#!/usr/bin/env bash
# Pull the recommended model roster on the Mac.
# Run this ON the Mac where Ollama is installed.
set -euo pipefail

OLLAMA_HOST="${OLLAMA_HOST:-http://localhost:11434}"

echo "=== Pulling models to ${OLLAMA_HOST} ==="

# Embedder (always resident, slot 1)
echo "--- Embedder: nomic-embed-text ---"
curl -s "${OLLAMA_HOST}/api/pull" -d '{"name":"nomic-embed-text"}' | jq -r '.status // empty'

# Worker models (rotate through slot 2)
echo "--- Worker: qwen3:14b (parse/data) ---"
curl -s "${OLLAMA_HOST}/api/pull" -d '{"name":"qwen3:14b"}' | jq -r '.status // empty'

echo "--- Worker: qwen3:30b (agent+code, default) ---"
curl -s "${OLLAMA_HOST}/api/pull" -d '{"name":"qwen3:30b"}' | jq -r '.status // empty'

# Optional — uncomment if needed:
# echo "--- Worker: gpt-oss:20b (fast coder) ---"
# curl -s "${OLLAMA_HOST}/api/pull" -d '{"name":"gpt-oss:20b"}' | jq -r '.status // empty'
# echo "--- Worker: qwen2.5-coder:32b (quality coder, slow) ---"
# curl -s "${OLLAMA_HOST}/api/pull" -d '{"name":"qwen2.5-coder:32b"}' | jq -r '.status // empty'

echo ""
echo "=== Mac-side Ollama environment (set via launchctl or .zshrc) ==="
echo "  OLLAMA_MAX_LOADED_MODELS=2"
echo "  OLLAMA_KEEP_ALIVE=-1        # for the embedder slot"
echo "  OLLAMA_CONTEXT_LENGTH=8192  # minimum recommended"
echo ""
echo "  Example: launchctl setenv OLLAMA_MAX_LOADED_MODELS 2"
echo "  Then restart Ollama."