy {
+			case StopImmediately:
+				process.StopImmediately()
+			default:
+				process.Stop()
+			}
+		}(process)
+	}
+	wg.Wait()
+}
+
+func (pg *ProcessGroup) Shutdown() {
+	var wg sync.WaitGroup
+	for _, process := range pg.processes {
+		wg.Add(1)
+		go func(process *Process) {
+			defer wg.Done()
+			process.Shutdown()
+		}(process)
+	}
+	wg.Wait()
+}

proxy/processgroup_test.go

@@ -0,0 +1,115 @@
+package proxy
+
+import (
+	"bytes"
+	"net/http"
+	"net/http/httptest"
+	"sync"
+	"testing"
+
+	"github.com/mostlygeek/llama-swap/proxy/config"
+	"github.com/stretchr/testify/assert"
+)
+
+var processGroupTestConfig = config.AddDefaultGroupToConfig(config.Config{
+	HealthCheckTimeout: 15,
+	Models: map[string]config.ModelConfig{
+		"model1": getTestSimpleResponderConfig("model1"),
+		"model2": getTestSimpleResponderConfig("model2"),
+		"model3": getTestSimpleResponderConfig("model3"),
+		"model4": getTestSimpleResponderConfig("model4"),
+		"model5": getTestSimpleResponderConfig("model5"),
+	},
+	Groups: map[string]config.GroupConfig{
+		"G1": {
+			Swap:      true,
+			Exclusive: true,
+			Members:   []string{"model1", "model2"},
+		},
+		"G2": {
+			Swap:      false,
+			Exclusive: true,
+			Members:   []string{"model3", "model4"},
+		},
+	},
+})
+
+func TestProcessGroup_DefaultHasCorrectModel(t *testing.T) {
+	pg := NewProcessGroup(config.DEFAULT_GROUP_ID, processGroupTestConfig, testLogger, testLogger)
+	assert.True(t, pg.HasMember("model5"))
+}
+
+func TestProcessGroup_HasMember(t *testing.T) {
+	pg := NewProcessGroup("G1", processGroupTestConfig, testLogger, testLogger)
+	assert.True(t, pg.HasMember("model1"))
+	assert.True(t, pg.HasMember("model2"))
+	assert.False(t, pg.HasMember("model3"))
+}
+
+// TestProcessGroup_ProxyRequestSwapIsTrueParallel tests that when swap is true
+// and multiple requests are made in parallel, only one process is running at a time.
+func TestProcessGroup_ProxyRequestSwapIsTrueParallel(t *testing.T) {
+	var processGroupTestConfig = config.AddDefaultGroupToConfig(config.Config{
+		HealthCheckTimeout: 15,
+		Models: map[string]config.ModelConfig{
+			// use the same listening so if a model is already running, it will fail
+			// this is a way to test that swap isolation is working
+			// properly when there are parallel requests made at the
+			// same time.
+			"model1": getTestSimpleResponderConfigPort("model1", 9832),
+			"model2": getTestSimpleResponderConfigPort("model2", 9832),
+			"model3": getTestSimpleResponderConfigPort("model3", 9832),
+			"model4": getTestSimpleResponderConfigPort("model4", 9832),
+			"model5": getTestSimpleResponderConfigPort("model5", 9832),
+		},
+		Groups: map[string]config.GroupConfig{
+			"G1": {
+				Swap:    true,
+				Members: []string{"model1", "model2", "model3", "model4", "model5"},
+			},
+		},
+	})
+
+	pg := NewProcessGroup("G1", processGroupTestConfig, testLogger, testLogger)
+	defer pg.StopProcesses(StopWaitForInflightRequest)
+
+	tests := []string{"model1", "model2", "model3", "model4", "model5"}
+
+	var wg sync.WaitGroup
+
+	wg.Add(len(tests))
+	for _, modelName := range tests {
+		go func(modelName string) {
+			defer wg.Done()
+			req := httptest.NewRequest("POST", "/v1/chat/completions", nil)
+			w := httptest.NewRecorder()
+			assert.NoError(t, pg.ProxyRequest(modelName, w, req))
+			assert.Equal(t, http.StatusOK, w.Code)
+			assert.Contains(t, w.Body.String(), modelName)
+		}(modelName)
+	}
+	wg.Wait()
+}
+
+func TestProcessGroup_ProxyRequestSwapIsFalse(t *testing.T) {
+	pg := NewProcessGroup("G2", processGroupTestConfig, testLogger, testLogger)
+	defer pg.StopProcesses(StopWaitForInflightRequest)
+
+	tests := []string{"model3", "model4"}
+
+	for _, modelName := range tests {
+		t.Run(modelName, func(t *testing.T) {
+			reqBody := `{"x", "y"}`
+			req := httptest.NewRequest("POST", "/v1/chat/completions", bytes.NewBufferString(reqBody))
+			w := httptest.NewRecorder()
+			assert.NoError(t, pg.ProxyRequest(modelName, w, req))
+			assert.Equal(t, http.StatusOK, w.Code)
+			assert.Contains(t, w.Body.String(), modelName)
+		})
+	}
+
+	// make sure all the processes are running
+	for _, process := range pg.processes {
+		assert.Equal(t, StateReady, process.CurrentState())
+	}
+}

proxy/proxymanager.go

@@ -0,0 +1,687 @@
+package proxy
+
+import (
+	"bytes"
+	"context"
+	"fmt"
+	"io"
+	"mime/multipart"
+	"net/http"
+	"os"
+	"sort"
+	"strconv"
+	"strings"
+	"sync"
+	"time"
+
+	"github.com/gin-gonic/gin"
+	"github.com/mostlygeek/llama-swap/event"
+	"github.com/mostlygeek/llama-swap/proxy/config"
+	"github.com/tidwall/gjson"
+	"github.com/tidwall/sjson"
+)
+
+const (
+	PROFILE_SPLIT_CHAR = ":"
+)
+
+type ProxyManager struct {
+	sync.Mutex
+
+	config    config.Config
+	ginEngine *gin.Engine
+
+	// logging
+	proxyLogger    *LogMonitor
+	upstreamLogger *LogMonitor
+	muxLogger      *LogMonitor
+
+	metricsMonitor *MetricsMonitor
+
+	processGroups map[string]*ProcessGroup
+
+	// shutdown signaling
+	shutdownCtx    context.Context
+	shutdownCancel context.CancelFunc
+}
+
+func New(config config.Config) *ProxyManager {
+	// set up loggers
+	stdoutLogger := NewLogMonitorWriter(os.Stdout)
+	upstreamLogger := NewLogMonitorWriter(stdoutLogger)
+	proxyLogger := NewLogMonitorWriter(stdoutLogger)
+
+	if config.LogRequests {
+		proxyLogger.Warn("LogRequests configuration is deprecated. Use logLevel instead.")
+	}
+
+	switch strings.ToLower(strings.TrimSpace(config.LogLevel)) {
+	case "debug":
+		proxyLogger.SetLogLevel(LevelDebug)
+		upstreamLogger.SetLogLevel(LevelDebug)
+	case "info":
+		proxyLogger.SetLogLevel(LevelInfo)
+		upstreamLogger.SetLogLevel(LevelInfo)
+	case "warn":
+		proxyLogger.SetLogLevel(LevelWarn)
+		upstreamLogger.SetLogLevel(LevelWarn)
+	case "error":
+		proxyLogger.SetLogLevel(LevelError)
+		upstreamLogger.SetLogLevel(LevelError)
+	default:
+		proxyLogger.SetLogLevel(LevelInfo)
+		upstreamLogger.SetLogLevel(LevelInfo)
+	}
+
+	shutdownCtx, shutdownCancel := context.WithCancel(context.Background())
+
+	pm := &ProxyManager{
+		config:    config,
+		ginEngine: gin.New(),
+
+		proxyLogger:    proxyLogger,
+		muxLogger:      stdoutLogger,
+		upstreamLogger: upstreamLogger,
+
+		metricsMonitor: NewMetricsMonitor(&config),
+
+		processGroups: make(map[string]*ProcessGroup),
+
+		shutdownCtx:    shutdownCtx,
+		shutdownCancel: shutdownCancel,
+	}
+
+	// create the process groups
+	for groupID := range config.Groups {
+		processGroup := NewProcessGroup(groupID, config, proxyLogger, upstreamLogger)
+		pm.processGroups[groupID] = processGroup
+	}
+
+	pm.setupGinEngine()
+
+	// run any startup hooks
+	if len(config.Hooks.OnStartup.Preload) > 0 {
+		// do it in the background, don't block startup -- not sure if good idea yet
+		go func() {
+			discardWriter := &DiscardWriter{}
+			for _, realModelName := range config.Hooks.OnStartup.Preload {
+				proxyLogger.Infof("Preloading model: %s", realModelName)
+				processGroup, _, err := pm.swapProcessGroup(realModelName)
+
+				if err != nil {
+					event.Emit(ModelPreloadedEvent{
+						ModelName: realModelName,
+						Success:   false,
+					})
+					proxyLogger.Errorf("Failed to preload model %s: %v", realModelName, err)
+					continue
+				} else {
+					req, _ := http.NewRequest("GET", "/", nil)
+					processGroup.ProxyRequest(realModelName, discardWriter, req)
+					event.Emit(ModelPreloadedEvent{
+						ModelName: realModelName,
+						Success:   true,
+					})
+				}
+			}
+		}()
+	}
+
+	return pm
+}
+
+func (pm *ProxyManager) setupGinEngine() {
+	pm.ginEngine.Use(func(c *gin.Context) {
+		// Start timer
+		start := time.Now()
+
+		// capture these because /upstream/:model rewrites them in c.Next()
+		clientIP := c.ClientIP()
+		method := c.Request.Method
+		path := c.Request.URL.Path
+
+		// Process request
+		c.Next()
+
+		// Stop timer
+		duration := time.Since(start)
+
+		statusCode := c.Writer.Status()
+		bodySize := c.Writer.Size()
+
+		pm.proxyLogger.Infof("Request %s \"%s %s %s\" %d %d \"%s\" %v",
+			clientIP,
+			method,
+			path,
+			c.Request.Proto,
+			statusCode,
+			bodySize,
+			c.Request.UserAgent(),
+			duration,
+		)
+	})
+
+	// see: issue: #81, #77 and #42 for CORS issues
+	// respond with permissive OPTIONS for any endpoint
+	pm.ginEngine.Use(func(c *gin.Context) {
+		if c.Request.Method == "OPTIONS" {
+			c.Header("Access-Control-Allow-Origin", "*")
+			c.Header("Access-Control-Allow-Methods", "GET, POST, PUT, PATCH, DELETE, OPTIONS")
+
+			// allow whatever the client requested by default
+			if headers := c.Request.Header.Get("Access-Control-Request-Headers"); headers != "" {
+				sanitized := SanitizeAccessControlRequestHeaderValues(headers)
+				c.Header("Access-Control-Allow-Headers", sanitized)
+			} else {
+				c.Header(
+					"Access-Control-Allow-Headers",
+					"Content-Type, Authorization, Accept, X-Requested-With",
+				)
+			}
+			c.Header("Access-Control-Max-Age", "86400")
+			c.AbortWithStatus(http.StatusNoContent)
+			return
+		}
+		c.Next()
+	})
+
+	mm := MetricsMiddleware(pm)
+
+	// Set up routes using the Gin engine
+	pm.ginEngine.POST("/v1/chat/completions", mm, pm.proxyOAIHandler)
+	// Support legacy /v1/completions api, see issue #12
+	pm.ginEngine.POST("/v1/completions", mm, pm.proxyOAIHandler)
+
+	// Support embeddings and reranking
+	pm.ginEngine.POST("/v1/embeddings", mm, pm.proxyOAIHandler)
+
+	// llama-server's /reranking endpoint + aliases
+	pm.ginEngine.POST("/reranking", mm, pm.proxyOAIHandler)
+	pm.ginEngine.POST("/rerank", mm, pm.proxyOAIHandler)
+	pm.ginEngine.POST("/v1/rerank", mm, pm.proxyOAIHandler)
+	pm.ginEngine.POST("/v1/reranking", mm, pm.proxyOAIHandler)
+
+	// llama-server's /infill endpoint for code infilling
+	pm.ginEngine.POST("/infill", mm, pm.proxyOAIHandler)
+
+	// llama-server's /completion endpoint
+	pm.ginEngine.POST("/completion", mm, pm.proxyOAIHandler)
+
+	// Support audio/speech endpoint
+	pm.ginEngine.POST("/v1/audio/speech", pm.proxyOAIHandler)
+	pm.ginEngine.POST("/v1/audio/transcriptions", pm.proxyOAIPostFormHandler)
+
+	pm.ginEngine.GET("/v1/models", pm.listModelsHandler)
+
+	// in proxymanager_loghandlers.go
+	pm.ginEngine.GET("/logs", pm.sendLogsHandlers)
+	pm.ginEngine.GET("/logs/stream", pm.streamLogsHandler)
+	pm.ginEngine.GET("/logs/stream/:logMonitorID", pm.streamLogsHandler)
+
+	/**
+	 * User Interface Endpoints
+	 */
+	pm.ginEngine.GET("/", func(c *gin.Context) {
+		c.Redirect(http.StatusFound, "/ui")
+	})
+
+	pm.ginEngine.GET("/upstream", func(c *gin.Context) {
+		c.Redirect(http.StatusFound, "/ui/models")
+	})
+	pm.ginEngine.Any("/upstream/*upstreamPath", pm.proxyToUpstream)
+	pm.ginEngine.GET("/unload", pm.unloadAllModelsHandler)
+	pm.ginEngine.GET("/running", pm.listRunningProcessesHandler)
+	pm.ginEngine.GET("/health", func(c *gin.Context) {
+		c.String(http.StatusOK, "OK")
+	})
+
+	pm.ginEngine.GET("/favicon.ico", func(c *gin.Context) {
+		if data, err := reactStaticFS.ReadFile("ui_dist/favicon.ico"); err == nil {
+			c.Data(http.StatusOK, "image/x-icon", data)
+		} else {
+			c.String(http.StatusInternalServerError, err.Error())
+		}
+	})
+
+	reactFS, err := GetReactFS()
+	if err != nil {
+		pm.proxyLogger.Errorf("Failed to load React filesystem: %v", err)
+	} else {
+
+		// serve files that exist under /ui/*
+		pm.ginEngine.StaticFS("/ui", reactFS)
+
+		// server SPA for UI under /ui/*
+		pm.ginEngine.NoRoute(func(c *gin.Context) {
+			if !strings.HasPrefix(c.Request.URL.Path, "/ui") {
+				c.AbortWithStatus(http.StatusNotFound)
+				return
+			}
+
+			file, err := reactFS.Open("index.html")
+			if err != nil {
+				c.String(http.StatusInternalServerError, err.Error())
+				return
+			}
+			defer file.Close()
+			http.ServeContent(c.Writer, c.Request, "index.html", time.Now(), file)
+
+		})
+	}
+
+	// see: proxymanager_api.go
+	// add API handler functions
+	addApiHandlers(pm)
+
+	// Disable console color for testing
+	gin.DisableConsoleColor()
+}
+
+// ServeHTTP implements http.Handler interface
+func (pm *ProxyManager) ServeHTTP(w http.ResponseWriter, r *http.Request) {
+	pm.ginEngine.ServeHTTP(w, r)
+}
+
+// StopProcesses acquires a lock and stops all running upstream processes.
+// This is the public method safe for concurrent calls.
+// Unlike Shutdown, this method only stops the processes but doesn't perform
+// a complete shutdown, allowing for process replacement without full termination.
+func (pm *ProxyManager) StopProcesses(strategy StopStrategy) {
+	pm.Lock()
+	defer pm.Unlock()
+
+	// stop Processes in parallel
+	var wg sync.WaitGroup
+	for _, processGroup := range pm.processGroups {
+		wg.Add(1)
+		go func(processGroup *ProcessGroup) {
+			defer wg.Done()
+			processGroup.StopProcesses(strategy)
+		}(processGroup)
+	}
+
+	wg.Wait()
+}
+
+// Shutdown stops all processes managed by this ProxyManager
+func (pm *ProxyManager) Shutdown() {
+	pm.Lock()
+	defer pm.Unlock()
+
+	pm.proxyLogger.Debug("Shutdown() called in proxy manager")
+
+	var wg sync.WaitGroup
+	// Send shutdown signal to all process in groups
+	for _, processGroup := range pm.processGroups {
+		wg.Add(1)
+		go func(processGroup *ProcessGroup) {
+			defer wg.Done()
+			processGroup.Shutdown()
+		}(processGroup)
+	}
+	wg.Wait()
+	pm.shutdownCancel()
+}
+
+func (pm *ProxyManager) swapProcessGroup(requestedModel string) (*ProcessGroup, string, error) {
+	// de-alias the real model name and get a real one
+	realModelName, found := pm.config.RealModelName(requestedModel)
+	if !found {
+		return nil, realModelName, fmt.Errorf("could not find real modelID for %s", requestedModel)
+	}
+
+	processGroup := pm.findGroupByModelName(realModelName)
+	if processGroup == nil {
+		return nil, realModelName, fmt.Errorf("could not find process group for model %s", requestedModel)
+	}
+
+	if processGroup.exclusive {
+		pm.proxyLogger.Debugf("Exclusive mode for group %s, stopping other process groups", processGroup.id)
+		for groupId, otherGroup := range pm.processGroups {
+			if groupId != processGroup.id && !otherGroup.persistent {
+				otherGroup.StopProcesses(StopWaitForInflightRequest)
+			}
+		}
+	}
+
+	return processGroup, realModelName, nil
+}
+
+func (pm *ProxyManager) listModelsHandler(c *gin.Context) {
+	data := make([]gin.H, 0, len(pm.config.Models))
+	createdTime := time.Now().Unix()
+
+	for id, modelConfig := range pm.config.Models {
+		if modelConfig.Unlisted {
+			continue
+		}
+
+		record := gin.H{
+			"id":       id,
+			"object":   "model",
+			"created":  createdTime,
+			"owned_by": "llama-swap",
+		}
+
+		if name := strings.TrimSpace(modelConfig.Name); name != "" {
+			record["name"] = name
+		}
+		if desc := strings.TrimSpace(modelConfig.Description); desc != "" {
+			record["description"] = desc
+		}
+
+		// Add metadata if present
+		if len(modelConfig.Metadata) > 0 {
+			record["meta"] = gin.H{
+				"llamaswap": modelConfig.Metadata,
+			}
+		}
+
+		data = append(data, record)
+	}
+
+	// Sort by the "id" key
+	sort.Slice(data, func(i, j int) bool {
+		si, _ := data[i]["id"].(string)
+		sj, _ := data[j]["id"].(string)
+		return si < sj
+	})
+
+	// Set CORS headers if origin exists
+	if origin := c.GetHeader("Origin"); origin != "" {
+		c.Header("Access-Control-Allow-Origin", origin)
+	}
+
+	// Use gin's JSON method which handles content-type and encoding
+	c.JSON(http.StatusOK, gin.H{
+		"object": "list",
+		"data":   data,
+	})
+}
+
+func (pm *ProxyManager) proxyToUpstream(c *gin.Context) {
+	upstreamPath := c.Param("upstreamPath")
+
+	// split the upstream path by / and search for the model name
+	parts := strings.Split(strings.TrimSpace(upstreamPath), "/")
+	if len(parts) == 0 {
+		pm.sendErrorResponse(c, http.StatusBadRequest, "model id required in path")
+		return
+	}
+
+	modelFound := false
+	searchModelName := ""
+	var modelName, remainingPath string
+	for i, part := range parts {
+		if parts[i] == "" {
+			continue
+		}
+
+		if searchModelName == "" {
+			searchModelName = part
+		} else {
+			searchModelName = searchModelName + "/" + parts[i]
+		}
+
+		if real, ok := pm.config.RealModelName(searchModelName); ok {
+			modelName = real
+			remainingPath = "/" + strings.Join(parts[i+1:], "/")
+			modelFound = true
+
+			// Check if this is exactly a model name with no additional path
+			// and doesn't end with a trailing slash
+			if remainingPath == "/" && !strings.HasSuffix(upstreamPath, "/") {
+				// Build new URL with query parameters preserved
+				newPath := "/upstream/" + searchModelName + "/"
+				if c.Request.URL.RawQuery != "" {
+					newPath += "?" + c.Request.URL.RawQuery
+				}
+
+				// Use 308 for non-GET/HEAD requests to preserve method
+				if c.Request.Method == http.MethodGet || c.Request.Method == http.MethodHead {
+					c.Redirect(http.StatusMovedPermanently, newPath)
+				} else {
+					c.Redirect(http.StatusPermanentRedirect, newPath)
+				}
+				return
+			}
+			break
+		}
+	}
+
+	if !modelFound {
+		pm.sendErrorResponse(c, http.StatusBadRequest, "model id required in path")
+		return
+	}
+
+	processGroup, realModelName, err := pm.swapProcessGroup(modelName)
+	if err != nil {
+		pm.sendErrorResponse(c, http.StatusInternalServerError, fmt.Sprintf("error swapping process group: %s", err.Error()))
+		return
+	}
+
+	// rewrite the path
+	c.Request.URL.Path = remainingPath
+	processGroup.ProxyRequest(realModelName, c.Writer, c.Request)
+}
+
+func (pm *ProxyManager) proxyOAIHandler(c *gin.Context) {
+	bodyBytes, err := io.ReadAll(c.Request.Body)
+	if err != nil {
+		pm.sendErrorResponse(c, http.StatusBadRequest, "could not ready request body")
+		return
+	}
+
+	requestedModel := gjson.GetBytes(bodyBytes, "model").String()
+	if requestedModel == "" {
+		pm.sendErrorResponse(c, http.StatusBadRequest, "missing or invalid 'model' key")
+		return
+	}
+
+	realModelName, found := pm.config.RealModelName(requestedModel)
+	if !found {
+		pm.sendErrorResponse(c, http.StatusBadRequest, fmt.Sprintf("could not find real modelID for %s", requestedModel))
+		return
+	}
+
+	processGroup, _, err := pm.swapProcessGroup(realModelName)
+	if err != nil {
+		pm.sendErrorResponse(c, http.StatusInternalServerError, fmt.Sprintf("error swapping process group: %s", err.Error()))
+		return
+	}
+
+	// issue #69 allow custom model names to be sent to upstream
+	useModelName := pm.config.Models[realModelName].UseModelName
+	if useModelName != "" {
+		bodyBytes, err = sjson.SetBytes(bodyBytes, "model", useModelName)
+		if err != nil {
+			pm.sendErrorResponse(c, http.StatusInternalServerError, fmt.Sprintf("error rewriting model name in JSON: %s", err.Error()))
+			return
+		}
+	}
+
+	// issue #174 strip parameters from the JSON body
+	stripParams, err := pm.config.Models[realModelName].Filters.SanitizedStripParams()
+	if err != nil { // just log it and continue
+		pm.proxyLogger.Errorf("Error sanitizing strip params string: %s, %s", pm.config.Models[realModelName].Filters.StripParams, err.Error())
+	} else {
+		for _, param := range stripParams {
+			pm.proxyLogger.Debugf("<%s> stripping param: %s", realModelName, param)
+			bodyBytes, err = sjson.DeleteBytes(bodyBytes, param)
+			if err != nil {
+				pm.sendErrorResponse(c, http.StatusInternalServerError, fmt.Sprintf("error deleting parameter %s from request", param))
+				return
+			}
+		}
+	}
+
+	c.Request.Body = io.NopCloser(bytes.NewBuffer(bodyBytes))
+
+	// dechunk it as we already have all the body bytes see issue #11
+	c.Request.Header.Del("transfer-encoding")
+	c.Request.Header.Set("content-length", strconv.Itoa(len(bodyBytes)))
+	c.Request.ContentLength = int64(len(bodyBytes))
+
+	if err := processGroup.ProxyRequest(realModelName, c.Writer, c.Request); err != nil {
+		pm.sendErrorResponse(c, http.StatusInternalServerError, fmt.Sprintf("error proxying request: %s", err.Error()))
+		pm.proxyLogger.Errorf("Error Proxying Request for processGroup %s and model %s", processGroup.id, realModelName)
+		return
+	}
+}
+
+func (pm *ProxyManager) proxyOAIPostFormHandler(c *gin.Context) {
+	// Parse multipart form
+	if err := c.Request.ParseMultipartForm(32 << 20); err != nil { // 32MB max memory, larger files go to tmp disk
+		pm.sendErrorResponse(c, http.StatusBadRequest, fmt.Sprintf("error parsing multipart form: %s", err.Error()))
+		return
+	}
+
+	// Get model parameter from the form
+	requestedModel := c.Request.FormValue("model")
+	if requestedModel == "" {
+		pm.sendErrorResponse(c, http.StatusBadRequest, "missing or invalid 'model' parameter in form data")
+		return
+	}
+
+	processGroup, realModelName, err := pm.swapProcessGroup(requestedModel)
+	if err != nil {
+		pm.sendErrorResponse(c, http.StatusInternalServerError, fmt.Sprintf("error swapping process group: %s", err.Error()))
+		return
+	}
+
+	// We need to reconstruct the multipart form in any case since the body is consumed
+	// Create a new buffer for the reconstructed request
+	var requestBuffer bytes.Buffer
+	multipartWriter := multipart.NewWriter(&requestBuffer)
+
+	// Copy all form values
+	for key, values := range c.Request.MultipartForm.Value {
+		for _, value := range values {
+			fieldValue := value
+			// If this is the model field and we have a profile, use just the model name
+			if key == "model" {
+				// # issue #69 allow custom model names to be sent to upstream
+				useModelName := pm.config.Models[realModelName].UseModelName
+
+				if useModelName != "" {
+					fieldValue = useModelName
+				} else {
+					fieldValue = requestedModel
+				}
+			}
+			field, err := multipartWriter.CreateFormField(key)
+			if err != nil {
+				pm.sendErrorResponse(c, http.StatusInternalServerError, "error recreating form field")
+				return
+			}
+			if _, err = field.Write([]byte(fieldValue)); err != nil {
+				pm.sendErrorResponse(c, http.StatusInternalServerError, "error writing form field")
+				return
+			}
+		}
+	}
+
+	// Copy all files from the original request
+	for key, fileHeaders := range c.Request.MultipartForm.File {
+		for _, fileHeader := range fileHeaders {
+			formFile, err := multipartWriter.CreateFormFile(key, fileHeader.Filename)
+			if err != nil {
+				pm.sendErrorResponse(c, http.StatusInternalServerError, "error recreating form file")
+				return
+			}
+
+			file, err := fileHeader.Open()
+			if err != nil {
+				pm.sendErrorResponse(c, http.StatusInternalServerError, "error opening uploaded file")
+				return
+			}
+
+			if _, err = io.Copy(formFile, file); err != nil {
+				file.Close()
+				pm.sendErrorResponse(c, http.StatusInternalServerError, "error copying file data")
+				return
+			}
+			file.Close()
+		}
+	}
+
+	// Close the multipart writer to finalize the form
+	if err := multipartWriter.Close(); err != nil {
+		pm.sendErrorResponse(c, http.StatusInternalServerError, "error finalizing multipart form")
+		return
+	}
+
+	// Create a new request with the reconstructed form data
+	modifiedReq, err := http.NewRequestWithContext(
+		c.Request.Context(),
+		c.Request.Method,
+		c.Request.URL.String(),
+		&requestBuffer,
+	)
+	if err != nil {
+		pm.sendErrorResponse(c, http.StatusInternalServerError, "error creating modified request")
+		return
+	}
+
+	// Copy the headers from the original request
+	modifiedReq.Header = c.Request.Header.Clone()
+	modifiedReq.Header.Set("Content-Type", multipartWriter.FormDataContentType())
+
+	// set the content length of the body
+	modifiedReq.Header.Set("Content-Length", strconv.Itoa(requestBuffer.Len()))
+	modifiedReq.ContentLength = int64(requestBuffer.Len())
+
+	// Use the modified request for proxying
+	if err := processGroup.ProxyRequest(realModelName, c.Writer, modifiedReq); err != nil {
+		pm.sendErrorResponse(c, http.StatusInternalServerError, fmt.Sprintf("error proxying request: %s", err.Error()))
+		pm.proxyLogger.Errorf("Error Proxying Request for processGroup %s and model %s", processGroup.id, realModelName)
+		return
+	}
+}
+
+func (pm *ProxyManager) sendErrorResponse(c *gin.Context, statusCode int, message string) {
+	acceptHeader := c.GetHeader("Accept")
+
+	if strings.Contains(acceptHeader, "application/json") {
+		c.JSON(statusCode, gin.H{"error": message})
+	} else {
+		c.String(statusCode, message)
+	}
+}
+
+func (pm *ProxyManager) unloadAllModelsHandler(c *gin.Context) {
+	pm.StopProcesses(StopImmediately)
+	c.String(http.StatusOK, "OK")
+}
+
+func (pm *ProxyManager) listRunningProcessesHandler(context *gin.Context) {
+	context.Header("Content-Type", "application/json")
+	runningProcesses := make([]gin.H, 0) // Default to an empty response.
+
+	for _, processGroup := range pm.processGroups {
+		for _, process := range processGroup.processes {
+			if process.CurrentState() == StateReady {
+				runningProcesses = append(runningProcesses, gin.H{
+					"model": process.ID,
+					"state": process.state,
+				})
+			}
+		}
+	}
+
+	// Put the results under the `running` key.
+	response := gin.H{
+		"running": runningProcesses,
+	}
+
+	context.JSON(http.StatusOK, response) // Always return 200 OK
+}
+
+func (pm *ProxyManager) findGroupByModelName(modelName string) *ProcessGroup {
+	for _, group := range pm.processGroups {
+		if group.HasMember(modelName) {
+			return group
+		}
+	}
+	return nil
+}

proxy/proxymanager_api.go

@@ -0,0 +1,229 @@
+package proxy
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"net/http"
+	"sort"
+	"strings"
+
+	"github.com/gin-gonic/gin"
+	"github.com/mostlygeek/llama-swap/event"
+)
+
+type Model struct {
+	Id          string `json:"id"`
+	Name        string `json:"name"`
+	Description string `json:"description"`
+	State       string `json:"state"`
+	Unlisted    bool   `json:"unlisted"`
+}
+
+func addApiHandlers(pm *ProxyManager) {
+	// Add API endpoints for React to consume
+	apiGroup := pm.ginEngine.Group("/api")
+	{
+		apiGroup.POST("/models/unload", pm.apiUnloadAllModels)
+		apiGroup.POST("/models/unload/*model", pm.apiUnloadSingleModelHandler)
+		apiGroup.GET("/events", pm.apiSendEvents)
+		apiGroup.GET("/metrics", pm.apiGetMetrics)
+	}
+}
+
+func (pm *ProxyManager) apiUnloadAllModels(c *gin.Context) {
+	pm.StopProcesses(StopImmediately)
+	c.JSON(http.StatusOK, gin.H{"msg": "ok"})
+}
+
+func (pm *ProxyManager) getModelStatus() []Model {
+	// Extract keys and sort them
+	models := []Model{}
+
+	modelIDs := make([]string, 0, len(pm.config.Models))
+	for modelID := range pm.config.Models {
+		modelIDs = append(modelIDs, modelID)
+	}
+	sort.Strings(modelIDs)
+
+	// Iterate over sorted keys
+	for _, modelID := range modelIDs {
+		// Get process state
+		processGroup := pm.findGroupByModelName(modelID)
+		state := "unknown"
+		if processGroup != nil {
+			process := processGroup.processes[modelID]
+			if process != nil {
+				var stateStr string
+				switch process.CurrentState() {
+				case StateReady:
+					stateStr = "ready"
+				case StateStarting:
+					stateStr = "starting"
+				case StateStopping:
+					stateStr = "stopping"
+				case StateShutdown:
+					stateStr = "shutdown"
+				case StateStopped:
+					stateStr = "stopped"
+				default:
+					stateStr = "unknown"
+				}
+				state = stateStr
+			}
+		}
+		models = append(models, Model{
+			Id:          modelID,
+			Name:        pm.config.Models[modelID].Name,
+			Description: pm.config.Models[modelID].Description,
+			State:       state,
+			Unlisted:    pm.config.Models[modelID].Unlisted,
+		})
+	}
+
+	return models
+}
+
+type messageType string
+
+const (
+	msgTypeModelStatus messageType = "modelStatus"
+	msgTypeLogData     messageType = "logData"
+	msgTypeMetrics     messageType = "metrics"
+)
+
+type messageEnvelope struct {
+	Type messageType `json:"type"`
+	Data string      `json:"data"`
+}
+
+// sends a stream of different message types that happen on the server
+func (pm *ProxyManager) apiSendEvents(c *gin.Context) {
+	c.Header("Content-Type", "text/event-stream")
+	c.Header("Cache-Control", "no-cache")
+	c.Header("Connection", "keep-alive")
+	c.Header("X-Content-Type-Options", "nosniff")
+	// prevent nginx from buffering SSE
+	c.Header("X-Accel-Buffering", "no")
+
+	sendBuffer := make(chan messageEnvelope, 25)
+	ctx, cancel := context.WithCancel(c.Request.Context())
+	sendModels := func() {
+		data, err := json.Marshal(pm.getModelStatus())
+		if err == nil {
+			msg := messageEnvelope{Type: msgTypeModelStatus, Data: string(data)}
+			select {
+			case sendBuffer <- msg:
+			case <-ctx.Done():
+				return
+			default:
+			}
+
+		}
+	}
+
+	sendLogData := func(source string, data []byte) {
+		data, err := json.Marshal(gin.H{
+			"source": source,
+			"data":   string(data),
+		})
+		if err == nil {
+			select {
+			case sendBuffer <- messageEnvelope{Type: msgTypeLogData, Data: string(data)}:
+			case <-ctx.Done():
+				return
+			default:
+			}
+		}
+	}
+
+	sendMetrics := func(metrics []TokenMetrics) {
+		jsonData, err := json.Marshal(metrics)
+		if err == nil {
+			select {
+			case sendBuffer <- messageEnvelope{Type: msgTypeMetrics, Data: string(jsonData)}:
+			case <-ctx.Done():
+				return
+			default:
+			}
+		}
+	}
+
+	/**
+	 * Send updated models list
+	 */
+	defer event.On(func(e ProcessStateChangeEvent) {
+		sendModels()
+	})()
+	defer event.On(func(e ConfigFileChangedEvent) {
+		sendModels()
+	})()
+
+	/**
+	 * Send Log data
+	 */
+	defer pm.proxyLogger.OnLogData(func(data []byte) {
+		sendLogData("proxy", data)
+	})()
+	defer pm.upstreamLogger.OnLogData(func(data []byte) {
+		sendLogData("upstream", data)
+	})()
+
+	/**
+	 * Send Metrics data
+	 */
+	defer event.On(func(e TokenMetricsEvent) {
+		sendMetrics([]TokenMetrics{e.Metrics})
+	})()
+
+	// send initial batch of data
+	sendLogData("proxy", pm.proxyLogger.GetHistory())
+	sendLogData("upstream", pm.upstreamLogger.GetHistory())
+	sendModels()
+	sendMetrics(pm.metricsMonitor.GetMetrics())
+
+	for {
+		select {
+		case <-c.Request.Context().Done():
+			cancel()
+			return
+		case <-pm.shutdownCtx.Done():
+			cancel()
+			return
+		case msg := <-sendBuffer:
+			c.SSEvent("message", msg)
+			c.Writer.Flush()
+		}
+	}
+}
+
+func (pm *ProxyManager) apiGetMetrics(c *gin.Context) {
+	jsonData, err := pm.metricsMonitor.GetMetricsJSON()
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to get metrics"})
+		return
+	}
+	c.Data(http.StatusOK, "application/json", jsonData)
+}
+
+func (pm *ProxyManager) apiUnloadSingleModelHandler(c *gin.Context) {
+	requestedModel := strings.TrimPrefix(c.Param("model"), "/")
+	realModelName, found := pm.config.RealModelName(requestedModel)
+	if !found {
+		pm.sendErrorResponse(c, http.StatusNotFound, "Model not found")
+		return
+	}
+
+	processGroup := pm.findGroupByModelName(realModelName)
+	if processGroup == nil {
+		pm.sendErrorResponse(c, http.StatusInternalServerError, fmt.Sprintf("process group not found for model %s", requestedModel))
+		return
+	}
+
+	if err := processGroup.StopProcess(realModelName, StopImmediately); err != nil {
+		pm.sendErrorResponse(c, http.StatusInternalServerError, fmt.Sprintf("error stopping process: %s", err.Error()))
+		return
+	} else {
+		c.String(http.StatusOK, "OK")
+	}
+}

proxy/proxymanager_loghandlers.go

@@ -0,0 +1,100 @@
+package proxy
+
+import (
+	"context"
+	"fmt"
+	"net/http"
+	"strings"
+
+	"github.com/gin-gonic/gin"
+)
+
+func (pm *ProxyManager) sendLogsHandlers(c *gin.Context) {
+	accept := c.GetHeader("Accept")
+	if strings.Contains(accept, "text/html") {
+		c.Redirect(http.StatusFound, "/ui/")
+	} else {
+		c.Header("Content-Type", "text/plain")
+		history := pm.muxLogger.GetHistory()
+		_, err := c.Writer.Write(history)
+		if err != nil {
+			c.AbortWithError(http.StatusInternalServerError, err)
+			return
+		}
+	}
+}
+
+func (pm *ProxyManager) streamLogsHandler(c *gin.Context) {
+	c.Header("Content-Type", "text/plain")
+	c.Header("Transfer-Encoding", "chunked")
+	c.Header("X-Content-Type-Options", "nosniff")
+	// prevent nginx from buffering streamed logs
+	c.Header("X-Accel-Buffering", "no")
+
+	logMonitorId := c.Param("logMonitorID")
+	logger, err := pm.getLogger(logMonitorId)
+	if err != nil {
+		c.String(http.StatusBadRequest, err.Error())
+		return
+	}
+
+	flusher, ok := c.Writer.(http.Flusher)
+	if !ok {
+		c.AbortWithError(http.StatusInternalServerError, fmt.Errorf("streaming unsupported"))
+		return
+	}
+
+	_, skipHistory := c.GetQuery("no-history")
+	// Send history first if not skipped
+
+	if !skipHistory {
+		history := logger.GetHistory()
+		if len(history) != 0 {
+			c.Writer.Write(history)
+			flusher.Flush()
+		}
+	}
+
+	sendChan := make(chan []byte, 10)
+	ctx, cancel := context.WithCancel(c.Request.Context())
+	defer logger.OnLogData(func(data []byte) {
+		select {
+		case sendChan <- data:
+		case <-ctx.Done():
+			return
+		default:
+		}
+	})()
+
+	for {
+		select {
+		case <-c.Request.Context().Done():
+			cancel()
+			return
+		case <-pm.shutdownCtx.Done():
+			cancel()
+			return
+		case data := <-sendChan:
+			c.Writer.Write(data)
+			flusher.Flush()
+		}
+	}
+}
+
+// getLogger searches for the appropriate logger based on the logMonitorId
+func (pm *ProxyManager) getLogger(logMonitorId string) (*LogMonitor, error) {
+	var logger *LogMonitor
+
+	if logMonitorId == "" {
+		// maintain the default
+		logger = pm.muxLogger
+	} else if logMonitorId == "proxy" {
+		logger = pm.proxyLogger
+	} else if logMonitorId == "upstream" {
+		logger = pm.upstreamLogger
+	} else {
+		return nil, fmt.Errorf("invalid logger. Use 'proxy' or 'upstream'")
+	}
+
+	return logger, nil
+}

proxy/sanitize_cors.go

@@ -0,0 +1,43 @@
+package proxy
+
+import (
+	"strings"
+)
+
+func isTokenChar(r rune) bool {
+	switch {
+	case r >= 'a' && r <= 'z':
+	case r >= 'A' && r <= 'Z':
+	case r >= '0' && r <= '9':
+	case strings.ContainsRune("!#$%&'*+-.^_`|~", r):
+	default:
+		return false
+	}
+	return true
+}
+
+func SanitizeAccessControlRequestHeaderValues(headerValues string) string {
+	parts := strings.Split(headerValues, ",")
+	valid := make([]string, 0, len(parts))
+
+	for _, p := range parts {
+		v := strings.TrimSpace(p)
+		if v == "" {
+			continue
+		}
+
+		validPart := true
+		for _, c := range v {
+			if !isTokenChar(c) {
+				validPart = false
+				break
+			}
+		}
+
+		if validPart {
+			valid = append(valid, v)
+		}
+	}
+
+	return strings.Join(valid, ", ")
+}

proxy/sanitize_cors_test.go

@@ -0,0 +1,77 @@
+package proxy
+
+import "testing"
+
+func TestSanitizeAccessControlRequestHeaderValues(t *testing.T) {
+	tests := []struct {
+		name     string
+		input    string
+		expected string
+	}{
+		{
+			name:     "empty string",
+			input:    "",
+			expected: "",
+		},
+		{
+			name:     "whitespace only",
+			input:    "   ",
+			expected: "",
+		},
+		{
+			name:     "single valid value",
+			input:    "content-type",
+			expected: "content-type",
+		},
+		{
+			name:     "multiple valid values",
+			input:    "content-type, authorization, x-requested-with",
+			expected: "content-type, authorization, x-requested-with",
+		},
+		{
+			name:     "values with extra spaces",
+			input:    "  content-type  ,  authorization  ",
+			expected: "content-type, authorization",
+		},
+		{
+			name:     "values with tabs",
+			input:    "content-type,\tauthorization",
+			expected: "content-type, authorization",
+		},
+		{
+			name:     "values with invalid characters",
+			input:    "content-type, auth\n, x-requested-with\r",
+			expected: "content-type, auth, x-requested-with",
+		},
+		{
+			name:     "empty values in list",
+			input:    "content-type,,authorization",
+			expected: "content-type, authorization",
+		},
+		{
+			name:     "leading and trailing commas",
+			input:    ",content-type,authorization,",
+			expected: "content-type, authorization",
+		},
+		{
+			name:     "mixed valid and invalid values",
+			input:    "content-type, \x00invalid, x-requested-with",
+			expected: "content-type, x-requested-with",
+		},
+		{
+			name:     "mixed case values",
+			input:    "Content-Type, my-Valid-Header, Another-hEader",
+			expected: "Content-Type, my-Valid-Header, Another-hEader",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			got := SanitizeAccessControlRequestHeaderValues(tt.input)
+			if got != tt.expected {
+				t.Errorf("SanitizeAccessControlRequestHeaderValues(%q) = %q, want %q",
+					tt.input, got, tt.expected)
+			}
+		})
+	}
+}

proxy/ui_embed.go

@@ -0,0 +1,24 @@
+package proxy
+
+import (
+	"embed"
+	"io/fs"
+	"net/http"
+)
+
+//go:embed ui_dist
+var reactStaticFS embed.FS
+
+// GetReactFS returns the embedded React filesystem
+func GetReactFS() (http.FileSystem, error) {
+	subFS, err := fs.Sub(reactStaticFS, "ui_dist")
+	if err != nil {
+		return nil, err
+	}
+	return http.FS(subFS), nil
+}
+
+// GetReactIndexHTML returns the main index.html for the React app
+func GetReactIndexHTML() ([]byte, error) {
+	return reactStaticFS.ReadFile("ui_dist/index.html")
+}

scripts/install.sh

@@ -0,0 +1,213 @@
+#!/bin/sh
+# This script installs llama-swap on Linux.
+# It detects the current operating system architecture and installs the appropriate version of llama-swap.
+
+set -eu
+
+LLAMA_SWAP_DEFAULT_ADDRESS=${LLAMA_SWAP_DEFAULT_ADDRESS:-"127.0.0.1:8080"}
+
+red="$( (/usr/bin/tput bold || :; /usr/bin/tput setaf 1 || :) 2>&-)"
+plain="$( (/usr/bin/tput sgr0 || :) 2>&-)"
+
+status() { echo ">>> $*" >&2; }
+error() { echo "${red}ERROR:${plain} $*"; exit 1; }
+warning() { echo "${red}WARNING:${plain} $*"; }
+
+available() { command -v "$1" >/dev/null; }
+require() {
+    _MISSING=''
+    for TOOL in "$@"; do
+        if ! available "$TOOL"; then
+            _MISSING="$_MISSING $TOOL"
+        fi
+    done
+
+    echo "$_MISSING"
+}
+
+SUDO=
+if [ "$(id -u)" -ne 0 ]; then
+    if ! available sudo; then
+        error "This script requires superuser permissions. Please re-run as root."
+    fi
+
+    SUDO="sudo"
+fi
+
+NEEDS=$(require tee tar python3 mktemp)
+if [ -n "$NEEDS" ]; then
+    status "ERROR: The following tools are required but missing:"
+    for NEED in $NEEDS; do
+        echo "  - $NEED"
+    done
+    exit 1
+fi
+
+[ "$(uname -s)" = "Linux" ] || error 'This script is intended to run on Linux only.'
+
+ARCH=$(uname -m)
+case "$ARCH" in
+    x86_64) ARCH="amd64" ;;
+    aarch64|arm64) ARCH="arm64" ;;
+    *) error "Unsupported architecture: $ARCH" ;;
+esac
+
+IS_WSL2=false
+
+KERN=$(uname -r)
+case "$KERN" in
+    *icrosoft*WSL2 | *icrosoft*wsl2) IS_WSL2=true;;
+    *icrosoft) error "Microsoft WSL1 is not currently supported. Please use WSL2 with 'wsl --set-version <distro> 2'" ;;
+    *) ;;
+esac
+
+download_binary() {
+    ASSET_NAME="linux_$ARCH"
+
+    TMPDIR=$(mktemp -d)
+    trap 'rm -rf "${TMPDIR}"' EXIT INT TERM HUP
+    PYTHON_SCRIPT=$(cat <<EOF
+import os
+import json
+import sys
+import urllib.request
+
+ASSET_NAME = "${ASSET_NAME}"
+
+with urllib.request.urlopen("https://api.github.com/repos/mostlygeek/llama-swap/releases/latest") as resp:
+    data = json.load(resp)
+    for asset in data.get("assets", []):
+        if ASSET_NAME in asset.get("name", ""):
+            url = asset["browser_download_url"]
+            break
+    else:
+        print("ERROR: Matching asset not found.", file=sys.stderr)
+        exit(1)
+
+print("Downloading:", url, file=sys.stderr)
+output_path = os.path.join("${TMPDIR}", "llama-swap.tar.gz")
+urllib.request.urlretrieve(url, output_path)
+print(output_path)
+EOF
+)
+
+    TARFILE=$(python3 -c "$PYTHON_SCRIPT")
+    if [ ! -f "$TARFILE" ]; then
+        error "Failed to download binary."
+    fi
+
+    status "Extracting to /usr/local/bin"
+    $SUDO tar -xzf "$TARFILE" -C /usr/local/bin llama-swap
+}
+download_binary
+
+configure_systemd() {
+    if ! id llama-swap >/dev/null 2>&1; then
+        status "Creating llama-swap user..."
+        $SUDO useradd -r -s /bin/false -U -m -d /usr/share/llama-swap llama-swap
+    fi
+    if getent group render >/dev/null 2>&1; then
+        status "Adding llama-swap user to render group..."
+        $SUDO usermod -a -G render llama-swap
+    fi
+    if getent group video >/dev/null 2>&1; then
+        status "Adding llama-swap user to video group..."
+        $SUDO usermod -a -G video llama-swap
+    fi
+    if getent group docker >/dev/null 2>&1; then
+        status "Adding llama-swap user to docker group..."
+        $SUDO usermod -a -G docker llama-swap
+    fi
+
+    status "Adding current user to llama-swap group..."
+    $SUDO usermod -a -G llama-swap "$(whoami)"
+
+    if [ ! -f "/usr/share/llama-swap/config.yaml" ]; then
+        status "Creating default config.yaml..."
+        cat <<EOF | $SUDO -u llama-swap tee /usr/share/llama-swap/config.yaml >/dev/null
+# default 15s likely to fail for default models due to downloading models
+healthCheckTimeout: 60
+
+models:
+  "qwen2.5":
+    cmd: |
+      docker run
+        --rm
+        -p \${PORT}:8080
+        --name qwen2.5
+      ghcr.io/ggml-org/llama.cpp:server
+        -hf bartowski/Qwen2.5-0.5B-Instruct-GGUF:Q4_K_M
+    cmdStop: docker stop qwen2.5
+
+  "smollm2":
+    cmd: |
+      docker run
+        --rm
+        -p \${PORT}:8080
+        --name smollm2
+      ghcr.io/ggml-org/llama.cpp:server
+        -hf bartowski/SmolLM2-135M-Instruct-GGUF:Q4_K_M
+    cmdStop: docker stop smollm2
+EOF
+    fi
+
+    status "Creating llama-swap systemd service..."
+    cat <<EOF | $SUDO tee /etc/systemd/system/llama-swap.service >/dev/null
+[Unit]
+Description=llama-swap
+After=network.target
+
+[Service]
+User=llama-swap
+Group=llama-swap
+
+# set this to match your environment
+ExecStart=/usr/local/bin/llama-swap --config /usr/share/llama-swap/config.yaml --watch-config -listen ${LLAMA_SWAP_DEFAULT_ADDRESS}
+
+Restart=on-failure
+RestartSec=3
+StartLimitBurst=3
+StartLimitInterval=30
+
+[Install]
+WantedBy=multi-user.target
+EOF
+    SYSTEMCTL_RUNNING="$(systemctl is-system-running || true)"
+    case $SYSTEMCTL_RUNNING in
+        running|degraded)
+            status "Enabling and starting llama-swap service..."
+            $SUDO systemctl daemon-reload
+            $SUDO systemctl enable llama-swap
+
+            start_service() { $SUDO systemctl restart llama-swap; }
+            trap start_service EXIT
+            ;;
+        *)
+            warning "systemd is not running"
+            if [ "$IS_WSL2" = true ]; then
+                warning "see https://learn.microsoft.com/en-us/windows/wsl/systemd#how-to-enable-systemd to enable it"
+            fi
+            ;;
+    esac
+}
+
+if available systemctl; then
+    configure_systemd
+fi
+
+install_success() {
+    status "The llama-swap API is now available at http://${LLAMA_SWAP_DEFAULT_ADDRESS}"
+    status 'Customize the config file at /usr/share/llama-swap/config.yaml.'
+    status 'Install complete.'
+}
+
+# WSL2 only supports GPUs via nvidia passthrough
+# so check for nvidia-smi to determine if GPU is available
+if [ "$IS_WSL2" = true ]; then
+    if available nvidia-smi && [ -n "$(nvidia-smi | grep -o "CUDA Version: [0-9]*\.[0-9]*")" ]; then
+        status "Nvidia GPU detected."
+    fi
+    exit 0
+fi
+
+install_success

scripts/uninstall.sh

@@ -0,0 +1,68 @@
+#!/bin/sh
+# This script uninstalls llama-swap on Linux.
+# It removes the binary, systemd service, config.yaml (optional), and llama-swap user and group.
+
+set -eu
+
+red="$( (/usr/bin/tput bold || :; /usr/bin/tput setaf 1 || :) 2>&-)"
+plain="$( (/usr/bin/tput sgr0 || :) 2>&-)"
+
+status() { echo ">>> $*" >&2; }
+error() { echo "${red}ERROR:${plain} $*"; exit 1; }
+warning() { echo "${red}WARNING:${plain} $*"; }
+
+available() { command -v $1 >/dev/null; }
+
+SUDO=
+if [ "$(id -u)" -ne 0 ]; then
+    if ! available sudo; then
+        error "This script requires superuser permissions. Please re-run as root."
+    fi
+
+    SUDO="sudo"
+fi
+
+configure_systemd() {
+    status "Stopping llama-swap service..."
+    $SUDO systemctl stop llama-swap
+
+    status "Disabling llama-swap service..."
+    $SUDO systemctl disable llama-swap
+}
+if available systemctl; then
+    configure_systemd
+fi
+
+if available llama-swap; then
+    status "Removing llama-swap binary..."
+    $SUDO rm $(which llama-swap)
+fi
+
+if [ -f "/usr/share/llama-swap/config.yaml" ]; then
+    while true; do
+        printf "Delete config.yaml (/usr/share/llama-swap/config.yaml)? [y/N] " >&2
+        read answer
+        case "$answer" in
+            [Yy]* ) 
+                $SUDO rm -r /usr/share/llama-swap
+                break
+                ;;
+            [Nn]* | "" ) 
+                break
+                ;;
+            * ) 
+                echo "Invalid input. Please enter y or n."
+                ;;
+        esac
+    done
+fi
+
+if id llama-swap >/dev/null 2>&1; then
+    status "Removing llama-swap user..."
+    $SUDO userdel llama-swap
+fi
+
+if getent group llama-swap >/dev/null 2>&1; then
+    status "Removing llama-swap group..."
+    $SUDO groupdel llama-swap
+fi

ui/.gitignore

@@ -0,0 +1,25 @@
+.vite
+# Logs
+logs
+*.log
+npm-debug.log*
+yarn-debug.log*
+yarn-error.log*
+pnpm-debug.log*
+lerna-debug.log*
+
+node_modules
+dist
+dist-ssr
+*.local
+
+# Editor directories and files
+.vscode/*
+!.vscode/extensions.json
+.idea
+.DS_Store
+*.suo
+*.ntvs*
+*.njsproj
+*.sln
+*.sw?

ui/README.md

@@ -0,0 +1,54 @@
+# React + TypeScript + Vite
+
+This template provides a minimal setup to get React working in Vite with HMR and some ESLint rules.
+
+Currently, two official plugins are available:
+
+- [@vitejs/plugin-react](https://github.com/vitejs/vite-plugin-react/blob/main/packages/plugin-react) uses [Babel](https://babeljs.io/) for Fast Refresh
+- [@vitejs/plugin-react-swc](https://github.com/vitejs/vite-plugin-react/blob/main/packages/plugin-react-swc) uses [SWC](https://swc.rs/) for Fast Refresh
+
+## Expanding the ESLint configuration
+
+If you are developing a production application, we recommend updating the configuration to enable type-aware lint rules:
+
+```js
+export default tseslint.config({
+  extends: [
+    // Remove ...tseslint.configs.recommended and replace with this
+    ...tseslint.configs.recommendedTypeChecked,
+    // Alternatively, use this for stricter rules
+    ...tseslint.configs.strictTypeChecked,
+    // Optionally, add this for stylistic rules
+    ...tseslint.configs.stylisticTypeChecked,
+  ],
+  languageOptions: {
+    // other options...
+    parserOptions: {
+      project: ['./tsconfig.node.json', './tsconfig.app.json'],
+      tsconfigRootDir: import.meta.dirname,
+    },
+  },
+})
+```
+
+You can also install [eslint-plugin-react-x](https://github.com/Rel1cx/eslint-react/tree/main/packages/plugins/eslint-plugin-react-x) and [eslint-plugin-react-dom](https://github.com/Rel1cx/eslint-react/tree/main/packages/plugins/eslint-plugin-react-dom) for React-specific lint rules:
+
+```js
+// eslint.config.js
+import reactX from 'eslint-plugin-react-x'
+import reactDom from 'eslint-plugin-react-dom'
+
+export default tseslint.config({
+  plugins: {
+    // Add the react-x and react-dom plugins
+    'react-x': reactX,
+    'react-dom': reactDom,
+  },
+  rules: {
+    // other rules...
+    // Enable its recommended typescript rules
+    ...reactX.configs['recommended-typescript'].rules,
+    ...reactDom.configs.recommended.rules,
+  },
+})
+```

ui/eslint.config.js

@@ -0,0 +1,28 @@
+import js from '@eslint/js'
+import globals from 'globals'
+import reactHooks from 'eslint-plugin-react-hooks'
+import reactRefresh from 'eslint-plugin-react-refresh'
+import tseslint from 'typescript-eslint'
+
+export default tseslint.config(
+  { ignores: ['dist'] },
+  {
+    extends: [js.configs.recommended, ...tseslint.configs.recommended],
+    files: ['**/*.{ts,tsx}'],
+    languageOptions: {
+      ecmaVersion: 2020,
+      globals: globals.browser,
+    },
+    plugins: {
+      'react-hooks': reactHooks,
+      'react-refresh': reactRefresh,
+    },
+    rules: {
+      ...reactHooks.configs.recommended.rules,
+      'react-refresh/only-export-components': [
+        'warn',
+        { allowConstantExport: true },
+      ],
+    },
+  },
+)

ui/index.html

@@ -0,0 +1,17 @@
+<!doctype html>
+<html lang="en">
+  <head>
+    <meta charset="UTF-8" />
+    <meta name="viewport" content="width=device-width, initial-scale=1.0" />
+    <link rel="icon" type="image/png" href="/favicon-96x96.png" sizes="96x96" />
+    <link rel="icon" type="image/svg+xml" href="/favicon.svg" />
+    <link rel="shortcut icon" href="/favicon.ico" />
+    <link rel="apple-touch-icon" sizes="180x180" href="/apple-touch-icon.png" />
+    <link rel="manifest" href="/site.webmanifest" />
+    <title>llama-swap</title>
+  </head>
+  <body >
+    <div id="root"></div>
+    <script type="module" src="/src/main.tsx"></script>
+  </body>
+</html>

ui/package.json

@@ -0,0 +1,34 @@
+{
+  "name": "ui",
+  "private": true,
+  "version": "0.0.0",
+  "type": "module",
+  "scripts": {
+    "start": "vite",
+    "build": "tsc -b && vite build --emptyOutDir",
+    "lint": "eslint .",
+    "preview": "vite preview"
+  },
+  "dependencies": {
+    "react": "^19.1.0",
+    "react-dom": "^19.1.0",
+    "react-icons": "^5.5.0",
+    "react-resizable-panels": "^3.0.4",
+    "react-router-dom": "^7.6.2"
+  },
+  "devDependencies": {
+    "@eslint/js": "^9.25.0",
+    "@tailwindcss/vite": "^4.1.8",
+    "@types/react": "^19.1.2",
+    "@types/react-dom": "^19.1.2",
+    "@vitejs/plugin-react": "^4.4.1",
+    "eslint": "^9.25.0",
+    "eslint-plugin-react-hooks": "^5.2.0",
+    "eslint-plugin-react-refresh": "^0.4.19",
+    "globals": "^16.0.0",
+    "tailwindcss": "^4.1.8",
+    "typescript": "~5.8.3",
+    "typescript-eslint": "^8.30.1",
+    "vite": "^6.3.5"
+  }
+}

ui/public/site.webmanifest

@@ -0,0 +1,21 @@
+{
+  "name": "llama-swap",
+  "short_name": "llama-swap",
+  "icons": [
+    {
+      "src": "/web-app-manifest-192x192.png",
+      "sizes": "192x192",
+      "type": "image/png",
+      "purpose": "maskable"
+    },
+    {
+      "src": "/web-app-manifest-512x512.png",
+      "sizes": "512x512",
+      "type": "image/png",
+      "purpose": "maskable"
+    }
+  ],
+  "theme_color": "#ffffff",
+  "background_color": "#ffffff",
+  "display": "standalone"
+}

ui/src/App.css

@@ -0,0 +1,6 @@
+#root {
+  max-width: 1280px;
+  margin: 0 auto;
+  padding: 2rem;
+  text-align: center;
+}

ui/src/App.tsx

@@ -0,0 +1,38 @@
+import { useEffect } from "react";
+import { Navigate, Route, BrowserRouter as Router, Routes } from "react-router-dom";
+import { Header } from "./components/Header";
+import { useAPI } from "./contexts/APIProvider";
+import { useTheme } from "./contexts/ThemeProvider";
+import ActivityPage from "./pages/Activity";
+import LogViewerPage from "./pages/LogViewer";
+import ModelPage from "./pages/Models";
+
+function App() {
+  const { setConnectionState } = useTheme();
+
+  const { connectionStatus } = useAPI();
+
+  // Synchronize the window.title connections state with the actual connection state
+  useEffect(() => {
+    setConnectionState(connectionStatus);
+  }, [connectionStatus]);
+
+  return (
+    <Router basename="/ui/">
+      <div className="flex flex-col h-screen">
+        <Header />
+
+        <main className="flex-1 overflow-auto p-4">
+          <Routes>
+            <Route path="/" element={<LogViewerPage />} />
+            <Route path="/models" element={<ModelPage />} />
+            <Route path="/activity" element={<ActivityPage />} />
+            <Route path="*" element={<Navigate to="/" replace />} />
+          </Routes>
+        </main>
+      </div>
+    </Router>
+  );
+}
+
+export default App;

ui/src/assets/react.svg

@@ -0,0 +1 @@
+<svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" class="iconify iconify--logos" width="35.93" height="32" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 228"><path fill="#00D8FF" d="M210.483 73.824a171.49 171.49 0 0 0-8.24-2.597c.465-1.9.893-3.777 1.273-5.621c6.238-30.281 2.16-54.676-11.769-62.708c-13.355-7.7-35.196.329-57.254 19.526a171.23 171.23 0 0 0-6.375 5.848a155.866 155.866 0 0 0-4.241-3.917C100.759 3.829 77.587-4.822 63.673 3.233C50.33 10.957 46.379 33.89 51.995 62.588a170.974 170.974 0 0 0 1.892 8.48c-3.28.932-6.445 1.924-9.474 2.98C17.309 83.498 0 98.307 0 113.668c0 15.865 18.582 31.778 46.812 41.427a145.52 145.52 0 0 0 6.921 2.165a167.467 167.467 0 0 0-2.01 9.138c-5.354 28.2-1.173 50.591 12.134 58.266c13.744 7.926 36.812-.22 59.273-19.855a145.567 145.567 0 0 0 5.342-4.923a168.064 168.064 0 0 0 6.92 6.314c21.758 18.722 43.246 26.282 56.54 18.586c13.731-7.949 18.194-32.003 12.4-61.268a145.016 145.016 0 0 0-1.535-6.842c1.62-.48 3.21-.974 4.76-1.488c29.348-9.723 48.443-25.443 48.443-41.52c0-15.417-17.868-30.326-45.517-39.844Zm-6.365 70.984c-1.4.463-2.836.91-4.3 1.345c-3.24-10.257-7.612-21.163-12.963-32.432c5.106-11 9.31-21.767 12.459-31.957c2.619.758 5.16 1.557 7.61 2.4c23.69 8.156 38.14 20.213 38.14 29.504c0 9.896-15.606 22.743-40.946 31.14Zm-10.514 20.834c2.562 12.94 2.927 24.64 1.23 33.787c-1.524 8.219-4.59 13.698-8.382 15.893c-8.067 4.67-25.32-1.4-43.927-17.412a156.726 156.726 0 0 1-6.437-5.87c7.214-7.889 14.423-17.06 21.459-27.246c12.376-1.098 24.068-2.894 34.671-5.345a134.17 134.17 0 0 1 1.386 6.193ZM87.276 214.515c-7.882 2.783-14.16 2.863-17.955.675c-8.075-4.657-11.432-22.636-6.853-46.752a156.923 156.923 0 0 1 1.869-8.499c10.486 2.32 22.093 3.988 34.498 4.994c7.084 9.967 14.501 19.128 21.976 27.15a134.668 134.668 0 0 1-4.877 4.492c-9.933 8.682-19.886 14.842-28.658 17.94ZM50.35 144.747c-12.483-4.267-22.792-9.812-29.858-15.863c-6.35-5.437-9.555-10.836-9.555-15.216c0-9.322 13.897-21.212 37.076-29.293c2.813-.98 5.757-1.905 8.812-2.773c3.204 10.42 7.406 21.315 12.477 32.332c-5.137 11.18-9.399 22.249-12.634 32.792a134.718 134.718 0 0 1-6.318-1.979Zm12.378-84.26c-4.811-24.587-1.616-43.134 6.425-47.789c8.564-4.958 27.502 2.111 47.463 19.835a144.318 144.318 0 0 1 3.841 3.545c-7.438 7.987-14.787 17.08-21.808 26.988c-12.04 1.116-23.565 2.908-34.161 5.309a160.342 160.342 0 0 1-1.76-7.887Zm110.427 27.268a347.8 347.8 0 0 0-7.785-12.803c8.168 1.033 15.994 2.404 23.343 4.08c-2.206 7.072-4.956 14.465-8.193 22.045a381.151 381.151 0 0 0-7.365-13.322Zm-45.032-43.861c5.044 5.465 10.096 11.566 15.065 18.186a322.04 322.04 0 0 0-30.257-.006c4.974-6.559 10.069-12.652 15.192-18.18ZM82.802 87.83a323.167 323.167 0 0 0-7.227 13.238c-3.184-7.553-5.909-14.98-8.134-22.152c7.304-1.634 15.093-2.97 23.209-3.984a321.524 321.524 0 0 0-7.848 12.897Zm8.081 65.352c-8.385-.936-16.291-2.203-23.593-3.793c2.26-7.3 5.045-14.885 8.298-22.6a321.187 321.187 0 0 0 7.257 13.246c2.594 4.48 5.28 8.868 8.038 13.147Zm37.542 31.03c-5.184-5.592-10.354-11.779-15.403-18.433c4.902.192 9.899.29 14.978.29c5.218 0 10.376-.117 15.453-.343c-4.985 6.774-10.018 12.97-15.028 18.486Zm52.198-57.817c3.422 7.8 6.306 15.345 8.596 22.52c-7.422 1.694-15.436 3.058-23.88 4.071a382.417 382.417 0 0 0 7.859-13.026a347.403 347.403 0 0 0 7.425-13.565Zm-16.898 8.101a358.557 358.557 0 0 1-12.281 19.815a329.4 329.4 0 0 1-23.444.823c-7.967 0-15.716-.248-23.178-.732a310.202 310.202 0 0 1-12.513-19.846h.001a307.41 307.41 0 0 1-10.923-20.627a310.278 310.278 0 0 1 10.89-20.637l-.001.001a307.318 307.318 0 0 1 12.413-19.761c7.613-.576 15.42-.876 23.31-.876H128c7.926 0 15.743.303 23.354.883a329.357 329.357 0 0 1 12.335 19.695a358.489 358.489 0 0 1 11.036 20.54a329.472 329.472 0 0 1-11 20.722Zm22.56-122.124c8.572 4.944 11.906 24.881 6.52 51.026c-.344 1.668-.73 3.367-1.15 5.09c-10.622-2.452-22.155-4.275-34.23-5.408c-7.034-10.017-14.323-19.124-21.64-27.008a160.789 160.789 0 0 1 5.888-5.4c18.9-16.447 36.564-22.941 44.612-18.3ZM128 90.808c12.625 0 22.86 10.235 22.86 22.86s-10.235 22.86-22.86 22.86s-22.86-10.235-22.86-22.86s10.235-22.86 22.86-22.86Z"></path></svg>

ui/src/components/ConnectionStatus.tsx

@@ -0,0 +1,26 @@
+import { useAPI } from "../contexts/APIProvider";
+import { useMemo } from "react";
+
+const ConnectionStatusIcon = () => {
+  const { connectionStatus } = useAPI();
+
+  const eventStatusColor = useMemo(() => {
+    switch (connectionStatus) {
+      case "connected":
+        return "bg-emerald-500";
+      case "connecting":
+        return "bg-amber-500";
+      case "disconnected":
+      default:
+        return "bg-red-500";
+    }
+  }, [connectionStatus]);
+
+  return (
+    <div className="flex items-center" title={`event stream: ${connectionStatus}`}>
+      <span className={`inline-block w-3 h-3 rounded-full ${eventStatusColor} mr-2`}></span>
+    </div>
+  );
+};
+
+export default ConnectionStatusIcon;

ui/src/components/Header.tsx

@@ -0,0 +1,56 @@
+import { useCallback } from "react";
+import { RiMoonFill, RiSunFill } from "react-icons/ri";
+import { NavLink, type NavLinkRenderProps } from "react-router-dom";
+import { useTheme } from "../contexts/ThemeProvider";
+import ConnectionStatusIcon from "./ConnectionStatus";
+
+export function Header() {
+  const { screenWidth, toggleTheme, isDarkMode, appTitle, setAppTitle, isNarrow } = useTheme();
+  const handleTitleChange = useCallback(
+    (newTitle: string) => {
+      setAppTitle(newTitle.replace(/\n/g, "").trim().substring(0, 64) || "llama-swap");
+    },
+    [setAppTitle]
+  );
+
+  const navLinkClass = ({ isActive }: NavLinkRenderProps) =>
+    `text-gray-600 hover:text-black dark:text-gray-300 dark:hover:text-gray-100 p-1 ${isActive ? "font-semibold" : ""}`;
+
+  return (
+    <header className={`flex items-center justify-between bg-surface border-b border-border px-4 ${isNarrow ? "py-1 h-[60px]" : "p-2 h-[75px]"}`}>
+      {screenWidth !== "xs" && screenWidth !== "sm" && (
+        <h1
+          contentEditable
+          suppressContentEditableWarning
+          className="p-0 outline-none hover:bg-gray-100 dark:hover:bg-gray-700 rounded"
+          onBlur={(e) => handleTitleChange(e.currentTarget.textContent || "(set title)")}
+          onKeyDown={(e) => {
+            if (e.key === "Enter") {
+              e.preventDefault();
+              handleTitleChange(e.currentTarget.textContent || "(set title)");
+              e.currentTarget.blur();
+            }
+          }}
+        >
+          {appTitle}
+        </h1>
+      )}
+
+      <menu className="flex items-center gap-4">
+        <NavLink to="/" className={navLinkClass} type="button">
+          Logs
+        </NavLink>
+        <NavLink to="/models" className={navLinkClass} type="button">
+          Models
+        </NavLink>
+        <NavLink to="/activity" className={navLinkClass} type="button">
+          Activity
+        </NavLink>
+        <button className="" onClick={toggleTheme}>
+          {isDarkMode ? <RiMoonFill /> : <RiSunFill />}
+        </button>
+        <ConnectionStatusIcon />
+      </menu>
+    </header>
+  );
+}

ui/src/contexts/APIProvider.tsx

@@ -0,0 +1,246 @@
+import { createContext, useState, useContext, useEffect, useCallback, useMemo, type ReactNode } from "react";
+import type { ConnectionState } from "../lib/types";
+
+type ModelStatus = "ready" | "starting" | "stopping" | "stopped" | "shutdown" | "unknown";
+const LOG_LENGTH_LIMIT = 1024 * 100; /* 100KB of log data */
+
+export interface Model {
+  id: string;
+  state: ModelStatus;
+  name: string;
+  description: string;
+  unlisted: boolean;
+}
+
+interface APIProviderType {
+  models: Model[];
+  listModels: () => Promise<Model[]>;
+  unloadAllModels: () => Promise<void>;
+  unloadSingleModel: (model: string) => Promise<void>;
+  loadModel: (model: string) => Promise<void>;
+  enableAPIEvents: (enabled: boolean) => void;
+  proxyLogs: string;
+  upstreamLogs: string;
+  metrics: Metrics[];
+  connectionStatus: ConnectionState;
+}
+
+interface Metrics {
+  id: number;
+  timestamp: string;
+  model: string;
+  cache_tokens: number;
+  input_tokens: number;
+  output_tokens: number;
+  prompt_per_second: number;
+  tokens_per_second: number;
+  duration_ms: number;
+}
+
+interface LogData {
+  source: "upstream" | "proxy";
+  data: string;
+}
+interface APIEventEnvelope {
+  type: "modelStatus" | "logData" | "metrics";
+  data: string;
+}
+
+const APIContext = createContext<APIProviderType | undefined>(undefined);
+type APIProviderProps = {
+  children: ReactNode;
+  autoStartAPIEvents?: boolean;
+};
+
+let apiEventSource: EventSource | null = null;
+
+export function APIProvider({ children, autoStartAPIEvents = true }: APIProviderProps) {
+  const [proxyLogs, setProxyLogs] = useState("");
+  const [upstreamLogs, setUpstreamLogs] = useState("");
+  const [metrics, setMetrics] = useState<Metrics[]>([]);
+  const [connectionStatus, setConnectionState] = useState<ConnectionState>("disconnected");
+  //const apiEventSource = useRef<EventSource | null>(null);
+
+  const [models, setModels] = useState<Model[]>([]);
+
+  const appendLog = useCallback((newData: string, setter: React.Dispatch<React.SetStateAction<string>>) => {
+    setter((prev) => {
+      const updatedLog = prev + newData;
+      return updatedLog.length > LOG_LENGTH_LIMIT ? updatedLog.slice(-LOG_LENGTH_LIMIT) : updatedLog;
+    });
+  }, []);
+
+  const enableAPIEvents = useCallback((enabled: boolean) => {
+    if (!enabled) {
+      apiEventSource?.close();
+      apiEventSource = null;
+      setMetrics([]);
+      return;
+    }
+
+    let retryCount = 0;
+    const initialDelay = 1000; // 1 second
+
+    const connect = () => {
+      apiEventSource?.close();
+      apiEventSource = new EventSource("/api/events");
+
+      setConnectionState("connecting");
+
+      apiEventSource.onopen = () => {
+        // clear everything out on connect to keep things in sync
+        setProxyLogs("");
+        setUpstreamLogs("");
+        setMetrics([]); // clear metrics on reconnect
+        setModels([]); // clear models on reconnect
+        retryCount = 0;
+        setConnectionState("connected");
+      };
+
+      apiEventSource.onmessage = (e: MessageEvent) => {
+        try {
+          const message = JSON.parse(e.data) as APIEventEnvelope;
+          switch (message.type) {
+            case "modelStatus":
+              {
+                const models = JSON.parse(message.data) as Model[];
+
+                // sort models by name and id
+                models.sort((a, b) => {
+                  return (a.name + a.id).localeCompare(b.name + b.id);
+                });
+
+                setModels(models);
+              }
+              break;
+
+            case "logData":
+              const logData = JSON.parse(message.data) as LogData;
+              switch (logData.source) {
+                case "proxy":
+                  appendLog(logData.data, setProxyLogs);
+                  break;
+                case "upstream":
+                  appendLog(logData.data, setUpstreamLogs);
+                  break;
+              }
+              break;
+
+            case "metrics":
+              {
+                const newMetrics = JSON.parse(message.data) as Metrics[];
+                setMetrics((prevMetrics) => {
+                  return [...newMetrics, ...prevMetrics];
+                });
+              }
+              break;
+          }
+        } catch (err) {
+          console.error(e.data, err);
+        }
+      };
+
+      apiEventSource.onerror = () => {
+        apiEventSource?.close();
+        retryCount++;
+        const delay = Math.min(initialDelay * Math.pow(2, retryCount - 1), 5000);
+        setConnectionState("disconnected");
+        setTimeout(connect, delay);
+      };
+    };
+
+    connect();
+  }, []);
+
+  useEffect(() => {
+    if (autoStartAPIEvents) {
+      enableAPIEvents(true);
+    }
+
+    return () => {
+      enableAPIEvents(false);
+    };
+  }, [enableAPIEvents, autoStartAPIEvents]);
+
+  const listModels = useCallback(async (): Promise<Model[]> => {
+    try {
+      const response = await fetch("/api/models/");
+      if (!response.ok) {
+        throw new Error(`HTTP error! status: ${response.status}`);
+      }
+      const data = await response.json();
+      return data || [];
+    } catch (error) {
+      console.error("Failed to fetch models:", error);
+      return []; // Return empty array as fallback
+    }
+  }, []);
+
+  const unloadAllModels = useCallback(async () => {
+    try {
+      const response = await fetch(`/api/models/unload`, {
+        method: "POST",
+      });
+      if (!response.ok) {
+        throw new Error(`Failed to unload models: ${response.status}`);
+      }
+    } catch (error) {
+      console.error("Failed to unload models:", error);
+      throw error; // Re-throw to let calling code handle it
+    }
+  }, []);
+
+  const unloadSingleModel = useCallback(async (model: string) => {
+    try {
+      const response = await fetch(`/api/models/unload/${model}`, {
+        method: "POST",
+      });
+      if (!response.ok) {
+        throw new Error(`Failed to unload model: ${response.status}`);
+      }
+    } catch (error) {
+      console.error("Failed to unload model", model, error);
+      throw error;
+    }
+  }, []);
+
+  const loadModel = useCallback(async (model: string) => {
+    try {
+      const response = await fetch(`/upstream/${model}/`, {
+        method: "GET",
+      });
+      if (!response.ok) {
+        throw new Error(`Failed to load model: ${response.status}`);
+      }
+    } catch (error) {
+      console.error("Failed to load model:", error);
+      throw error; // Re-throw to let calling code handle it
+    }
+  }, []);
+
+  const value = useMemo(
+    () => ({
+      models,
+      listModels,
+      unloadAllModels,
+      unloadSingleModel,
+      loadModel,
+      enableAPIEvents,
+      proxyLogs,
+      upstreamLogs,
+      metrics,
+      connectionStatus,
+    }),
+    [models, listModels, unloadAllModels, loadModel, enableAPIEvents, proxyLogs, upstreamLogs, metrics]
+  );
+
+  return <APIContext.Provider value={value}>{children}</APIContext.Provider>;
+}
+
+export function useAPI() {
+  const context = useContext(APIContext);
+  if (context === undefined) {
+    throw new Error("useAPI must be used within an APIProvider");
+  }
+  return context;
+}

ui/src/contexts/ThemeProvider.tsx

@@ -0,0 +1,97 @@
+import { createContext, useContext, useEffect, type ReactNode, useMemo, useState } from "react";
+import { usePersistentState } from "../hooks/usePersistentState";
+import type { ConnectionState } from "../lib/types";
+
+type ScreenWidth = "xs" | "sm" | "md" | "lg" | "xl" | "2xl";
+type ThemeContextType = {
+  isDarkMode: boolean;
+  screenWidth: ScreenWidth;
+  isNarrow: boolean;
+  toggleTheme: () => void;
+
+  // for managing the window title and connection state information
+  appTitle: string;
+  setAppTitle: (title: string) => void;
+  setConnectionState: (state: ConnectionState) => void;
+};
+
+const ThemeContext = createContext<ThemeContextType | undefined>(undefined);
+
+type ThemeProviderProps = {
+  children: ReactNode;
+};
+
+export function ThemeProvider({ children }: ThemeProviderProps) {
+  const [appTitle, setAppTitle] = usePersistentState("app-title", "llama-swap");
+  const [connectionState, setConnectionState] = useState<ConnectionState>("disconnected");
+
+  /**
+   * Set the document.title with informative information
+   */
+  useEffect(() => {
+    const connectionIcon = connectionState === "connecting" ? "🟡" : connectionState === "connected" ? "🟢" : "🔴";
+    document.title = connectionIcon + " " + appTitle; // Set initial title
+  }, [appTitle, connectionState]);
+
+  const [isDarkMode, setIsDarkMode] = usePersistentState<boolean>("theme", false);
+  const [screenWidth, setScreenWidth] = useState<ScreenWidth>("md"); // Default to md
+
+  // matches tailwind classes
+  // https://tailwindcss.com/docs/responsive-design
+  useEffect(() => {
+    const checkInnerWidth = () => {
+      const innerWidth = window.innerWidth;
+      if (innerWidth < 640) {
+        setScreenWidth("xs");
+      } else if (innerWidth < 768) {
+        setScreenWidth("sm");
+      } else if (innerWidth < 1024) {
+        setScreenWidth("md");
+      } else if (innerWidth < 1280) {
+        setScreenWidth("lg");
+      } else if (innerWidth < 1536) {
+        setScreenWidth("xl");
+      } else {
+        setScreenWidth("2xl");
+      }
+    };
+
+    checkInnerWidth();
+    window.addEventListener("resize", checkInnerWidth);
+
+    return () => window.removeEventListener("resize", checkInnerWidth);
+  }, []);
+
+  useEffect(() => {
+    document.documentElement.setAttribute("data-theme", isDarkMode ? "dark" : "light");
+  }, [isDarkMode]);
+
+  const toggleTheme = () => setIsDarkMode((prev) => !prev);
+  const isNarrow = useMemo(() => {
+    return screenWidth === "xs" || screenWidth === "sm" || screenWidth === "md";
+  }, [screenWidth]);
+
+  return (
+    <ThemeContext.Provider
+      value={{
+        isDarkMode,
+        toggleTheme,
+        screenWidth,
+        isNarrow,
+        appTitle,
+        setAppTitle,
+        setConnectionState,
+      }}
+    >
+      {children}
+    </ThemeContext.Provider>
+  );
+}
+
+export function useTheme(): ThemeContextType {
+  const context = useContext(ThemeContext);
+  if (context === undefined) {
+    throw new Error("useTheme must be used within a ThemeProvider");
+  }
+  return context;
+}

ui/src/hooks/usePersistentState.ts

@@ -0,0 +1,39 @@
+import { useState, useEffect, useCallback } from "react";
+
+export function usePersistentState<T>(key: string, initialValue: T): [T, (value: T | ((prevState: T) => T)) => void] {
+  const [state, setState] = useState<T>(() => {
+    if (typeof window === "undefined") return initialValue;
+    try {
+      const saved = localStorage.getItem(key);
+      return saved !== null ? JSON.parse(saved) : initialValue;
+    } catch (e) {
+      console.error(`Error parsing stored value for ${key}`, e);
+      return initialValue;
+    }
+  });
+
+  const setPersistentState = useCallback(
+    (value: T | ((prevState: T) => T)) => {
+      setState((prev) => {
+        const nextValue = typeof value === "function" ? (value as (prevState: T) => T)(prev) : value;
+        try {
+          localStorage.setItem(key, JSON.stringify(nextValue));
+        } catch (e) {
+          console.error(`Error saving value for ${key}`, e);
+        }
+        return nextValue;
+      });
+    },
+    [key]
+  );
+
+  useEffect(() => {
+    try {
+      localStorage.setItem(key, JSON.stringify(state));
+    } catch (e) {
+      console.error(`Error saving value for ${key}`, e);
+    }
+  }, [key, state]);
+
+  return [state, setPersistentState];
+}

ui/src/index.css

@@ -0,0 +1,176 @@
+@import "tailwindcss";
+@custom-variant dark (&:where([data-theme=dark], [data-theme=dark] *));
+
+@theme {
+  --color-background: rgba(252, 252, 249, 1);
+  --color-surface: rgba(255, 255, 253, 1);
+
+  /* text colors */
+  --color-txtmain: rgba(19, 52, 59, 1);
+  --color-txtsecondary: rgba(98, 108, 113, 1);
+  --color-navlink-active: rgba(245, 245, 245, 1);
+
+  --color-primary: rgba(50, 184, 198, 1);
+
+  --color-primary-hover: rgba(29, 116, 128, 1);
+  --color-primary-active: rgba(26, 104, 115, 1);
+  --color-secondary: rgba(94, 82, 64, 0.12);
+  --color-secondary-hover: rgba(94, 82, 64, 0.2);
+  --color-secondary-active: rgba(94, 82, 64, 0.25);
+  --color-border: rgba(94, 82, 64, 0.3);
+  --color-btn-primary-text: rgba(252, 252, 249, 1);
+  --color-card-border: rgba(94, 82, 64, 0.12);
+  --color-card-border-inner: rgba(94, 82, 64, 0.12);
+  --color-error: rgba(192, 21, 47, 1);
+  --color-success: rgba(33, 128, 141, 1);
+  --color-warning: rgb(244, 155, 0);
+  --color-info: rgba(98, 108, 113, 1);
+  --color-focus-ring: rgba(33, 128, 141, 0.4);
+  --color-select-caret: rgba(19, 52, 59, 0.8);
+  --color-btn-border: rgba(94, 82, 64, 0.7);
+}
+
+@layer theme {
+  /* over ride theme for dark mode */
+  [data-theme="dark"] {
+    --color-background: rgba(31, 33, 33, 1);
+    --color-surface: rgba(38, 40, 40, 1);
+    /* text colors */
+    --color-txtmain: rgba(245, 245, 245, 1);
+    --color-txtsecondary: rgba(167, 169, 169, 0.7);
+
+    --color-navlink-active: rgba(245, 245, 245, 1);
+
+    --color-primary: rgba(33, 128, 141, 1);
+    --color-primary-hover: rgba(45, 166, 178, 1);
+    --color-primary-active: rgba(41, 150, 161, 1);
+    --color-secondary: rgba(119, 124, 124, 0.15);
+    --color-secondary-hover: rgba(119, 124, 124, 0.25);
+    --color-secondary-active: rgba(119, 124, 124, 0.3);
+    --color-border: rgba(119, 124, 124, 0.3);
+    --color-error: rgba(255, 84, 89, 1);
+    --color-success: rgba(50, 184, 198, 1);
+    --color-warning: rgb(244, 155, 0);
+    --color-info: rgba(167, 169, 169, 1);
+    --color-focus-ring: rgba(50, 184, 198, 0.4);
+    --color-btn-primary-text: rgba(19, 52, 59, 1);
+    --color-card-border: rgba(119, 124, 124, 0.2);
+    --color-card-border-inner: rgba(119, 124, 124, 0.15);
+    --shadow-inset-sm: inset 0 1px 0 rgba(255, 255, 255, 0.1), inset 0 -1px 0 rgba(0, 0, 0, 0.15);
+    --button-border-secondary: rgba(119, 124, 124, 0.2);
+  }
+}
+
+@layer base {
+  body {
+    /* example of how colors using theme colors*/
+    @apply bg-background text-txtmain;
+  }
+
+  h1 {
+    @apply text-4xl text-txtmain font-bold pb-4;
+  }
+  h2 {
+    @apply text-3xl text-txtmain font-bold pb-4;
+  }
+  h3 {
+    @apply text-2xl text-txtmain font-bold pb-4;
+  }
+  h4 {
+    @apply text-xl text-txtmain font-bold pb-4;
+  }
+  h5 {
+    @apply text-lg text-txtmain font-bold pb-4;
+  }
+  h6 {
+    @apply text-base text-txtmain font-bold pb-4;
+  }
+}
+
+/* define CSS classes here for specific types of components */
+@layer components {
+  .container {
+    @apply px-4;
+  }
+
+  /* Tables */
+  table th {
+    @apply p-2 font-semibold;
+  }
+  table td {
+    @apply p-2;
+  }
+
+  /* Navigation Header */
+
+  .navlink {
+    @apply text-txtsecondary hover:bg-secondary hover:text-txtmain rounded-lg p-2;
+  }
+  .navlink.active {
+    @apply bg-primary text-navlink-active;
+  }
+
+  /* Card component */
+  .card {
+    @apply bg-surface rounded-lg border border-card-border shadow-sm overflow-hidden p-4;
+  }
+
+  .card:hover {
+    @apply shadow-md;
+  }
+
+  .card__body {
+    @apply p-4;
+  }
+
+  .card__header,
+  .card__footer {
+    @apply p-4 border-b border-card-border-inner;
+  }
+
+  /* Status Badges */
+  .status {
+    @apply inline-block px-2 py-1 text-xs font-medium rounded-lg;
+  }
+
+  .status--ready {
+    @apply bg-success/10 text-success;
+  }
+
+  .status--starting,
+  .status--stopping {
+    @apply bg-warning/10 text-warning;
+  }
+
+  .status--stopped {
+    @apply bg-error/10 text-error;
+  }
+
+  /* Buttons */
+  .btn {
+    @apply bg-surface py-2 px-4 text-sm rounded-md border transition-colors duration-200 border-btn-border;
+  }
+
+  .btn:hover {
+    cursor: pointer;
+  }
+
+  .btn--sm {
+    @apply px-2 py-0.5 text-xs;
+  }
+
+  .btn:disabled {
+    @apply opacity-50 cursor-not-allowed;
+  }
+}
+
+@layer utilities {
+  .ml-2 {
+    margin-left: 0.5rem;
+  }
+
+  .my-8 {
+    margin-top: 2rem;
+    margin-bottom: 2rem;
+  }
+}

ui/src/lib/types.ts

@@ -0,0 +1 @@
+export type ConnectionState = "connected" | "connecting" | "disconnected";

ui/src/main.tsx

@@ -0,0 +1,16 @@
+import { StrictMode } from "react";
+import { createRoot } from "react-dom/client";
+import "./index.css";
+import App from "./App.tsx";
+import { ThemeProvider } from "./contexts/ThemeProvider";
+import { APIProvider } from "./contexts/APIProvider";
+
+createRoot(document.getElementById("root")!).render(
+  <StrictMode>
+    <ThemeProvider>
+      <APIProvider>
+        <App />
+      </APIProvider>
+    </ThemeProvider>
+  </StrictMode>
+);

ui/src/pages/Activity.tsx

@@ -0,0 +1,120 @@
+import { useMemo } from "react";
+import { useAPI } from "../contexts/APIProvider";
+
+const formatSpeed = (speed: number): string => {
+  return speed < 0 ? "unknown" : speed.toFixed(2) + " t/s";
+};
+
+const formatDuration = (ms: number): string => {
+  return (ms / 1000).toFixed(2) + "s";
+};
+
+const formatRelativeTime = (timestamp: string): string => {
+  const now = new Date();
+  const date = new Date(timestamp);
+  const diffInSeconds = Math.floor((now.getTime() - date.getTime()) / 1000);
+
+  // Handle future dates by returning "just now"
+  if (diffInSeconds < 5) {
+    return "now";
+  }
+
+  if (diffInSeconds < 60) {
+    return `${diffInSeconds}s ago`;
+  }
+
+  const diffInMinutes = Math.floor(diffInSeconds / 60);
+  if (diffInMinutes < 60) {
+    return `${diffInMinutes}m ago`;
+  }
+
+  const diffInHours = Math.floor(diffInMinutes / 60);
+  if (diffInHours < 24) {
+    return `${diffInHours}h ago`;
+  }
+
+  return "a while ago";
+};
+
+const ActivityPage = () => {
+  const { metrics } = useAPI();
+  const sortedMetrics = useMemo(() => {
+    return [...metrics].sort((a, b) => b.id - a.id);
+  }, [metrics]);
+
+  return (
+    <div className="p-2">
+      <h1 className="text-2xl font-bold">Activity</h1>
+
+      {metrics.length === 0 && (
+        <div className="text-center py-8">
+          <p className="text-gray-600">No metrics data available</p>
+        </div>
+      )}
+      {metrics.length > 0 && (
+        <div className="card overflow-auto">
+          <table className="min-w-full divide-y">
+            <thead className="border-gray-200 dark:border-white/10">
+              <tr className="text-left text-xs uppercase tracking-wider">
+                <th className="px-6 py-3">ID</th>
+                <th className="px-6 py-3">Time</th>
+                <th className="px-6 py-3">Model</th>
+                <th className="px-6 py-3">
+                  Cached <Tooltip content="prompt tokens from cache" />
+                </th>
+                <th className="px-6 py-3">
+                  Prompt <Tooltip content="new prompt tokens processed" />
+                </th>
+                <th className="px-6 py-3">Generated</th>
+                <th className="px-6 py-3">Prompt Processing</th>
+                <th className="px-6 py-3">Generation Speed</th>
+                <th className="px-6 py-3">Duration</th>
+              </tr>
+            </thead>
+            <tbody className="divide-y">
+              {sortedMetrics.map((metric) => (
+                <tr key={metric.id} className="whitespace-nowrap text-sm border-gray-200 dark:border-white/10">
+                  <td className="px-4 py-4">{metric.id + 1 /* un-zero index */}</td>
+                  <td className="px-6 py-4">{formatRelativeTime(metric.timestamp)}</td>
+                  <td className="px-6 py-4">{metric.model}</td>
+                  <td className="px-6 py-4">{metric.cache_tokens > 0 ? metric.cache_tokens.toLocaleString() : "-"}</td>
+                  <td className="px-6 py-4">{metric.input_tokens.toLocaleString()}</td>
+                  <td className="px-6 py-4">{metric.output_tokens.toLocaleString()}</td>
+                  <td className="px-6 py-4">{formatSpeed(metric.prompt_per_second)}</td>
+                  <td className="px-6 py-4">{formatSpeed(metric.tokens_per_second)}</td>
+                  <td className="px-6 py-4">{formatDuration(metric.duration_ms)}</td>
+                </tr>
+              ))}
+            </tbody>
+          </table>
+        </div>
+      )}
+    </div>
+  );
+};
+
+interface TooltipProps {
+  content: string;
+}
+
+const Tooltip: React.FC<TooltipProps> = ({ content }) => {
+  return (
+    <div className="relative group inline-block">
+      ⓘ
+      <div
+        className="absolute top-full left-1/2 transform -translate-x-1/2 mt-2
+                     px-3 py-2 bg-gray-900 text-white text-sm rounded-md
+                     opacity-0 group-hover:opacity-100 transition-opacity
+                     duration-200 pointer-events-none whitespace-nowrap z-50 normal-case"
+      >
+        {content}
+        <div
+          className="absolute bottom-full left-1/2 transform -translate-x-1/2
+                       border-4 border-transparent border-b-gray-900"
+        ></div>
+      </div>
+    </div>
+  );
+};
+
+export default ActivityPage;