Address Gadfly findings on #85

- PublicGardenPage had the same 100vh mobile bug I fixed in the editor — 100dvh
  there too, so the fix is consistent across both full-height pages.
- Cap the toast stack at 4 (drop oldest). Now that error toasts don't auto-
  dismiss, a burst of failures could otherwise grow the stack unbounded and push
  the newest — the one that just happened — off-screen.
- Hoist the From/To date-input styling to a shared dateInputClass const so the
  two don't drift.

Not taken: reusing safeRedirectPath for the 401 redirect (it validates an
incoming redirect param, it doesn't build the outgoing one — the two uses don't
overlap), and refactoring the three-branch filter spread (it's clear at three
fields; a helper earns its keep when there are more).

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>
Claude-Session: https://claude.ai/code/session_01H3zbym8Doka2d7D48maSgZ

CLAUDE.md

@@ -85,6 +85,12 @@ Frontend: React 19 + Vite + Tailwind 4 + TanStack Router/Query, built into
   deliberately. `ErrForbidden` means "you can see it but may not do that".
 - **Plops (plantings) live in their parent object's local frame**, origin at the
   object's center, `-y` is north. Moving or rotating a bed moves its plants free.
+- **A plop is a clump, not a plant.** `defaultPlopRadius` is `1.5 × spacing`, so a
+  plop is three spacings across and holds `π·r²/spacing²` plants. Reasoning about
+  fills as if one plop were one plant gets the geometry wrong every time — which
+  is how #75 happened: requiring the whole circle inside the bed inset the outer
+  row by 1.5 spacings when the horticultural rule is *half* a spacing. Spacing is
+  a constraint between neighbouring plants; a bed edge is nobody's neighbour.
 - **Soft removal**: "clear bed" sets `removed_at`; the editor reads
   `removed_at IS NULL`. Hard delete is a different operation.
 - **Migrations** are numbered `.sql` files in `internal/store/migrations/`, run
@@ -92,6 +98,29 @@ Frontend: React 19 + Vite + Tailwind 4 + TanStack Router/Query, built into
 - **Every service mutation lands in history** (#48). If you add one, record it —
   see `internal/service/revisions.go`. Multi-row operations pass all their
   changes to a single `record` call so they undo as one unit.
+- **The history write is detached from cancellation on purpose.** `commitScope`
+  calls `context.WithoutCancel` — that is not a mistake to tidy up. By the time
+  a commit runs, the rows it describes are already written, so cancelling it
+  cannot undo anything; it can only leave real changes with no way to undo them.
+  This was a live bug twice (#73): a client disconnect mid-request orphaned 18
+  plantings. Fixing it per-call-site is how it came back, which is why the rule
+  lives in `commitScope` where no caller can forget it.
+
+## Testing
+
+Match the test to the failure it would catch:
+
+- **Anything addressed by its own id needs an API-level test through the router.**
+  Service tests can't see a route that was never registered — PATCH/DELETE
+  `/journal/:id` once shipped fully implemented, fully unit-tested, and
+  completely unreachable.
+- **Watch for fixtures that assert your assumptions instead of the API.** A test
+  for the undo message passed because the fixture I wrote populated a field the
+  real response leaves empty. If a test builds the thing it's testing against,
+  it is checking your mental model, not the system.
+- Some things only real use finds. The agent's whole loop is covered by
+  majordomo's scriptable fake provider (`provider/fake`), which is worth using —
+  but the three worst v2 bugs all turned up in one live session afterwards.
 
 ## Workflow
 
@@ -100,6 +129,20 @@ fix what's real → merge when the pipeline is green. Do not grade Gadfly findin
 A push to `main` builds the image and deploys to Komodo; the live instance at
 `pansy.orgrimmar.dudenhoeffer.casa` updates a few minutes later.
 
+**One sweep, not a loop.** Take Gadfly's *initial* review, fix what's real, and
+merge once the build is green. Do NOT re-trigger Gadfly on the fix commits and
+wait for it again — a re-review-per-fix loop burns ~10 min a pass and drags a
+small PR out for an hour (Steve's explicit call, 2026-07-22). The initial review
+is the check; your own build/test/judgment covers the fixes. Gadfly is advisory
+and never blocks merge, so green build + fixes applied is enough.
+
+(Mechanics, if you ever *do* need a manual re-run: the workflow triggers on
+`opened`/`reopened`/`ready_for_review`, not `synchronize`, so pushes don't
+re-review; a `@gadfly review` comment does. A comment without that exact phrase
+still runs and exits green in ~2s — a skip that looks like a pass, so judge a
+real review by its ~10-min duration, not its status. Gadfly edits its consensus
+comment in place, so `updated_at` moves but `created_at` doesn't.)
+
 Workflow- and config-only changes (CI, this file, docs) go straight to `main`
 without the PR dance.
 
@@ -113,7 +156,35 @@ implemented in later per-issue sessions.
 the `_PATH`/`_ADDR` names you might guess. Authentik is the primary IdP;
 OIDC-first with local passwords as fallback.
 
-Agent config (v2): `OLLAMA_CLOUD_API_KEY` and `PANSY_AGENT_MODEL` (default
+Agent config: `OLLAMA_CLOUD_API_KEY`, `PANSY_AGENT_MODEL` (default
-`ollama-cloud/glm-5.2:cloud`), set in Komodo. Model strings pass verbatim to
+`ollama-cloud/glm-5.2:cloud`) and `PANSY_AGENT_ENABLED`, set in Komodo. Model
-`majordomo.Parse`, so a comma-separated spec gives failover for free.
+strings pass verbatim to `majordomo.Parse`, so a comma-separated spec gives
-`majordomo` and `executus` are sibling repos at `../`.
+failover for free — don't parse that grammar in pansy.
+
+The model and enabled flag are also **admin-editable at runtime** in Settings
+(#79); the env vars are just defaults (precedence: DB setting → env → default).
+Two things this makes load-bearing:
+
+- **The live Runner is hot-swapped, not built once.** It sits behind an
+  `atomic.Pointer` in `internal/api` (`agentHolder`), and its routes are
+  registered *unconditionally* with a nil-check on `agent.get()`. Do NOT go back
+  to registering the chat routes only when a key is present — a settings change
+  has to be able to turn the assistant on without a restart, which a missing
+  route can't. `/capabilities` reads the pointer, so it reflects the live state.
+- **`OLLAMA_CLOUD_API_KEY` stays in the environment, never the DB.** Model
+  selection is a setting; the key is not. A secret in `instance_settings` lands
+  in every backup and in the undo history's blast radius. The Settings API
+  reports whether a key is *present*, never its value.
+- The "how to turn a spec into a model" knowledge lives once in
+  `internal/agentmodel`, imported by both `agent` (to run) and `service` (to
+  validate a spec before storing it). It can't live in `agent` — `agent` imports
+  `service`, so a `service`→`agent` import would cycle.
+
+`majordomo` is a **real dependency** now, resolved from the Gitea instance as a
+pseudo-version. There is no `replace` directive and there must not be one: a
+`replace` pointing at `../majordomo` builds on your laptop and breaks the Docker
+build, which has no sibling checkout. `executus` is a sibling repo at `../` and
+is not a dependency.
+
+The `majordomo` build tag is **gone**. Don't reintroduce it — an untagged CI that
+never compiles the agent is worse than a slightly larger binary.

DESIGN.md

@@ -7,9 +7,11 @@ Work is tracked in Gitea issues; the tracking epic links every piece in dependen
 ## Decisions
 
 - **Placement model:** freeform plops (not a square-foot grid), scaled by real plant spacing. Grid snapping may come later as a toggle.
+- **Spacing is a plant-to-plant rule, so bed edges get half of it.** A bed edge is not a competitor for soil, light or water, so the outer row owes it half the spacing rather than a full one. `FillRegion` centres its lattice accordingly, and lets a plop — a *clump* three spacings across — cross the edge by up to half a spacing so its outermost plants land at that half-spacing. The rule, the square-foot-chart arithmetic behind it, and the failure mode it prevents are written out once in `hexCenters`; #75 is what getting it wrong looked like.
 - **Stack:** Go 1.26.x backend, module `gitea.stevedudenhoeffer.com/steve/pansy`; React + TypeScript + Vite + Tailwind frontend, production build embedded via `embed.FS` → one static binary (`CGO_ENABLED=0`).
-- **Users:** multi-user with ownership. Users own gardens; a garden can be shared with other users as viewer (read) or editor (edit content). Owner additionally shares/deletes.
+- **Users:** multi-user with ownership. Users own gardens; a garden can be shared with other users as viewer (read) or editor (edit content). Owner additionally shares/deletes. The first registered user is `is_admin` (set race-free inside the INSERT); admin gates instance-wide Settings (`requireAdmin`), the only thing that reads that flag.
 - **Auth:** OIDC-first (Authentik is the primary IdP), local argon2id passwords as an optional fallback.
+- **Instance settings (#79):** admin-editable, instance-wide config in a single-row `instance_settings` table — pansy's first DB-stored *instance* state (everything else hangs off a garden/object). Today it holds the agent model + on/off; **secrets never move here** — `OLLAMA_CLOUD_API_KEY` stays in the env so it doesn't land in backups or the undo history. Precedence: Settings value → env → default. The live agent Runner sits behind an `atomic.Pointer` in the API layer (`agentHolder`) with its routes always registered, so a settings change swaps it with no restart and no race against in-flight requests; `/capabilities` reads that pointer, so it reports what's live rather than what was configured at boot. The model/registry knowledge lives in one leaf package (`internal/agentmodel`) that both the runner and the settings validator import — agent imports service, so it can live in neither.
 - **Agentic future:** integration with majordomo/executus via typed Go tools (`llm.DefineTool[Args]`) wrapping the same service layer the REST API uses — not MCP/OpenAPI.
 
 ## Domain model
@@ -23,6 +25,7 @@ SQLite, centimeters everywhere (display-side imperial conversion only), `version
 - **garden_objects** — one polymorphic table: `kind` (`bed`|`grow_bag`|`container`|`in_ground`|`tree`|`path`|`structure`), name, `shape` (`rect`|`circle`; `polygon` + `points` JSON reserved in schema, not in the v1 editor), center `x_cm`,`y_cm` in garden space, `width_cm`,`height_cm` (circle: width=diameter), `rotation_deg`, `z_index`, `plantable` bool, nullable `color` override, `props` JSON for kind-specific extras (bed height, tree species…), notes.
 - **seed_lots** — one purchase: vendor, source URL, sku, lot code, purchase date, packed-for year, quantity + unit, cost, germination %. `owner_id` is on the LOT, not inherited from the plant, so you can record buying generic built-in Garlic from Johnny's and keep it private. `plantings.seed_lot_id` (ON DELETE SET NULL) attributes a plop to a purchase; **`remaining` is derived** (quantity − summed effective count of active plantings), never stored — a decremented column drifts the moment a planting is edited behind its back and can't be audited afterwards. Lots are never shared along with a garden.
 - **plants** — catalog, plus `source_url`/`vendor` provenance for the variety itself. `owner_id NULL` = built-in (~30 seeded via migration, read-only, clone to customize). name, category (`vegetable`|`herb`|`flower`|`fruit`|`tree_shrub`|`cover`), `spacing_cm` (mature spacing), `color` hex, `icon` (emoji string — zero assets in v1), nullable `days_to_maturity`, notes. Users see built-ins + their own.
+- **agent_conversations** / **agent_messages** — the assistant's chat, one thread per (user, garden). Only the user/assistant TEXT of each turn is stored, not the model's full transcript: continuity needs what was said and what came back, and replaying a stored tool call would replay a decision made against a garden that has since moved on. `agent_messages.change_set_id` is the handle the chat panel uses to offer Undo on the turn that caused it.
 - **journal_entries** — the grow log: `garden_id` (NOT NULL), nullable `object_id`/`planting_id` to narrow the target, author, body, `observed_at`. `notes` on a garden/object/plant says what the thing IS and a new note overwrites the old; a journal entry says what HAPPENED and when, and entries accumulate. `garden_id` is set even for a bed- or plop-level entry so permission checks reuse `requireGardenRole` unchanged rather than inventing a second ACL path. `observed_at` is distinct from `created_at` — you write up Saturday's observations on Sunday. Deliberately **not** in the revision history: entries are already append-shaped and individually versioned.
 - **change_sets** / **revisions** — the undo substrate. A change set groups the row-level revisions one logical operation produced (`source` ui/agent/api, `summary`, nullable `agent_run_id`, nullable `reverts_id`); a revision is `(entity_type, entity_id, op, before, after)` with full JSON row snapshots. The unit of undo is the **operation, not the row**. Revert is itself a change set, so an undo can be undone; it is version-guarded per entity and reports conflicts rather than clobbering a later edit. Garden *deletion* is deliberately not revertible (the cascade is invisible to the service, and would take the history with it).
 - **plantings** ("plops") — object_id, plant_id, center `x_cm`,`y_cm` **in the parent object's local frame** (moving/rotating a bed moves its plants for free), `radius_cm` (a plop is a circular patch), nullable `count` (NULL = derived: `max(1, round(π·r² / spacing_cm²))`), nullable label, `planted_at`/`removed_at` dates. "Clear bed" sets `removed_at`; v1 shows rows where `removed_at IS NULL`. Year-over-year history and a future plan/scenario layer build on these dates without schema surgery.
@@ -62,9 +65,19 @@ POST /change-sets/:id/revert  ← undo an operation; 201, or 409 + the conflicts
 POST /gardens/:id/copy        ← deep-copy a garden you own (objects + active plops; not shares/link)
 POST /gardens/:id/objects     PATCH,DELETE /objects/:id
 POST /objects/:id/plantings   PATCH,DELETE /plantings/:id
+POST /objects/:id/fill        ← hex-pack a region with one plant; region by compass name or rect
+POST /objects/:id/clear       ← soft-remove every active plop, as ONE change set
 GET,POST /plants              PATCH,DELETE /plants/:id            (own plants only)
 GET,POST /seed-lots           GET,PATCH,DELETE /seed-lots/:id     (own lots only; private)
+GET,POST /gardens/:id/journal PATCH,DELETE /journal/:id           (editor writes; author edits own)
+GET  /gardens/:id/journal/counts  ← entries per object, for the "has notes" indicator
+POST /agent/chat              ← SSE: step events, then the finished turn (editor only)
+GET,DELETE /gardens/:id/agent/history                             (the actor's own thread)
+GET  /capabilities            ← what this instance can do RIGHT NOW (tracks the live agent, not just config)
+GET,PATCH /settings           ← instance-wide config (admin only): agent model + on/off
 GET,POST /gardens/:id/shares  PATCH,DELETE /gardens/:id/shares/:userId    (invite by email)
+GET,POST,DELETE /gardens/:id/share-link   ← the public read-only token for this garden
+GET  /public/gardens/:token   ← UNAUTHENTICATED read-only /full; the token is the capability
 ```
 
 **Sync:** plain REST + optimistic UI + last-write-wins with a version guard. Every PATCH/DELETE carries the row's `version`; the server increments on write and returns **409 + the current row** on mismatch; the client rolls back and refetches. No websockets/CRDT — the right cost for household-scale co-editing. Drags PATCH once on drop, not per frame.
@@ -77,6 +90,7 @@ GET,POST /gardens/:id/shares  PATCH,DELETE /gardens/:id/shares/:userId    (invit
 cmd/pansy/main.go        config load → store.Open → migrate → service → api → ListenAndServe
 internal/config/         PANSY_PORT, PANSY_DB, PANSY_BASE_URL, PANSY_REGISTRATION,
                          PANSY_OIDC_* (issuer/client_id/secret/button label), PANSY_LOCAL_AUTH,
+                         PANSY_AGENT_MODEL / OLLAMA_CLOUD_API_KEY / PANSY_AGENT_ENABLED,
                          trusted proxies
 internal/store/          sqlite.go (open/pragmas), migrate.go, migrations/*.sql (embedded),
                          queries per entity (users.go, gardens.go, objects.go, plantings.go,
@@ -89,7 +103,8 @@ internal/service/        ★ THE SEAM. All permission checks + business ops. aut
 internal/api/            thin gin handlers (decode → service → encode), session middleware,
                          ginserver setup, routes.go, oidc.go, spa.go (embed fallback)
 internal/webdist/        dist.go: //go:embed all:dist   (web build copied in by Makefile)
-internal/agent/          later: llm.DefineTool[Args] wrappers over the SAME service methods
+internal/agent/          tools.go (llm.DefineTool wrappers over the SAME service methods),
+                         runtime.go (model resolution, run loop, one change set per turn)
 web/                     Vite app
 Makefile                 (cd web && npm run build) → copy dist → CGO_ENABLED=0 go build
 ```
@@ -111,7 +126,7 @@ React 19 + TypeScript + Vite + Tailwind 4 (`@tailwindcss/vite`), `@tanstack/reac
 - **Routes:** `/login`, `/register`, `/gardens` (list), `/gardens/:id` (editor, `?focus=objectId`), `/plants` (catalog). Auth guard on the router root via `/auth/me`.
 - **State:** TanStack Query for all server state (editor keyed on `gardens/:id/full`; optimistic mutations with version-conflict rollback). One small Zustand store for ephemeral editor state only: viewport, selection, focused object, active tool, in-flight drag.
 - **Editor components (`web/src/editor/`):** `GardenCanvas` (svg root + viewport g), `useViewport` (use-gesture pan/zoom/pinch), `ObjectShape`, `PlopMarker` (semantic-zoom branching), `Palette` (drag-to-place object kinds), `EditorRail` (the one side panel), `Inspector`, `HistoryPanel`, `PlantPicker`.
-- **One rail, tabs inside it.** The inspector, history, journal, and later the chat panel all want the same strip of screen; rather than each bolting on its own chrome they are tabs in `EditorRail` — so the canvas is one width instead of a different width per panel, and adding a panel is adding a tab. Selecting an object switches to the Inspector tab automatically, so the rail is never something you operate before you can edit; on a phone the same tabs render in the bottom sheet the inspector already used. Pure geometry helpers (local↔world transforms, unit formatting) in `web/src/lib/geometry.ts`, unit-tested.
+- **One rail, tabs inside it.** The inspector, history, journal and assistant all want the same strip of screen; rather than each bolting on its own chrome they are tabs in `EditorRail` — so the canvas is one width instead of a different width per panel, and adding a panel is adding a tab. Selecting an object switches to the Inspector tab automatically, so the rail is never something you operate before you can edit; on a phone the same tabs render in the bottom sheet the inspector already used. Pure geometry helpers (local↔world transforms, unit formatting) in `web/src/lib/geometry.ts`, unit-tested.
 
 ## Roadmap
 
@@ -123,7 +138,8 @@ React 19 + TypeScript + Vite + Tailwind 4 (`@tailwindcss/vite`), `@tanstack/reac
 6. **Plops** — focus-zoom, place/move/resize, derived counts, semantic zoom. *The core vision lands here.*
 7. **Sharing** — invite by email, roles, viewer read-only mode.
 8. **Polish** — imperial toggle, mobile ergonomics, clear-bed, keyboard nudging.
-9. **Agent seam** — `ops.go` bulk ops + `internal/agent` DefineTool wrappers; optional mort-style API-key agent endpoint. The agent harness itself lives in majordomo/executus, outside pansy.
+9. **Agent seam** — `ops.go` bulk ops + `internal/agent` DefineTool wrappers.
+10. **Garden assistant** — majordomo in-process, Ollama Cloud, streaming chat. Each turn runs inside ONE change set (`source='agent'`), so a turn that clears a bed and replants it undoes as one action; that is what makes acting without a confirmation prompt defensible. Bounded by a step cap and a timeout — loop safety, not spend control. The `majordomo` build tag is gone: a tag that keeps the agent out of the binary only earns its keep if you'd ship a build without it, and the agent is the point.
 
 ## Deliberate v1 limits
 

README.md

@@ -63,12 +63,19 @@ All configuration is via environment variables; every value has a default, so `.
 | `PANSY_OIDC_BUTTON_LABEL` | `Sign in with Authentik` | Label for the OIDC button on the login page.                |
 | `PANSY_TRUSTED_PROXIES`   | *(none)*           | Comma-separated proxy CIDRs/IPs to trust for client-IP resolution. |
 
-The garden assistant reads two more. Both are **read only once the assistant lands** ([#56](https://gitea.stevedudenhoeffer.com/steve/pansy/issues/56)); setting them earlier is harmless and setting neither leaves the assistant off.
+The garden assistant reads three more. Setting none of them leaves the assistant off; the app runs exactly as it does without it.
 
 | Variable                | Default                        | Description                                                                 |
-| ----------------------- | ----------------------------- | --------------------------------------------------------------------------- |
+| ----------------------- | ------------------------------ | --------------------------------------------------------------------------- |
-| `OLLAMA_CLOUD_API_KEY`  | *(empty)*                     | Ollama Cloud API key. Without it the assistant is disabled, not broken.      |
+| `OLLAMA_CLOUD_API_KEY`  | *(empty)*                      | Ollama Cloud API key. Without it the assistant is off, not broken. This is the one agent value that stays in the environment — it is **never** stored in the database or editable in Settings. |
-| `PANSY_AGENT_MODEL`     | `ollama-cloud/glm-5.2:cloud`  | Model spec, passed verbatim to `majordomo.Parse` — a comma-separated list is a failover chain. |
+| `PANSY_AGENT_MODEL`     | `ollama-cloud/glm-5.2:cloud`   | Default model spec, passed verbatim to `majordomo.Parse` — a comma-separated list is a failover chain, e.g. `ollama-cloud/glm-5.2:cloud,ollama-cloud/kimi-k2.6:cloud`. An admin can override this per-instance in **Settings** without a redeploy; a blank Settings value inherits this. |
+| `PANSY_AGENT_ENABLED`   | on when a key is present       | Default on/off for the assistant. Also overridable in Settings (which can inherit this default). |
+
+The model and enabled flag can be changed at runtime by an admin under **Settings** (the gear appears in the nav for admins) — the change swaps the live assistant with no restart. The env vars above are the defaults an untouched instance uses, and the API key is intentionally not among the runtime-editable settings: a secret in the database would land in every backup. Precedence for the model and enabled flag is **Settings value, if set → env var → built-in default**.
+
+The assistant acts without asking first, which is only reasonable because every turn is one undoable change set — see the History panel in the editor.
+
+**If you set the key and the assistant still doesn't appear**, check that the variable reaches the *container*, not just your orchestrator's stack config — Compose needs it listed under the service's `environment:`. pansy logs why the assistant is off at startup, and Settings shows the same status (a key present, the resolved model, and whether it's actually running).
 
 Local email/password auth is live (`POST /api/v1/auth/register`, `/auth/login`, `/auth/logout`, `GET /auth/me`, `GET /auth/providers`); the session is an HttpOnly cookie (`Secure` when `PANSY_BASE_URL` is https). The first account registered becomes admin, and it may register even when `PANSY_REGISTRATION=closed` to bootstrap the instance.
 
@@ -108,7 +115,8 @@ services:
       # PANSY_OIDC_ISSUER: https://auth.example.com/application/o/pansy/
       # PANSY_OIDC_CLIENT_ID: ...
       # PANSY_OIDC_CLIENT_SECRET: ...
-      # OLLAMA_CLOUD_API_KEY: ...     # enables the garden assistant
+      # OLLAMA_CLOUD_API_KEY: ${OLLAMA_CLOUD_API_KEY}   # enables the garden assistant
+      # PANSY_AGENT_MODEL: ollama-cloud/glm-5.2:cloud
     restart: unless-stopped
 volumes:
   pansy-data:

go.mod

@@ -3,20 +3,39 @@ module gitea.stevedudenhoeffer.com/steve/pansy
 go 1.26.2
 
 require (
+	gitea.stevedudenhoeffer.com/steve/majordomo v0.0.0-20260718232210-a941f5ff4a3f
 	github.com/coreos/go-oidc/v3 v3.20.0
+	github.com/gen2brain/heic v0.7.1
 	github.com/gin-gonic/gin v1.10.1
 	github.com/samber/slog-gin v1.15.0
-	golang.org/x/crypto v0.31.0
+	golang.org/x/crypto v0.36.0
+	golang.org/x/image v0.44.0
 	golang.org/x/oauth2 v0.36.0
 	modernc.org/sqlite v1.34.4
 )
 
+require (
+	cloud.google.com/go v0.116.0 // indirect
+	cloud.google.com/go/auth v0.9.3 // indirect
+	cloud.google.com/go/compute/metadata v0.5.0 // indirect
+	github.com/golang/groupcache v0.0.0-20210331224755-41bb18bfe9da // indirect
+	github.com/google/go-cmp v0.6.0 // indirect
+	github.com/google/s2a-go v0.1.8 // indirect
+	github.com/googleapis/enterprise-certificate-proxy v0.3.4 // indirect
+	github.com/gorilla/websocket v1.5.3 // indirect
+	go.opencensus.io v0.24.0 // indirect
+	google.golang.org/genai v1.59.0 // indirect
+	google.golang.org/genproto/googleapis/rpc v0.0.0-20240903143218-8af14fe29dc1 // indirect
+	google.golang.org/grpc v1.66.2 // indirect
+)
+
 require (
 	github.com/bytedance/sonic v1.11.9 // 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/dustin/go-humanize v1.0.1 // indirect
+	github.com/ebitengine/purego v0.10.1 // indirect
 	github.com/gabriel-vasile/mimetype v1.4.4 // indirect
 	github.com/gin-contrib/sse v0.1.0 // indirect
 	github.com/go-jose/go-jose/v4 v4.1.4 // indirect
@@ -35,14 +54,15 @@ require (
 	github.com/ncruces/go-strftime v0.1.9 // indirect
 	github.com/pelletier/go-toml/v2 v2.2.2 // indirect
 	github.com/remyoudompheng/bigfft v0.0.0-20230129092748-24d4a6f8daec // indirect
+	github.com/tetratelabs/wazero v1.12.0 // indirect
 	github.com/twitchyliquid64/golang-asm v0.15.1 // indirect
 	github.com/ugorji/go/codec v1.2.12 // indirect
 	go.opentelemetry.io/otel v1.29.0 // indirect
 	go.opentelemetry.io/otel/trace v1.29.0 // indirect
 	golang.org/x/arch v0.8.0 // indirect
-	golang.org/x/net v0.33.0 // indirect
+	golang.org/x/net v0.38.0 // indirect
-	golang.org/x/sys v0.28.0 // indirect
+	golang.org/x/sys v0.44.0 // indirect
-	golang.org/x/text v0.21.0 // indirect
+	golang.org/x/text v0.40.0 // indirect
 	google.golang.org/protobuf v1.34.2 // indirect
 	gopkg.in/yaml.v3 v3.0.1 // indirect
 	modernc.org/gc/v3 v3.0.0-20240107210532-573471604cb6 // indirect

go.sum

@@ -1,11 +1,24 @@
+cloud.google.com/go v0.26.0/go.mod h1:aQUYkXzVsufM+DwF1aE+0xfcU+56JwCaLick0ClmMTw=
+cloud.google.com/go v0.116.0 h1:B3fRrSDkLRt5qSHWe40ERJvhvnQwdZiHu0bJOpldweE=
+cloud.google.com/go v0.116.0/go.mod h1:cEPSRWPzZEswwdr9BxE6ChEn01dWlTaF05LiC2Xs70U=
+cloud.google.com/go/auth v0.9.3 h1:VOEUIAADkkLtyfr3BLa3R8Ed/j6w1jTBmARx+wb5w5U=
+cloud.google.com/go/auth v0.9.3/go.mod h1:7z6VY+7h3KUdRov5F1i8NDP5ZzWKYmEPO842BgCsmTk=
+cloud.google.com/go/compute/metadata v0.5.0 h1:Zr0eK8JbFv6+Wi4ilXAR8FJ3wyNdpxHKJNPos6LTZOY=
+cloud.google.com/go/compute/metadata v0.5.0/go.mod h1:aHnloV2TPI38yx4s9+wAZhHykWvVCfu7hQbF+9CWoiY=
+gitea.stevedudenhoeffer.com/steve/majordomo v0.0.0-20260718232210-a941f5ff4a3f h1:Zt91pngDr+TIU14sxv3fk1QS7bMQetvccFNg8kPaltA=
+gitea.stevedudenhoeffer.com/steve/majordomo v0.0.0-20260718232210-a941f5ff4a3f/go.mod h1:UZLveG17SmENt4sne2RSLIbioix30RZbRIQUzBAnOyY=
+github.com/BurntSushi/toml v0.3.1/go.mod h1:xHWCNGjB5oqiDr8zfno3MHue2Ht5sIBksp03qcyfWMU=
 github.com/bytedance/sonic v1.11.9 h1:LFHENlIY/SLzDWverzdOvgMztTxcfcF+cqNsz9pK5zg=
 github.com/bytedance/sonic v1.11.9/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/census-instrumentation/opencensus-proto v0.2.1/go.mod h1:f6KPmirojxKA12rnyqOA5BBL4O983OfeGPqjHWSTneU=
+github.com/client9/misspell v0.3.4/go.mod h1:qj6jICC3Q7zFZvVWo7KLAzC3yx5G7kyvSDkc90ppPyw=
 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/cncf/udpa/go v0.0.0-20191209042840-269d4d468f6f/go.mod h1:M8M6+tZqaGXZJjfX53e64911xZQV5JYwmTeXPW+k8Sc=
 github.com/coreos/go-oidc/v3 v3.20.0 h1:EtE0WIBHk03N+DqGkY4+UONzzZHk7amKt6IyNd7OsZE=
 github.com/coreos/go-oidc/v3 v3.20.0/go.mod h1:DYCf24+ncYi+XkIH97GY1+dqoRlbaSI26KVTCI9SrY4=
 github.com/davecgh/go-spew v1.1.0/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
@@ -13,8 +26,16 @@ 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/dustin/go-humanize v1.0.1 h1:GzkhY7T5VNhEkwH0PVJgjz+fX1rhBrR7pRT3mDkpeCY=
 github.com/dustin/go-humanize v1.0.1/go.mod h1:Mu1zIs6XwVuF/gI1OepvI0qD18qycQx+mFykh5fBlto=
+github.com/ebitengine/purego v0.10.1 h1:dewVBCBT2GaMu1SrNTYxQhgQBethzfhiwvZiLGP/qyY=
+github.com/ebitengine/purego v0.10.1/go.mod h1:iIjxzd6CiRiOG0UyXP+V1+jWqUXVjPKLAI0mRfJZTmQ=
+github.com/envoyproxy/go-control-plane v0.9.0/go.mod h1:YTl/9mNaCwkRvm6d1a2C3ymFceY/DCBVvsKhRF0iEA4=
+github.com/envoyproxy/go-control-plane v0.9.1-0.20191026205805-5f8ba28d4473/go.mod h1:YTl/9mNaCwkRvm6d1a2C3ymFceY/DCBVvsKhRF0iEA4=
+github.com/envoyproxy/go-control-plane v0.9.4/go.mod h1:6rpuAdCZL397s3pYoYcLgu1mIlRU8Am5FuJP05cCM98=
+github.com/envoyproxy/protoc-gen-validate v0.1.0/go.mod h1:iSmxcyjqTsJpI2R4NaDN7+kN2VEUnK/pcBlmesArF7c=
 github.com/gabriel-vasile/mimetype v1.4.4 h1:QjV6pZ7/XZ7ryI2KuyeEDE8wnh7fHP9YnQy+R0LnH8I=
 github.com/gabriel-vasile/mimetype v1.4.4/go.mod h1:JwLei5XPtWdGiMFB5Pjle1oEeoSeEuJfJE+TtfvdB/s=
+github.com/gen2brain/heic v0.7.1 h1:Aha1sZdKEeZeWl5o0xkSg7NBRhhkrlokGVCRri+2Qcc=
+github.com/gen2brain/heic v0.7.1/go.mod h1:ja42wMJc4fpnKsfdUJxeZa2YqqRnes1wS0xqs5+8o5w=
 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 v1.10.1 h1:T0ujvqyCSqRopADpgPgiTT63DUQVSfojyME59Ei63pQ=
@@ -31,13 +52,40 @@ github.com/go-playground/validator/v10 v10.22.0 h1:k6HsTZ0sTnROkhS//R0O+55JgM8C4
 github.com/go-playground/validator/v10 v10.22.0/go.mod h1:dbuPbCMFw/DrkbEynArYaCwl3amGuJotoKCe95atGMM=
 github.com/goccy/go-json v0.10.3 h1:KZ5WoDbxAIgm2HNbYckL0se1fHD6rz5j4ywS6ebzDqA=
 github.com/goccy/go-json v0.10.3/go.mod h1:oq7eo15ShAhp70Anwd5lgX2pLfOS3QCiwU/PULtXL6M=
+github.com/golang/glog v0.0.0-20160126235308-23def4e6c14b/go.mod h1:SBH7ygxi8pfUlaOkMMuAQtPIUF8ecWP5IEl/CR7VP2Q=
+github.com/golang/groupcache v0.0.0-20200121045136-8c9f03a8e57e/go.mod h1:cIg4eruTrX1D+g88fzRXU5OdNfaM+9IcxsU14FzY7Hc=
+github.com/golang/groupcache v0.0.0-20210331224755-41bb18bfe9da h1:oI5xCqsCo564l8iNU+DwB5epxmsaqB+rhGL0m5jtYqE=
+github.com/golang/groupcache v0.0.0-20210331224755-41bb18bfe9da/go.mod h1:cIg4eruTrX1D+g88fzRXU5OdNfaM+9IcxsU14FzY7Hc=
+github.com/golang/mock v1.1.1/go.mod h1:oTYuIxOrZwtPieC+H1uAHpcLFnEyAGVDL/k47Jfbm0A=
+github.com/golang/protobuf v1.2.0/go.mod h1:6lQm79b+lXiMfvg/cZm0SGofjICqVBUtrP5yJMmIC1U=
+github.com/golang/protobuf v1.3.2/go.mod h1:6lQm79b+lXiMfvg/cZm0SGofjICqVBUtrP5yJMmIC1U=
+github.com/golang/protobuf v1.4.0-rc.1/go.mod h1:ceaxUfeHdC40wWswd/P6IGgMaK3YpKi5j83Wpe3EHw8=
+github.com/golang/protobuf v1.4.0-rc.1.0.20200221234624-67d41d38c208/go.mod h1:xKAWHe0F5eneWXFV3EuXVDTCmh+JuBKY0li0aMyXATA=
+github.com/golang/protobuf v1.4.0-rc.2/go.mod h1:LlEzMj4AhA7rCAGe4KMBDvJI+AwstrUpVNzEA03Pprs=
+github.com/golang/protobuf v1.4.0-rc.4.0.20200313231945-b860323f09d0/go.mod h1:WU3c8KckQ9AFe+yFwt9sWVRKCVIyN9cPHBJSNnbL67w=
+github.com/golang/protobuf v1.4.0/go.mod h1:jodUvKwWbYaEsadDk5Fwe5c77LiNKVO9IDvqG2KuDX0=
+github.com/golang/protobuf v1.4.1/go.mod h1:U8fpvMrcmy5pZrNK1lt4xCsGvpyWQ/VVv6QDs8UjoX8=
+github.com/golang/protobuf v1.4.3/go.mod h1:oDoupMAO8OvCJWAcko0GGGIgR6R6ocIYbsSw735rRwI=
+github.com/google/go-cmp v0.2.0/go.mod h1:oXzfMopK8JAjlY9xF4vHSVASa0yLyX7SntLO5aqRK0M=
+github.com/google/go-cmp v0.3.0/go.mod h1:8QqcDgzrUqlUb/G2PQTWiueGozuR1884gddMywk6iLU=
+github.com/google/go-cmp v0.3.1/go.mod h1:8QqcDgzrUqlUb/G2PQTWiueGozuR1884gddMywk6iLU=
+github.com/google/go-cmp v0.4.0/go.mod h1:v8dTdLbMG2kIc/vJvl+f65V22dbkXbowE6jgT/gNBxE=
+github.com/google/go-cmp v0.5.0/go.mod h1:v8dTdLbMG2kIc/vJvl+f65V22dbkXbowE6jgT/gNBxE=
+github.com/google/go-cmp v0.5.3/go.mod h1:v8dTdLbMG2kIc/vJvl+f65V22dbkXbowE6jgT/gNBxE=
 github.com/google/go-cmp v0.6.0 h1:ofyhxvXcZhMsU5ulbFiLKl/XBFqE1GSq7atu8tAmTRI=
 github.com/google/go-cmp v0.6.0/go.mod h1:17dUlkBOakJ0+DkrSSNjCkIjxS6bF9zb3elmeNGIjoY=
 github.com/google/gofuzz v1.0.0/go.mod h1:dBl0BpW6vV/+mYPU4Po3pmUjxk6FQPldtuIdl/M65Eg=
 github.com/google/pprof v0.0.0-20240409012703-83162a5b38cd h1:gbpYu9NMq8jhDVbvlGkMFWCjLFlqqEZjEmObmhUy6Vo=
 github.com/google/pprof v0.0.0-20240409012703-83162a5b38cd/go.mod h1:kf6iHlnVGwgKolg33glAes7Yg/8iWP8ukqeldJSO7jw=
+github.com/google/s2a-go v0.1.8 h1:zZDs9gcbt9ZPLV0ndSyQk6Kacx2g/X+SKYovpnz3SMM=
+github.com/google/s2a-go v0.1.8/go.mod h1:6iNWHTpQ+nfNRN5E00MSdfDwVesa8hhS32PhPO8deJA=
+github.com/google/uuid v1.1.2/go.mod h1:TIyPZe4MgqvfeYDBFedMoGGpEw/LqOeaOT+nhxU+yHo=
 github.com/google/uuid v1.6.0 h1:NIvaJDMOsjHA8n1jAhLSgzrAzy1Hgr+hNrb57e+94F0=
 github.com/google/uuid v1.6.0/go.mod h1:TIyPZe4MgqvfeYDBFedMoGGpEw/LqOeaOT+nhxU+yHo=
+github.com/googleapis/enterprise-certificate-proxy v0.3.4 h1:XYIDZApgAnrN1c855gTgghdIA6Stxb52D5RnLI1SLyw=
+github.com/googleapis/enterprise-certificate-proxy v0.3.4/go.mod h1:YKe7cfqYXjKGpGvmSg28/fFvhNzinZQm8DGnaburhGA=
+github.com/gorilla/websocket v1.5.3 h1:saDtZ6Pbx/0u+bgYQ3q96pZgCzfhKXGPqt7kZ72aNNg=
+github.com/gorilla/websocket v1.5.3/go.mod h1:YR8l580nyteQvAITg2hZ9XVh4b55+EU/adAjf1fMHhE=
 github.com/hashicorp/golang-lru/v2 v2.0.7 h1:a+bsQ5rvGLjzHuww6tVxozPZFVghXaHOwFs4luLUK2k=
 github.com/hashicorp/golang-lru/v2 v2.0.7/go.mod h1:QeFd9opnmA6QUJc5vARoKUSoFhyfM2/ZepoAG6RGpeM=
 github.com/json-iterator/go v1.1.12 h1:PV8peI4a0ysnczrg+LtxykD8LfKY9ML6u2jnxaEnrnM=
@@ -65,6 +113,7 @@ github.com/pelletier/go-toml/v2 v2.2.2 h1:aYUidT7k73Pcl9nb2gScu7NSrKCSHIDE89b3+6
 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/prometheus/client_model v0.0.0-20190812154241-14fe0d1b01d4/go.mod h1:xMI15A0UPsDsEKsMN9yxemIoYk6Tm2C1GtYGdfGttqA=
 github.com/remyoudompheng/bigfft v0.0.0-20230129092748-24d4a6f8daec h1:W09IVJc94icq4NjY3clb7Lk8O1qJ8BdBEF8z0ibU0rE=
 github.com/remyoudompheng/bigfft v0.0.0-20230129092748-24d4a6f8daec/go.mod h1:qqbHyh8v60DhA7CoWK5oRCqLrMHRGoxYCSS9EjAz6Eo=
 github.com/samber/slog-gin v1.15.0 h1:Kqs/ilXd9divtslWjbz5DVptmLlzyntbBiXUAta2SFg=
@@ -81,10 +130,14 @@ github.com/stretchr/testify v1.8.1/go.mod h1:w2LPCIKwWwSfY2zedu0+kehJoqGctiVI29o
 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/tetratelabs/wazero v1.12.0 h1:DuWcpNu/FzgEXgGBDp8J1Spc+CWOvvtvVyjKlaZopYU=
+github.com/tetratelabs/wazero v1.12.0/go.mod h1:LvKtzl2RqO4gyF27BiXU+nKAjcV8f38U+kP/q2vgxh0=
 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=
+go.opencensus.io v0.24.0 h1:y73uSU6J157QMP2kn2r30vwW1A2W2WFwSCGnAVxeaD0=
+go.opencensus.io v0.24.0/go.mod h1:vNK8G9p7aAivkbmorf4v+7Hgx+Zs0yY+0fOtgBfjQKo=
 go.opentelemetry.io/otel v1.29.0 h1:PdomN/Al4q/lN6iBJEN3AwPvUiHPMlt93c8bqTG5Llw=
 go.opentelemetry.io/otel v1.29.0/go.mod h1:N/WtXPs1CNCUEx+Agz5uouwCba+i+bJGFicT8SR4NP8=
 go.opentelemetry.io/otel/trace v1.29.0 h1:J/8ZNK4XgR7a21DZUAsbF8pZ5Jcw1VhACmnYt39JTi4=
@@ -92,24 +145,79 @@ go.opentelemetry.io/otel/trace v1.29.0/go.mod h1:eHl3w0sp3paPkYstJOmAimxhiFXPg+M
 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 v0.31.0 h1:ihbySMvVjLAeSH1IbfcRTkD/iNscyz8rGzjF/E5hV6U=
+golang.org/x/crypto v0.0.0-20190308221718-c2843e01d9a2/go.mod h1:djNgcEr1/C05ACkg1iLfiJU5Ep61QUkGW8qpdssI0+w=
-golang.org/x/crypto v0.31.0/go.mod h1:kDsLvtWBEx7MV9tJOj9bnXsPbxwJQ6csT/x4KIN4Ssk=
+golang.org/x/crypto v0.0.0-20200622213623-75b288015ac9/go.mod h1:LzIPMQfyMNhhGPhUkYOs5KpL4U8rLKemX1yGLhDgUto=
-golang.org/x/mod v0.17.0 h1:zY54UmvipHiNd+pm+m0x9KhZ9hl1/7QNMyxXbc6ICqA=
+golang.org/x/crypto v0.36.0 h1:AnAEvhDddvBdpY+uR+MyHmuZzzNqXSe/GvuDeob5L34=
-golang.org/x/mod v0.17.0/go.mod h1:hTbmBsO62+eylJbnUtE2MGJUyE7QWk4xUqPFrRgJ+7c=
+golang.org/x/crypto v0.36.0/go.mod h1:Y4J0ReaxCR1IMaabaSMugxJES1EpwhBHhv2bDHklZvc=
-golang.org/x/net v0.33.0 h1:74SYHlV8BIgHIFC/LrYkOGIwL19eTYXQ5wc6TBuO36I=
+golang.org/x/exp v0.0.0-20190121172915-509febef88a4/go.mod h1:CJ0aWSM057203Lf6IL+f9T1iT9GByDxfZKAQTCR3kQA=
-golang.org/x/net v0.33.0/go.mod h1:HXLR5J+9DxmrqMwG9qjGCxZ+zKXxBru04zlTvWlWuN4=
+golang.org/x/image v0.44.0 h1:+tDekMZED9+LrtB3G5xzRggpVh9CARjZqROla3R3R+I=
+golang.org/x/image v0.44.0/go.mod h1:V8K3KE9KKKE+pLpQDOeN18w9oacNSvy1tDOirTu4xtY=
+golang.org/x/lint v0.0.0-20181026193005-c67002cb31c3/go.mod h1:UVdnD1Gm6xHRNCYTkRU2/jEulfH38KcIWyp/GAMgvoE=
+golang.org/x/lint v0.0.0-20190227174305-5b3e6a55c961/go.mod h1:wehouNa3lNwaWXcvxsM5YxQ5yQlVC4a0KAMCusXpPoU=
+golang.org/x/lint v0.0.0-20190313153728-d0100b6bd8b3/go.mod h1:6SW0HCj/g11FgYtHlgUYUwCkIfeOF89ocIRzGO/8vkc=
+golang.org/x/mod v0.37.0 h1:vF1DjpVEshcIqoEaauuHebaLk1O1forxjxBaVn884JQ=
+golang.org/x/mod v0.37.0/go.mod h1:m8S8VeM9r4dzDwjrKO0a1sZP3YjeMamRRlD+fmR2Q/0=
+golang.org/x/net v0.0.0-20180724234803-3673e40ba225/go.mod h1:mL1N/T3taQHkDXs73rZJwtUhF3w3ftmwwsq0BUmARs4=
+golang.org/x/net v0.0.0-20180826012351-8a410e7b638d/go.mod h1:mL1N/T3taQHkDXs73rZJwtUhF3w3ftmwwsq0BUmARs4=
+golang.org/x/net v0.0.0-20190213061140-3a22650c66bd/go.mod h1:mL1N/T3taQHkDXs73rZJwtUhF3w3ftmwwsq0BUmARs4=
+golang.org/x/net v0.0.0-20190311183353-d8887717615a/go.mod h1:t9HGtf8HONx5eT2rtn7q6eTqICYqUVnKs3thJo3Qplg=
+golang.org/x/net v0.0.0-20190404232315-eb5bcb51f2a3/go.mod h1:t9HGtf8HONx5eT2rtn7q6eTqICYqUVnKs3thJo3Qplg=
+golang.org/x/net v0.0.0-20201110031124-69a78807bb2b/go.mod h1:sp8m0HH+o8qH0wwXwYZr8TS3Oi6o0r6Gce1SSxlDquU=
+golang.org/x/net v0.38.0 h1:vRMAPTMaeGqVhG5QyLJHqNDwecKTomGeqbnfZyKlBI8=
+golang.org/x/net v0.38.0/go.mod h1:ivrbrMbzFq5J41QOQh0siUuly180yBYtLp+CKbEaFx8=
+golang.org/x/oauth2 v0.0.0-20180821212333-d2e6202438be/go.mod h1:N/0e6XlmueqKjAGxoOufVs8QHGRruUQn6yWY3a++T0U=
 golang.org/x/oauth2 v0.36.0 h1:peZ/1z27fi9hUOFCAZaHyrpWG5lwe0RJEEEeH0ThlIs=
 golang.org/x/oauth2 v0.36.0/go.mod h1:YDBUJMTkDnJS+A4BP4eZBjCqtokkg1hODuPjwiGPO7Q=
-golang.org/x/sync v0.10.0 h1:3NQrjDixjgGwUOCaF8w2+VYHv0Ve/vGYSbdkTa98gmQ=
+golang.org/x/sync v0.0.0-20180314180146-1d60e4601c6f/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
-golang.org/x/sync v0.10.0/go.mod h1:Czt+wKu1gCyEFDUtn0jG5QVvpJ6rzVqr5aXyt9drQfk=
+golang.org/x/sync v0.0.0-20181108010431-42b317875d0f/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
+golang.org/x/sync v0.0.0-20190423024810-112230192c58/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
+golang.org/x/sync v0.22.0 h1:SZjpbeLmrCk4xhRSZFNZW5gFUeCeFgjekvI/+gfScek=
+golang.org/x/sync v0.22.0/go.mod h1:9xrNwdLfx4jkKbNva9FpL6vEN7evnE43NNNJQ2LF3+0=
+golang.org/x/sys v0.0.0-20180830151530-49385e6e1522/go.mod h1:STP8DvDyc/dI5b8T5hshtkjS+E42TnysNCUPdjciGhY=
+golang.org/x/sys v0.0.0-20190215142949-d0b11bdaac8a/go.mod h1:STP8DvDyc/dI5b8T5hshtkjS+E42TnysNCUPdjciGhY=
+golang.org/x/sys v0.0.0-20190412213103-97732733099d/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
+golang.org/x/sys v0.0.0-20200930185726-fdedc70b468f/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 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 v0.28.0 h1:Fksou7UEQUWlKvIdsqzJmUmCX3cZuD2+P3XyyzwMhlA=
+golang.org/x/sys v0.44.0 h1:ildZl3J4uzeKP07r2F++Op7E9B29JRUy+a27EibtBTQ=
-golang.org/x/sys v0.28.0/go.mod h1:/VUhepiaJMQUp4+oa/7Zr1D23ma6VTLIYjOOTFZPUcA=
+golang.org/x/sys v0.44.0/go.mod h1:4GL1E5IUh+htKOUEOaiffhrAeqysfVGipDYzABqnCmw=
-golang.org/x/text v0.21.0 h1:zyQAAkrwaneQ066sspRyJaG9VNi/YJ1NfzcGB3hZ/qo=
+golang.org/x/text v0.3.0/go.mod h1:NqM8EUOU14njkJ3fqMW+pc6Ldnwhi/IjpwHt7yyuwOQ=
-golang.org/x/text v0.21.0/go.mod h1:4IBbMaMmOPCJ8SecivzSH54+73PCFmPWxNTLm+vZkEQ=
+golang.org/x/text v0.3.3/go.mod h1:5Zoc/QRtKVWzQhOtBMvqHzDpF6irO9z98xDceosuGiQ=
-golang.org/x/tools v0.21.1-0.20240508182429-e35e4ccd0d2d h1:vU5i/LfpvrRCpgM/VPfJLg5KjxD3E+hfT1SH+d9zLwg=
+golang.org/x/text v0.40.0 h1:Ub2Z6/xjgF1WrYQz2nuITOEegKFtiIy+rieRJ5lHZKs=
-golang.org/x/tools v0.21.1-0.20240508182429-e35e4ccd0d2d/go.mod h1:aiJjzUbINMkxbQROHiO6hDPo2LHcIPhhQsa9DLh0yGk=
+golang.org/x/text v0.40.0/go.mod h1:hpnzDAfGV753zIKo+wk3u1bVKCGPbrnF7+7LBF/UHVY=
+golang.org/x/tools v0.0.0-20180917221912-90fa682c2a6e/go.mod h1:n7NCudcB/nEzxVGmLbDWY5pfWTLqBcC2KZ6jyYvM4mQ=
+golang.org/x/tools v0.0.0-20190114222345-bf090417da8b/go.mod h1:n7NCudcB/nEzxVGmLbDWY5pfWTLqBcC2KZ6jyYvM4mQ=
+golang.org/x/tools v0.0.0-20190226205152-f727befe758c/go.mod h1:9Yl7xja0Znq3iFh3HoIrodX9oNMXvdceNzlUR8zjMvY=
+golang.org/x/tools v0.0.0-20190311212946-11955173bddd/go.mod h1:LCzVGOaR6xXOjkQ3onu1FJEFr0SW1gC7cKk1uF8kGRs=
+golang.org/x/tools v0.0.0-20190524140312-2c0ae7006135/go.mod h1:RgjU9mgBXZiqYHBnxXauZ1Gv1EHHAz9KjViQ78xBX0Q=
+golang.org/x/tools v0.47.0 h1:7Kn5x/d1svx/PzryTsqeoZN4TZwqeH5pGWjefhLi/1Q=
+golang.org/x/tools v0.47.0/go.mod h1:dFHnyTvFWY212G+h7ZY4Vsp/K3U4/7W9TyVaAul8uCA=
+golang.org/x/xerrors v0.0.0-20191204190536-9bdfabe68543/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
+google.golang.org/appengine v1.1.0/go.mod h1:EbEs0AVv82hx2wNQdGPgUI5lhzA/G0D9YwlJXL52JkM=
+google.golang.org/appengine v1.4.0/go.mod h1:xpcJRLb0r/rnEns0DIKYYv+WjYCduHsrkT7/EB5XEv4=
+google.golang.org/genai v1.59.0 h1:xp+ydkJFW8hO0hTUaAkr8TrLM9HFP3NYAwFhPd0nDqA=
+google.golang.org/genai v1.59.0/go.mod h1:mDdPDFXo1Ats7f1WXVyZgWb/CkMzFWTWJruIMy7hGIU=
+google.golang.org/genproto v0.0.0-20180817151627-c66870c02cf8/go.mod h1:JiN7NxoALGmiZfu7CAH4rXhgtRTLTxftemlI0sWmxmc=
+google.golang.org/genproto v0.0.0-20190819201941-24fa4b261c55/go.mod h1:DMBHOl98Agz4BDEuKkezgsaosCRResVns1a3J2ZsMNc=
+google.golang.org/genproto v0.0.0-20200526211855-cb27e3aa2013/go.mod h1:NbSheEEYHJ7i3ixzK3sjbqSGDJWnxyFXZblF3eUsNvo=
+google.golang.org/genproto/googleapis/rpc v0.0.0-20240903143218-8af14fe29dc1 h1:pPJltXNxVzT4pK9yD8vR9X75DaWYYmLGMsEvBfFQZzQ=
+google.golang.org/genproto/googleapis/rpc v0.0.0-20240903143218-8af14fe29dc1/go.mod h1:UqMtugtsSgubUsoxbuAoiCXvqvErP7Gf0so0mK9tHxU=
+google.golang.org/grpc v1.19.0/go.mod h1:mqu4LbDTu4XGKhr4mRzUsmM4RtVoemTSY81AxZiDr8c=
+google.golang.org/grpc v1.23.0/go.mod h1:Y5yQAOtifL1yxbo5wqy6BxZv8vAUGQwXBOALyacEbxg=
+google.golang.org/grpc v1.25.1/go.mod h1:c3i+UQWmh7LiEpx4sFZnkU36qjEYZ0imhYfXVyQciAY=
+google.golang.org/grpc v1.27.0/go.mod h1:qbnxyOmOxrQa7FizSgH+ReBfzJrCY1pSN7KXBS8abTk=
+google.golang.org/grpc v1.33.2/go.mod h1:JMHMWHQWaTccqQQlmk3MJZS+GWXOdAesneDmEnv2fbc=
+google.golang.org/grpc v1.66.2 h1:3QdXkuq3Bkh7w+ywLdLvM56cmGvQHUMZpiCzt6Rqaoo=
+google.golang.org/grpc v1.66.2/go.mod h1:s3/l6xSSCURdVfAnL+TqCNMyTDAGN6+lZeVxnZR128Y=
+google.golang.org/protobuf v0.0.0-20200109180630-ec00e32a8dfd/go.mod h1:DFci5gLYBciE7Vtevhsrf46CRTquxDuWsQurQQe4oz8=
+google.golang.org/protobuf v0.0.0-20200221191635-4d8936d0db64/go.mod h1:kwYJMbMJ01Woi6D6+Kah6886xMZcty6N08ah7+eCXa0=
+google.golang.org/protobuf v0.0.0-20200228230310-ab0ca4ff8a60/go.mod h1:cfTl7dwQJ+fmap5saPgwCLgHXTUD7jkjRqWcaiX5VyM=
+google.golang.org/protobuf v1.20.1-0.20200309200217-e05f789c0967/go.mod h1:A+miEFZTKqfCUM6K7xSMQL9OKL/b6hQv+e19PK+JZNE=
+google.golang.org/protobuf v1.21.0/go.mod h1:47Nbq4nVaFHyn7ilMalzfO3qCViNmqZ2kzikPIcrTAo=
+google.golang.org/protobuf v1.22.0/go.mod h1:EGpADcykh3NcUnDUJcl1+ZksZNG86OlYog2l/sGQquU=
+google.golang.org/protobuf v1.23.0/go.mod h1:EGpADcykh3NcUnDUJcl1+ZksZNG86OlYog2l/sGQquU=
+google.golang.org/protobuf v1.23.1-0.20200526195155-81db48ad09cc/go.mod h1:EGpADcykh3NcUnDUJcl1+ZksZNG86OlYog2l/sGQquU=
+google.golang.org/protobuf v1.25.0/go.mod h1:9JNX74DMeImyA3h4bdi1ymwjUzf21/xIlbajtzgsN7c=
 google.golang.org/protobuf v1.34.2 h1:6xV6lTsCfpGD21XK49h7MhtcApnLqkfYgPcdHftf6hg=
 google.golang.org/protobuf v1.34.2/go.mod h1:qYOHts0dSfpeUzUFpOMr/WGzszTmLH+DiWniOlNbLDw=
 gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0=
@@ -118,6 +226,8 @@ gopkg.in/check.v1 v1.0.0-20180628173108-788fd7840127/go.mod h1:Co6ibVJAznAaIkqp8
 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=
+honnef.co/go/tools v0.0.0-20190102054323-c2f93a96b099/go.mod h1:rf3lG4BRIbNafJWhAfAdb/ePZxsR/4RtNHQocxwk9r4=
+honnef.co/go/tools v0.0.0-20190523083050-ea95bdfd59fc/go.mod h1:rf3lG4BRIbNafJWhAfAdb/ePZxsR/4RtNHQocxwk9r4=
 modernc.org/cc/v4 v4.21.4 h1:3Be/Rdo1fpr8GrQ7IVw9OHtplU4gWbb+wNgeoBMmGLQ=
 modernc.org/cc/v4 v4.21.4/go.mod h1:HM7VJTZbUCR3rV8EYBi9wxnJ0ZBRiGE5OeGXNA0IsLQ=
 modernc.org/ccgo/v4 v4.19.2 h1:lwQZgvboKD0jBwdaeVCTouxhxAyN6iawF3STraAal8Y=

internal/agent/doc.go

@@ -1,18 +1,36 @@
-// Package agent adapts pansy's service layer to majordomo tools so an agent can
+// Package agent is pansy's garden assistant: a majordomo toolbox over the
-// drive a garden in natural language ("fill the NE corner with garlic, the NW
+// service layer, plus the run loop that drives a model with it.
-// with basil, the south half with beans"). Each tool runs as a fixed actor and
-// therefore inherits pansy's ACL checks for free — a viewer's fill_region returns
-// ErrForbidden, exactly as the REST API would.
 //
-// Two deliberate separations keep the core server lean:
+// Every tool runs as a fixed actor, so the agent inherits pansy's permission
+// checks for free — a viewer's fill_region returns ErrForbidden, exactly as the
+// REST API would. That is the whole reason the tools are thin adapters over
+// internal/service and never reach past it; the moment one does, the ACL story
+// stops being automatic and becomes something to remember.
 //
-//   - cmd/pansy does NOT import this package, so the server binary carries no
+// # A turn is one change set
-//     agent/LLM dependencies.
-//   - the tool wiring (tools.go) sits behind the `majordomo` build tag, so the
-//     default `go build ./...` / `go test ./...` compiles without the majordomo
-//     module. Build the agent tools with `-tags majordomo` once majordomo is a
-//     dependency (`go get gitea.stevedudenhoeffer.com/steve/majordomo`).
 //
-// The agent harness itself (model loop, chat surface) lives in the
+// Runs wrap their whole turn in a single change set (source "agent"), so
-// majordomo/executus stack, outside this repo; this package is only the toolbox.
+// "empty the garlic bed and plant cucumbers" — one object edit and a dozen
+// planting inserts — undoes as one action rather than thirteen. That is what
+// makes acting without a confirmation prompt defensible: the answer to "it did
+// the wrong thing" is one click, not an archaeology exercise.
+//
+// # No build tag
+//
+// This package used to sit behind a `majordomo` build tag, with cmd/pansy
+// deliberately not importing it, so the default build carried no LLM
+// dependency. Both separations are gone, on purpose: a tag that keeps the agent
+// out of the binary only earns its keep if you would ever ship a build without
+// the agent, and the agent is the point. Keeping it would have meant an
+// untagged CI that never compiled the code that matters.
+//
+// majordomo is stdlib-first and pure Go, so CGO_ENABLED=0 and the single static
+// binary survive it.
+//
+// # Unconfigured instances
+//
+// With no API key the assistant is simply not offered: the chat route isn't
+// registered and the capability isn't advertised — the same shape as OIDC
+// 404ing when unconfigured. An instance without a key starts and serves the app
+// exactly as it did before.
 package agent

internal/agent/runtime.go

@@ -0,0 +1,232 @@
+package agent
+
+import (
+	"context"
+	"crypto/rand"
+	"encoding/hex"
+	"errors"
+	"fmt"
+	"log/slog"
+	"strings"
+	"time"
+
+	"gitea.stevedudenhoeffer.com/steve/majordomo/agent"
+	"gitea.stevedudenhoeffer.com/steve/majordomo/llm"
+
+	"gitea.stevedudenhoeffer.com/steve/pansy/internal/agentmodel"
+	"gitea.stevedudenhoeffer.com/steve/pansy/internal/domain"
+	"gitea.stevedudenhoeffer.com/steve/pansy/internal/service"
+)
+
+// Loop bounds. These exist so a stuck run terminates, NOT to control spend —
+// pansy is a personal tool and multi-tenant cost control is explicitly not a
+// concern here. A model that gets wedged calling describe_garden forever should
+// stop on its own, and a hung upstream shouldn't hold a connection open all day.
+const (
+	maxSteps = 24
+	// A turn that legitimately fills several beds does a lot of round trips, so
+	// this is generous; it is a backstop, not a budget.
+	runTimeout = 4 * time.Minute
+	// Successive all-error steps, and identical repeated calls, that end a run.
+	maxConsecutiveToolErrors = 4
+	maxSameCallRepeats       = 3
+)
+
+// Runner drives a model over pansy's toolbox. One per process; Run is safe to
+// call concurrently.
+type Runner struct {
+	svc   *service.Service
+	model llm.Model
+}
+
+// NewRunner resolves modelSpec against pansy's registry and returns a Runner, or
+// an error if the assistant can't be offered. Callers should treat an error as
+// "no assistant" rather than a startup failure — an instance with no key must
+// still serve the app.
+//
+// It takes the key and spec explicitly rather than a *config.Config so the same
+// constructor serves both boot (from env) and a runtime settings change (from
+// the DB) — the Runner has no idea which one configured it.
+func NewRunner(svc *service.Service, apiKey, modelSpec string) (*Runner, error) {
+	if apiKey == "" {
+		return nil, errors.New("agent: not configured")
+	}
+	// agentmodel.Resolve already rejects an empty/blank spec, so don't duplicate
+	// that guard here — one place decides what a valid spec is.
+	model, err := agentmodel.Resolve(apiKey, modelSpec)
+	if err != nil {
+		return nil, err
+	}
+	return &Runner{svc: svc, model: model}, nil
+}
+
+// Turn is the outcome of one exchange.
+type Turn struct {
+	// Reply is what to show the user.
+	Reply string `json:"reply"`
+	// ChangeSetID is the change set this turn produced, if it changed anything —
+	// the handle the UI needs to offer Undo.
+	ChangeSetID *int64 `json:"changeSetId,omitempty"`
+	// Steps is how many model round trips it took.
+	Steps int `json:"steps"`
+	// Truncated is set when the run hit its step cap rather than finishing.
+	Truncated bool `json:"truncated,omitempty"`
+}
+
+// Run executes one turn against a garden, as actorID.
+//
+// The whole turn runs inside ONE change set, so everything the model did undoes
+// together. That is what makes acting without a confirmation prompt defensible.
+// The scope is opened even for a turn that turns out to be a question — a change
+// set with no revisions is never written, so asking costs nothing.
+func (r *Runner) Run(ctx context.Context, actorID, gardenID int64, message string, history []llm.Message, onStep func(agent.Step)) (*Turn, error) {
+	message = strings.TrimSpace(message)
+	if message == "" {
+		return nil, domain.ErrInvalidInput
+	}
+
+	ctx, cancel := context.WithTimeout(ctx, runTimeout)
+	defer cancel()
+
+	garden, err := r.svc.GetGarden(ctx, actorID, gardenID)
+	if err != nil {
+		return nil, err
+	}
+
+	// An id for this run, stamped on the change set so a row in the history list
+	// can be matched to the log lines that produced it. Without it, "the agent
+	// did something odd on Tuesday" has no thread back to what it was thinking.
+	runID := newRunID()
+	slog.Info("agent: run start", "run", runID, "garden", gardenID, "actor", actorID)
+
+	var (
+		result   *agent.Result
+		runErr   error
+		truncErr bool
+	)
+	changeSet, err := r.svc.WithChangeSet(ctx, actorID, gardenID, service.ChangeSetOptions{
+		Source:     domain.SourceAgent,
+		Summary:    turnSummary(message),
+		AgentRunID: &runID,
+	}, func(ctx context.Context) error {
+		box := NewToolbox(r.svc, actorID)
+		a := agent.New(r.model, systemPrompt(garden),
+			agent.WithMaxSteps(maxSteps),
+			agent.WithToolErrorLimits(maxConsecutiveToolErrors, maxSameCallRepeats),
+		)
+		a.AddToolbox(box)
+
+		opts := []agent.RunOption{agent.WithHistory(history)}
+		if onStep != nil {
+			opts = append(opts, agent.OnStep(onStep))
+		}
+		result, runErr = a.Run(ctx, message, opts...)
+		// A run that ran out of steps still DID things, and those things must be
+		// recorded and undoable. So the loop-guard errors don't fail the scope —
+		// they're reported to the user instead.
+		if runErr != nil && isLoopLimit(runErr) {
+			truncErr = true
+			return nil
+		}
+		return runErr
+	})
+	if err != nil {
+		// WithChangeSet records whatever committed before failing, so the partial
+		// work is undoable even though the turn errored.
+		return nil, err
+	}
+
+	turn := &Turn{Truncated: truncErr}
+	if changeSet != nil {
+		turn.ChangeSetID = &changeSet.ID
+	}
+	if result != nil {
+		turn.Reply = result.Output
+		turn.Steps = len(result.Steps)
+	}
+	if turn.Reply == "" {
+		turn.Reply = fallbackReply(turn)
+	}
+	return turn, nil
+}
+
+// isLoopLimit reports whether an error is one of majordomo's loop guards firing
+// rather than a genuine failure. Those runs have a partial result worth keeping.
+func isLoopLimit(err error) bool {
+	return errors.Is(err, agent.ErrMaxSteps) || errors.Is(err, agent.ErrToolLoop)
+}
+
+// fallbackReply covers a run that finished with no text — a model that made its
+// last tool call and then stopped. Silence reads as a failure, so say what
+// happened.
+func fallbackReply(t *Turn) string {
+	switch {
+	case t.Truncated:
+		return "I stopped partway through — that turned into more steps than I should take in one go. " +
+			"Have a look at what changed, and tell me what to do next."
+	case t.ChangeSetID != nil:
+		return "Done — have a look at the canvas."
+	default:
+		return "I didn't change anything."
+	}
+}
+
+// newRunID returns a short random identifier for one run.
+func newRunID() string {
+	var b [8]byte
+	if _, err := rand.Read(b[:]); err != nil {
+		// The id is for correlating logs, not for security. A clock-based
+		// fallback is worse than random and better than an empty string.
+		return fmt.Sprintf("t%d", time.Now().UnixNano())
+	}
+	return hex.EncodeToString(b[:])
+}
+
+// turnSummary is what the history list shows for this turn. The user's own words
+// are the most useful label available, trimmed to fit a list row.
+//
+// Trimmed by RUNES, not bytes: slicing a byte offset would cut a multibyte
+// character in half and store invalid UTF-8 in the summary — which is not a
+// hypothetical for text people type.
+func turnSummary(message string) string {
+	const max = 120
+	s := strings.Join(strings.Fields(message), " ")
+	runes := []rune(s)
+	if len(runes) > max {
+		s = strings.TrimSpace(string(runes[:max])) + "…"
+	}
+	return s
+}
+
+// systemPrompt gives the model the conventions it cannot infer.
+//
+// The compass convention in particular is not guessable: -y is north because
+// screen y grows downward, and a model that assumes otherwise plants the south
+// half when asked for the north one.
+func systemPrompt(g *domain.Garden) string {
+	units := "metric — all measurements are centimeters"
+	if g.UnitPref == domain.UnitImperial {
+		units = "imperial for display, but every measurement you send or receive is in CENTIMETERS"
+	}
+	return fmt.Sprintf(`You are pansy's garden assistant. You help plan and edit a real garden by calling tools.
+
+The garden you are working on is %q (id %d), %.0f x %.0f cm. The user's units are %s.
+
+Conventions you cannot guess and must not assume:
+- Positions in a garden are centimeters from its top-left corner: x grows east, y grows SOUTH.
+- Inside an object (a bed), positions are relative to that object's CENTER, and -y is NORTH.
+  So the north half of a bed is negative y. Getting this backwards plants the wrong end.
+- Objects and plantings are version-guarded. Use the version from describe_garden when editing.
+
+How to work:
+- Start from describe_garden to see what is actually there. Do not guess ids.
+- Use find_plant to turn a plant name into an id. If it returns several candidates,
+  pick the one that matches what the user said, or ask them which they meant.
+- To replant a bed with something else: clear_object, then fill_region with region "all".
+- When a tool refuses (for example, the user only has view access to this garden),
+  explain what happened in plain words. Do not retry it.
+
+When you are done, say briefly what you changed — the user is watching the canvas
+and wants to know what to look at. If you changed nothing, say that too.`,
+		g.Name, g.ID, g.WidthCM, g.HeightCM, units)
+}

internal/agent/runtime_test.go

@@ -0,0 +1,358 @@
+package agent
+
+import (
+	"context"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"strings"
+	"testing"
+	"time"
+	"unicode/utf8"
+
+	"gitea.stevedudenhoeffer.com/steve/majordomo/llm"
+	"gitea.stevedudenhoeffer.com/steve/majordomo/provider/fake"
+
+	"gitea.stevedudenhoeffer.com/steve/pansy/internal/domain"
+	"gitea.stevedudenhoeffer.com/steve/pansy/internal/service"
+)
+
+// scriptedRunner wires a Runner to a fake provider so the whole loop — tools,
+// change-set scoping, loop guards — is exercised with no live model.
+func scriptedRunner(t *testing.T, svc *service.Service, steps ...fake.Step) *Runner {
+	t.Helper()
+	p := fake.New("fake")
+	for _, s := range steps {
+		p.Enqueue("m", s)
+	}
+	model, err := p.Model("m")
+	if err != nil {
+		t.Fatalf("fake model: %v", err)
+	}
+	return &Runner{svc: svc, model: model}
+}
+
+// toolCall scripts one model turn that calls a tool.
+func toolCall(name string, args any) fake.Step {
+	raw, _ := json.Marshal(args)
+	return fake.ReplyWith(llm.Response{
+		FinishReason: llm.FinishToolCalls,
+		ToolCalls:    []llm.ToolCall{{ID: "c1", Name: name, Arguments: raw}},
+	})
+}
+
+// TestTurnIsOneChangeSet is the acceptance criterion the whole "act freely"
+// posture rests on: a turn that clears a bed and replants it — one object edit
+// and many planting inserts — has to undo as ONE action, not thirteen.
+func TestTurnIsOneChangeSet(t *testing.T) {
+	ctx := context.Background()
+	svc, owner := newAgentTestService(t)
+
+	g, err := svc.CreateGarden(ctx, owner, service.GardenInput{Name: "Plot", WidthCM: 2000, HeightCM: 2000})
+	if err != nil {
+		t.Fatalf("garden: %v", err)
+	}
+	garlic := mustPlant(t, svc, owner, "Garlic", 15, "🧄")
+	cucumber := mustPlant(t, svc, owner, "Cucumber", 45, "🥒")
+	bed, err := svc.CreateObject(ctx, owner, g.ID, service.ObjectInput{
+		Kind: domain.KindBed, Name: "Garlic bed", XCM: 1000, YCM: 1000, WidthCM: 400, HeightCM: 400,
+	})
+	if err != nil {
+		t.Fatalf("bed: %v", err)
+	}
+	if _, err := svc.FillNamedRegion(ctx, owner, bed.ID, "all", garlic.ID, nil); err != nil {
+		t.Fatalf("seed garlic: %v", err)
+	}
+
+	before, _, err := svc.GardenHistory(ctx, owner, g.ID, 0, 0)
+	if err != nil {
+		t.Fatalf("history: %v", err)
+	}
+
+	r := scriptedRunner(t, svc,
+		toolCall("clear_object", map[string]any{"objectId": bed.ID}),
+		toolCall("fill_region", map[string]any{"objectId": bed.ID, "region": "all", "plantId": cucumber.ID}),
+		fake.Reply("Cleared the garlic and replanted the bed with cucumbers."),
+	)
+
+	turn, err := r.Run(ctx, owner, g.ID, "change the garlic bed to cucumbers this year", nil, nil)
+	if err != nil {
+		t.Fatalf("Run: %v", err)
+	}
+	if turn.ChangeSetID == nil {
+		t.Fatal("a turn that changed things produced no change set")
+	}
+	if !strings.Contains(turn.Reply, "cucumbers") {
+		t.Errorf("reply = %q", turn.Reply)
+	}
+
+	// Exactly ONE new change set, whatever the model did inside the turn.
+	after, _, err := svc.GardenHistory(ctx, owner, g.ID, 0, 0)
+	if err != nil {
+		t.Fatalf("history: %v", err)
+	}
+	if len(after)-len(before) != 1 {
+		t.Fatalf("turn produced %d change sets, want exactly 1", len(after)-len(before))
+	}
+	cs := after[0]
+	if cs.Source != domain.SourceAgent {
+		t.Errorf("source = %q, want agent", cs.Source)
+	}
+	if cs.Summary != "change the garlic bed to cucumbers this year" {
+		t.Errorf("summary = %q, want the user's own words", cs.Summary)
+	}
+
+	// The bed really is cucumbers now.
+	full, _ := svc.GardenFull(ctx, owner, g.ID, nil)
+	if len(full.Plantings) == 0 {
+		t.Fatal("bed ended up empty")
+	}
+	for _, p := range full.Plantings {
+		if p.PlantID != cucumber.ID {
+			t.Errorf("unexpected plant %d still in the bed", p.PlantID)
+		}
+	}
+
+	// And one undo puts it back.
+	if _, conflicts, err := svc.RevertChangeSet(ctx, owner, *turn.ChangeSetID, domain.SourceUI); err != nil || len(conflicts) != 0 {
+		t.Fatalf("undo: err=%v conflicts=%+v", err, conflicts)
+	}
+	full, _ = svc.GardenFull(ctx, owner, g.ID, nil)
+	for _, p := range full.Plantings {
+		if p.PlantID != garlic.ID {
+			t.Errorf("after undo the bed holds plant %d, want the garlic back", p.PlantID)
+		}
+	}
+}
+
+// TestViewerGetsAnExplainableRefusal — the ACL story only works if the model can
+// narrate the refusal, so a tool denial has to reach it as a tool RESULT it can
+// read, not as a 500 that ends the run.
+func TestViewerGetsAnExplainableRefusal(t *testing.T) {
+	ctx := context.Background()
+	svc, owner := newAgentTestService(t)
+	viewer, err := svc.Register(ctx, service.RegisterInput{Email: "[email protected]", DisplayName: "V", Password: "password123"})
+	if err != nil {
+		t.Fatalf("register: %v", err)
+	}
+	g, err := svc.CreateGarden(ctx, owner, service.GardenInput{Name: "Plot", WidthCM: 2000, HeightCM: 2000})
+	if err != nil {
+		t.Fatalf("garden: %v", err)
+	}
+	bed, err := svc.CreateObject(ctx, owner, g.ID, service.ObjectInput{
+		Kind: domain.KindBed, XCM: 1000, YCM: 1000, WidthCM: 400, HeightCM: 400,
+	})
+	if err != nil {
+		t.Fatalf("bed: %v", err)
+	}
+	if _, err := svc.AddShare(ctx, owner, g.ID, "[email protected]", domain.RoleViewer); err != nil {
+		t.Fatalf("share: %v", err)
+	}
+
+	// A viewer can't open a change set at all, so the turn is refused up front —
+	// before any model call — and the API turns that into a plain explanation.
+	r := scriptedRunner(t, svc, fake.Reply("unused"))
+	_, err = r.Run(ctx, viewer.ID, g.ID, "plant garlic in that bed", nil, nil)
+	if !errors.Is(err, domain.ErrForbidden) {
+		t.Fatalf("viewer turn err = %v, want ErrForbidden", err)
+	}
+
+	// And at the tool layer, a refusal comes back as a readable tool result
+	// rather than killing the run.
+	box := NewToolbox(svc, viewer.ID)
+	raw, _ := json.Marshal(map[string]any{"objectId": bed.ID})
+	res := box.Execute(ctx, llm.ToolCall{ID: "1", Name: "clear_object", Arguments: raw})
+	if !res.IsError {
+		t.Fatal("a viewer's clear_object succeeded")
+	}
+	if res.Content == "" {
+		t.Error("the refusal carried no text for the model to explain")
+	}
+}
+
+// TestRunStopsAtTheStepCap — a model that gets wedged should stop on its own,
+// and what it managed to do must still be recorded and undoable.
+func TestRunStopsAtTheStepCap(t *testing.T) {
+	ctx := context.Background()
+	svc, owner := newAgentTestService(t)
+	g, err := svc.CreateGarden(ctx, owner, service.GardenInput{Name: "Plot", WidthCM: 2000, HeightCM: 2000})
+	if err != nil {
+		t.Fatalf("garden: %v", err)
+	}
+
+	// More describe_garden calls than the cap allows, forever.
+	steps := make([]fake.Step, 0, maxSteps+4)
+	for i := 0; i < maxSteps+4; i++ {
+		steps = append(steps, toolCall("describe_garden", map[string]any{"gardenId": g.ID}))
+	}
+	r := scriptedRunner(t, svc, steps...)
+
+	turn, err := r.Run(ctx, owner, g.ID, "look at the garden", nil, nil)
+	if err != nil {
+		t.Fatalf("a capped run should end cleanly, got %v", err)
+	}
+	if !turn.Truncated {
+		t.Error("turn.Truncated = false; the run hit the cap and should say so")
+	}
+	if turn.Reply == "" {
+		t.Error("a capped run said nothing; silence reads as a hang")
+	}
+	// It only read, so there is nothing to undo — and no empty change set either.
+	if turn.ChangeSetID != nil {
+		t.Errorf("a read-only turn produced change set %d", *turn.ChangeSetID)
+	}
+}
+
+// TestReadOnlyTurnWritesNoChangeSet — asking a question must not litter the
+// history with empty entries.
+func TestReadOnlyTurnWritesNoChangeSet(t *testing.T) {
+	ctx := context.Background()
+	svc, owner := newAgentTestService(t)
+	g, err := svc.CreateGarden(ctx, owner, service.GardenInput{Name: "Plot", WidthCM: 2000, HeightCM: 2000})
+	if err != nil {
+		t.Fatalf("garden: %v", err)
+	}
+	before, _, _ := svc.GardenHistory(ctx, owner, g.ID, 0, 0)
+
+	r := scriptedRunner(t, svc,
+		toolCall("describe_garden", map[string]any{"gardenId": g.ID}),
+		fake.Reply("It's empty — nothing planted yet."),
+	)
+	turn, err := r.Run(ctx, owner, g.ID, "what's in the garden?", nil, nil)
+	if err != nil {
+		t.Fatalf("Run: %v", err)
+	}
+	if turn.ChangeSetID != nil {
+		t.Errorf("a question produced change set %d", *turn.ChangeSetID)
+	}
+	after, _, _ := svc.GardenHistory(ctx, owner, g.ID, 0, 0)
+	if len(after) != len(before) {
+		t.Errorf("history grew by %d for a read-only turn", len(after)-len(before))
+	}
+}
+
+// TestNewRunnerNeedsConfiguration — an instance with no key or no model must not
+// get a half-built runner; the caller treats the error as "no assistant" and
+// carries on. (Whether the assistant is ENABLED is resolved before NewRunner is
+// reached, so it isn't NewRunner's concern any more.)
+func TestNewRunnerNeedsConfiguration(t *testing.T) {
+	svc, _ := newAgentTestService(t)
+	for _, tc := range []struct{ key, model string }{
+		{"", "ollama-cloud/x"},
+		{"k", ""},
+		{"k", "   "},
+	} {
+		if _, err := NewRunner(svc, tc.key, tc.model); err == nil {
+			t.Errorf("NewRunner accepted key=%q model=%q", tc.key, tc.model)
+		}
+	}
+	// A model spec naming a provider that doesn't exist is a configuration
+	// error, not a panic at first use.
+	if _, err := NewRunner(svc, "k", "nonesuch/model"); err == nil {
+		t.Error("NewRunner accepted an unresolvable model spec")
+	}
+}
+
+// TestTurnSummaryFitsAHistoryRow — the user's own words are the most useful
+// label for a change set, but a paragraph would wreck the list.
+func TestTurnSummaryFitsAHistoryRow(t *testing.T) {
+	if got := turnSummary("  change   the garlic bed\n to cucumbers "); got != "change the garlic bed to cucumbers" {
+		t.Errorf("turnSummary = %q", got)
+	}
+	long := turnSummary(strings.Repeat("plant garlic ", 40))
+	if len(long) > 130 || !strings.HasSuffix(long, "…") {
+		t.Errorf("long summary = %q (%d chars)", long, len(long))
+	}
+}
+
+// TestSystemPromptStatesTheCompassConvention — -y being north is not guessable,
+// and a model that assumes otherwise plants the wrong end of the bed.
+func TestSystemPromptStatesTheCompassConvention(t *testing.T) {
+	p := systemPrompt(&domain.Garden{ID: 1, Name: "Plot", WidthCM: 500, HeightCM: 400, UnitPref: domain.UnitImperial})
+	for _, want := range []string{"NORTH", "-y", "centimeters", "Plot", "version"} {
+		if !strings.Contains(p, want) {
+			t.Errorf("system prompt is missing %q:\n%s", want, p)
+		}
+	}
+	if !strings.Contains(p, fmt.Sprintf("%.0f", 500.0)) {
+		t.Error("system prompt doesn't state the garden's size")
+	}
+}
+
+// TestPartialWorkSurvivesATimeout is the finding that mattered most on this PR.
+//
+// The run context carries a timeout. When it fires, WithChangeSet's recovery
+// path has to record what already committed — and doing that with the SAME
+// dead context would fail, losing the history for changes that really happened.
+// The user-facing message says "anything I'd already changed is in History", so
+// this isn't just a gap, it's a promise the code has to keep.
+func TestPartialWorkSurvivesATimeout(t *testing.T) {
+	ctx := context.Background()
+	svc, owner := newAgentTestService(t)
+	g, err := svc.CreateGarden(ctx, owner, service.GardenInput{Name: "Plot", WidthCM: 2000, HeightCM: 2000})
+	if err != nil {
+		t.Fatalf("garden: %v", err)
+	}
+	bed, err := svc.CreateObject(ctx, owner, g.ID, service.ObjectInput{
+		Kind: domain.KindBed, Name: "Bed", XCM: 1000, YCM: 1000, WidthCM: 400, HeightCM: 400,
+	})
+	if err != nil {
+		t.Fatalf("bed: %v", err)
+	}
+	before, _, _ := svc.GardenHistory(ctx, owner, g.ID, 0, 0)
+
+	// A turn that renames the bed, then dies with the context already cancelled.
+	cancelled, cancel := context.WithCancel(ctx)
+	r := scriptedRunner(t, svc,
+		toolCall("move_object", map[string]any{
+			"objectId": bed.ID, "xCm": 600.0, "yCm": 600.0, "version": bed.Version,
+		}),
+		fake.Step{Err: context.DeadlineExceeded},
+	)
+	// Cancel once the first tool call has landed, so the failure path runs with a
+	// dead context — exactly the timeout case.
+	go func() {
+		time.Sleep(50 * time.Millisecond)
+		cancel()
+	}()
+	_, err = r.Run(cancelled, owner, g.ID, "move the bed", nil, nil)
+	if err == nil {
+		t.Fatal("expected the turn to fail")
+	}
+
+	// The move committed, so it must be in history and undoable.
+	after, _, herr := svc.GardenHistory(ctx, owner, g.ID, 0, 0)
+	if herr != nil {
+		t.Fatalf("history: %v", herr)
+	}
+	if len(after) != len(before)+1 {
+		t.Fatalf("the failed turn recorded %d change sets, want 1 — its work is otherwise un-undoable",
+			len(after)-len(before))
+	}
+	if !strings.Contains(after[0].Summary, "failed partway") {
+		t.Errorf("summary = %q, want it marked as partial", after[0].Summary)
+	}
+	if _, conflicts, rerr := svc.RevertChangeSet(ctx, owner, after[0].ID, domain.SourceUI); rerr != nil || len(conflicts) != 0 {
+		t.Fatalf("the partial turn should be undoable: err=%v conflicts=%+v", rerr, conflicts)
+	}
+	o, _ := svc.DescribeGarden(ctx, owner, g.ID)
+	if len(o.Objects) > 0 && o.Objects[0].XCM != bed.XCM {
+		t.Errorf("undo left the bed at %v, want %v", o.Objects[0].XCM, bed.XCM)
+	}
+}
+
+// TestTurnSummaryTrimsByRunes — slicing a byte offset would cut a multibyte
+// character in half and store invalid UTF-8 in the history summary.
+func TestTurnSummaryTrimsByRunes(t *testing.T) {
+	// 200 multibyte runes: a byte slice at 120 would land mid-character.
+	got := turnSummary(strings.Repeat("🌱", 200))
+	if !utf8.ValidString(got) {
+		t.Errorf("turnSummary produced invalid UTF-8: %q", got)
+	}
+	if !strings.HasSuffix(got, "…") {
+		t.Errorf("long summary should be elided, got %q", got)
+	}
+	if n := utf8.RuneCountInString(got); n > 121 {
+		t.Errorf("summary is %d runes, want it trimmed", n)
+	}
+}

internal/agent/tools.go

@@ -1,5 +1,3 @@
-//go:build majordomo
-
 package agent
 
 import (

internal/agent/tools_test.go

@@ -1,5 +1,3 @@
-//go:build majordomo
-
 package agent
 
 import (

internal/agentmodel/agentmodel.go

@@ -0,0 +1,57 @@
+// Package agentmodel holds the ONE place that knows how pansy turns a model spec
+// into a majordomo model: which provider to register and under which token.
+//
+// It exists as a leaf so both internal/agent (which builds the run loop) and
+// internal/service (which validates a spec before storing it as a setting) can
+// share that knowledge without an import cycle — agent imports service, so the
+// shared bit can live in neither of them.
+package agentmodel
+
+import (
+	"errors"
+	"fmt"
+	"strings"
+
+	"gitea.stevedudenhoeffer.com/steve/majordomo"
+	"gitea.stevedudenhoeffer.com/steve/majordomo/llm"
+	"gitea.stevedudenhoeffer.com/steve/majordomo/provider/ollama"
+)
+
+// registry builds the private majordomo registry pansy uses.
+//
+// Private, not the package-level default: pansy passes the key it was configured
+// with rather than depending on ambient environment, and majordomo's own
+// ollama-cloud preset reads OLLAMA_API_KEY while pansy (like gadfly) is
+// configured with OLLAMA_CLOUD_API_KEY. Registering the provider explicitly
+// makes that bridge visible instead of a mysterious empty token.
+func registry(apiKey string) *majordomo.Registry {
+	reg := majordomo.New()
+	reg.RegisterProvider(ollama.Cloud(ollama.WithToken(apiKey)))
+	return reg
+}
+
+// Resolve parses a model spec against pansy's registry into a live model. The
+// spec goes to Parse VERBATIM — the grammar, including comma-separated failover
+// chains, is majordomo's, and re-implementing any of it here would only mean two
+// places to update when it grows.
+func Resolve(apiKey, spec string) (llm.Model, error) {
+	if strings.TrimSpace(spec) == "" {
+		return nil, errors.New("agentmodel: empty model spec")
+	}
+	m, err := registry(apiKey).Parse(spec)
+	if err != nil {
+		return nil, fmt.Errorf("agentmodel: resolve %q: %w", spec, err)
+	}
+	return m, nil
+}
+
+// Validate reports whether a spec resolves, without building anything the caller
+// keeps — the cheap, deterministic check a settings save runs to reject a typo.
+//
+// It does NOT make a live call, so it needs no working key and won't catch a
+// model that is merely absent upstream; that surfaces on first use. Parse
+// resolving (known provider, well-formed spec) is the half worth doing eagerly.
+func Validate(apiKey, spec string) error {
+	_, err := Resolve(apiKey, spec)
+	return err
+}

internal/agentmodel/agentmodel_test.go

@@ -0,0 +1,35 @@
+package agentmodel
+
+import "testing"
+
+// TestValidate is what the settings PATCH relies on to reject a typo at save
+// time rather than on the next chat turn.
+func TestValidate(t *testing.T) {
+	if err := Validate("k", "ollama-cloud/glm-5.2:cloud"); err != nil {
+		t.Errorf("valid spec rejected: %v", err)
+	}
+	// No key needed to validate — Parse resolves the provider, it doesn't call it.
+	if err := Validate("", "ollama-cloud/glm-5.2:cloud"); err != nil {
+		t.Errorf("valid spec rejected without a key: %v", err)
+	}
+	// A comma-separated failover chain is majordomo grammar and must resolve.
+	if err := Validate("k", "ollama-cloud/glm-5.2:cloud,ollama-cloud/kimi-k2.6:cloud"); err != nil {
+		t.Errorf("failover chain rejected: %v", err)
+	}
+	for _, bad := range []string{"", "   ", "nonesuch/model"} {
+		if err := Validate("k", bad); err == nil {
+			t.Errorf("Validate accepted %q", bad)
+		}
+	}
+}
+
+// TestResolveReturnsAModel confirms a good spec yields a usable model handle.
+func TestResolveReturnsAModel(t *testing.T) {
+	m, err := Resolve("k", "ollama-cloud/glm-5.2:cloud")
+	if err != nil {
+		t.Fatalf("resolve: %v", err)
+	}
+	if m == nil {
+		t.Fatal("resolve returned a nil model with no error")
+	}
+}

internal/api/agent.go

@@ -0,0 +1,306 @@
+package api
+
+import (
+	"context"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"io"
+	"log/slog"
+	"net/http"
+	"sync"
+	"time"
+
+	mdagent "gitea.stevedudenhoeffer.com/steve/majordomo/agent"
+	"gitea.stevedudenhoeffer.com/steve/majordomo/llm"
+	"github.com/gin-gonic/gin"
+
+	"gitea.stevedudenhoeffer.com/steve/pansy/internal/agent"
+	"gitea.stevedudenhoeffer.com/steve/pansy/internal/domain"
+)
+
+// The garden assistant's chat surface (#56).
+//
+// Streaming, because a turn that clears a bed and replants it makes a dozen tool
+// calls over tens of seconds. Without streaming that is a long silence followed
+// by everything at once, which reads as a hang — and the whole design rests on
+// watching the canvas change as it happens.
+
+// keepAliveInterval is how often a quiet stream emits a comment frame. Well
+// under the 30–60s idle timeout typical of reverse proxies, which is the thing
+// it exists to stay ahead of.
+const keepAliveInterval = 20 * time.Second
+
+// chatRequest is the body of POST /agent/chat.
+type chatRequest struct {
+	GardenID int64  `json:"gardenId" binding:"required"`
+	Message  string `json:"message" binding:"required"`
+}
+
+// chatEvent is one server-sent event. Exactly one field is set.
+type chatEvent struct {
+	// Step reports a completed model round trip: which tools it called.
+	Step *stepEvent `json:"step,omitempty"`
+	// Done carries the finished turn.
+	Done *agent.Turn `json:"done,omitempty"`
+	// Error is a turn that failed, in words meant for a person.
+	Error string `json:"error,omitempty"`
+	// Warning rides alongside Done: the turn worked, but something adjacent to it
+	// didn't, and saying nothing would be the quieter lie.
+	Warning string `json:"warning,omitempty"`
+}
+
+type stepEvent struct {
+	Index int      `json:"index"`
+	Tools []string `json:"tools"`
+}
+
+func (h *handlers) agentChat(c *gin.Context) {
+	// The route is always registered, so the assistant being off is a runtime
+	// state, not a missing route: answer it plainly rather than 404ing a path
+	// that exists. Loaded once here so a settings-driven swap mid-request can't
+	// make it flip between the guard and the Run call.
+	runner := h.agent.get()
+	if runner == nil {
+		writeAPIError(c, http.StatusServiceUnavailable, "AGENT_DISABLED", "the garden assistant isn't enabled on this instance")
+		return
+	}
+
+	var req chatRequest
+	if err := c.ShouldBindJSON(&req); err != nil {
+		writeAPIError(c, http.StatusBadRequest, "INVALID_INPUT", "a gardenId and a message are required")
+		return
+	}
+	actor := mustActor(c)
+
+	history, err := h.svc.AgentHistory(c.Request.Context(), actor.ID, req.GardenID)
+	if err != nil {
+		writeServiceError(c, err)
+		return
+	}
+
+	stream := openEventStream(c)
+	send := stream.send
+
+	// A model thinking hard between tool calls sends nothing for a while, and an
+	// idle proxy will cut a quiet connection. Deferred so a panic in the run
+	// can't leak the ticker goroutine; stopping it twice is harmless.
+	stopBeat := stream.keepAlive(keepAliveInterval)
+	defer stopBeat()
+
+	turn, err := runner.Run(c.Request.Context(), actor.ID, req.GardenID, req.Message,
+		replayHistory(history),
+		func(s mdagent.Step) {
+			send(chatEvent{Step: &stepEvent{Index: s.Index, Tools: toolNames(s)}})
+		})
+	stopBeat()
+	if err != nil {
+		// The stream is already open, so an error is an event rather than a
+		// status code — the client has committed to reading a stream by now.
+		send(chatEvent{Error: chatErrorMessage(err)})
+		return
+	}
+
+	// The turn itself succeeded — the garden really did change — so Done goes out
+	// regardless. But if the transcript couldn't be saved, say so: a clean "done"
+	// followed by a conversation that has forgotten the exchange after a reload is
+	// exactly the kind of quiet inconsistency that makes a tool feel unreliable.
+	//
+	// Detached from the request context, because the commonest reason this fails
+	// is the client having gone away — and the exchange is worth keeping either
+	// way, since the change set it produced certainly is.
+	if _, err := h.svc.RecordAgentExchange(context.WithoutCancel(c.Request.Context()), actor.ID, req.GardenID,
+		req.Message, turn.Reply, turn.ChangeSetID); err != nil {
+		slog.Error("api: record agent exchange", "error", err, "garden", req.GardenID)
+		send(chatEvent{Done: turn, Warning: "I couldn't save this exchange, so it won't be here after a reload. Anything I changed is still on the canvas, and in History."})
+		return
+	}
+	send(chatEvent{Done: turn})
+}
+
+// sseWriteTimeout bounds ONE write to the stream, not the stream itself.
+//
+// It is refreshed per frame, which is the only shape that satisfies both ends:
+// the server's absolute WriteTimeout would cut a long turn (#78), while removing
+// the deadline entirely would let a client that stops reading block a write
+// forever once the socket buffer fills — pinning the run goroutine and this
+// stream's mutex with it, and taking the keep-alive down too since it needs the
+// same lock. Generous, because it is a backstop against a stuck peer and not a
+// pacing mechanism.
+//
+// A var, not a const, ONLY so the test can shrink it to prove the deadline is
+// refreshed per frame rather than set once — a set-once 30s deadline would pass
+// a test whose whole run is under a second. Production never reassigns it.
+var sseWriteTimeout = 30 * time.Second
+
+// eventStream serializes writes to one SSE response.
+//
+// The mutex is load-bearing, not decoration: step events are sent from the
+// agent's run goroutine while the keep-alive ticker writes from its own, and two
+// goroutines writing a ResponseWriter concurrently is a data race that corrupts
+// frames long before it crashes anything.
+type eventStream struct {
+	c  *gin.Context
+	rc *http.ResponseController
+	mu sync.Mutex
+}
+
+// openEventStream puts the response into SSE mode.
+//
+// Headers go out before the first write and the stream is flushed immediately,
+// so a proxy holding the response until it looks complete can't reintroduce
+// exactly the silence streaming exists to remove.
+//
+// Taking the write deadline off the server's absolute WriteTimeout and onto a
+// per-write one is what makes a turn longer than 30s possible at all (#78).
+// WriteTimeout is an ABSOLUTE deadline from when the request header was read,
+// not an idle timeout, so a streaming response is cut mid-turn however recently
+// it wrote. Without this the 4-minute runTimeout is unreachable and the
+// keep-alive below tops out at one tick — pacing a connection that is destroyed
+// underneath it.
+//
+// That failure is INVISIBLE from in here: writes past the deadline return
+// err == nil and their bytes are dropped, so there is nothing to detect on the
+// write path. Only the client sees it, as a truncated stream it reports as a
+// dropped connection. Hence a deadline set up front and refreshed per frame,
+// rather than anything checked after the fact.
+func openEventStream(c *gin.Context) *eventStream {
+	c.Header("Content-Type", "text/event-stream")
+	c.Header("Cache-Control", "no-cache")
+	c.Header("X-Accel-Buffering", "no")
+	s := &eventStream{c: c, rc: http.NewResponseController(c.Writer)}
+	// Probe once here rather than reporting per frame: a writer that can't take
+	// deadlines will fail identically on every write, and the operator needs to
+	// hear it once. If this fails the stream still works — it is just back to
+	// being cut at WriteTimeout, which is worth saying out loud.
+	if err := s.rc.SetWriteDeadline(time.Now().Add(sseWriteTimeout)); err != nil {
+		slog.Error("api: SSE write deadlines unavailable; long turns will be truncated at the server WriteTimeout", "error", err)
+	}
+	c.Writer.Flush()
+	return s
+}
+
+func (s *eventStream) send(ev chatEvent) {
+	b, err := json.Marshal(ev)
+	if err != nil {
+		slog.Error("api: encode chat event", "error", err)
+		return
+	}
+	s.write(fmt.Sprintf("data: %s\n\n", b))
+}
+
+func (s *eventStream) write(frame string) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	// Refresh for THIS write, so the stream as a whole is unbounded but no single
+	// write is. Error deliberately unchecked: openEventStream already reported
+	// whether deadlines work at all, and this call can only fail the same way, so
+	// checking here would log once per frame to say the same thing.
+	_ = s.rc.SetWriteDeadline(time.Now().Add(sseWriteTimeout))
+	_, _ = io.WriteString(s.c.Writer, frame)
+	s.c.Writer.Flush()
+}
+
+// keepAlive writes an SSE comment frame on an interval until the returned
+// function is called, so a long silence while the model thinks doesn't look like
+// a dead connection to whatever sits in between. SSE ignores comment frames, so
+// this costs the client nothing.
+func (s *eventStream) keepAlive(every time.Duration) func() {
+	done := make(chan struct{})
+	stopped := make(chan struct{})
+	go func() {
+		defer close(stopped)
+		t := time.NewTicker(every)
+		defer t.Stop()
+		for {
+			select {
+			case <-done:
+				return
+			case <-s.c.Request.Context().Done():
+				return
+			case <-t.C:
+				s.write(": keep-alive\n\n")
+			}
+		}
+	}()
+	// Idempotent: the handler stops it explicitly when the run returns and again
+	// via defer, so a panic can't leak the goroutine.
+	var once sync.Once
+	return func() {
+		once.Do(func() { close(done) })
+		<-stopped
+	}
+}
+
+// getAgentHistory returns the actor's thread for a garden.
+func (h *handlers) getAgentHistory(c *gin.Context) {
+	gardenID, ok := parseIDParam(c, "id")
+	if !ok {
+		return
+	}
+	msgs, err := h.svc.AgentHistory(c.Request.Context(), mustActor(c).ID, gardenID)
+	if err != nil {
+		writeServiceError(c, err)
+		return
+	}
+	c.JSON(http.StatusOK, gin.H{"messages": msgs})
+}
+
+// deleteAgentHistory is the "start over" escape hatch.
+func (h *handlers) deleteAgentHistory(c *gin.Context) {
+	gardenID, ok := parseIDParam(c, "id")
+	if !ok {
+		return
+	}
+	if err := h.svc.ClearAgentHistory(c.Request.Context(), mustActor(c).ID, gardenID); err != nil {
+		writeServiceError(c, err)
+		return
+	}
+	c.Status(http.StatusNoContent)
+}
+
+// replayHistory turns stored text into model messages.
+//
+// Only the text is replayed — no stored tool calls. Continuity needs what was
+// said and what came back; replaying a tool call would be replaying a decision
+// made against a garden that has since moved on.
+func replayHistory(msgs []domain.AgentMessage) []llm.Message {
+	out := make([]llm.Message, 0, len(msgs))
+	for _, m := range msgs {
+		if m.Role == domain.AgentRoleUser {
+			out = append(out, llm.UserText(m.Body))
+		} else {
+			out = append(out, llm.AssistantText(m.Body))
+		}
+	}
+	return out
+}
+
+func toolNames(s mdagent.Step) []string {
+	names := make([]string, 0, len(s.Results))
+	for _, r := range s.Results {
+		names = append(names, r.Name)
+	}
+	return names
+}
+
+// chatErrorMessage turns a failure into something worth reading. Permission
+// errors in particular get named: "the agent broke" and "you can only view this
+// garden" want very different reactions.
+func chatErrorMessage(err error) string {
+	switch {
+	case errors.Is(err, domain.ErrForbidden):
+		return "You can only view this garden, so I can't change anything in it."
+	case errors.Is(err, domain.ErrNotFound):
+		return "I can't find that garden."
+	case errors.Is(err, domain.ErrInvalidInput):
+		return "I didn't get a message to work from."
+	case errors.Is(err, io.EOF), errors.Is(err, context.Canceled):
+		return "The connection dropped partway through. Anything I'd already changed is on the canvas, and in History."
+	case errors.Is(err, context.DeadlineExceeded):
+		return "That took too long and I stopped. Anything I'd already changed is on the canvas, and in History."
+	default:
+		slog.Error("api: agent run failed", "error", err)
+		return "Something went wrong talking to the model. Anything I'd already changed is on the canvas, and in History."
+	}
+}

internal/api/agent_holder.go

@@ -0,0 +1,89 @@
+package api
+
+import (
+	"context"
+	"log/slog"
+	"sync"
+	"sync/atomic"
+
+	"gitea.stevedudenhoeffer.com/steve/pansy/internal/agent"
+	"gitea.stevedudenhoeffer.com/steve/pansy/internal/service"
+)
+
+// agentHolder owns the live assistant Runner and lets an admin swap it at
+// runtime when the model settings change (#79).
+//
+// The Runner used to be a plain handlers field set once at boot, with the chat
+// routes registered only when it existed. That made the assistant permanently
+// whatever the environment said at startup. Now the routes are always
+// registered and the Runner lives behind an atomic pointer, so a settings change
+// can turn the assistant on, off, or onto a different model without a restart —
+// and without a data race against in-flight requests reading the pointer.
+//
+// A nil pointer means "no assistant right now"; handlers nil-check get() rather
+// than assuming a Runner is present.
+type agentHolder struct {
+	svc *service.Service
+
+	ptr atomic.Pointer[agent.Runner]
+	// rebuildMu serializes rebuilds so two concurrent settings saves can't
+	// interleave into a torn "resolve A, resolve B, store A, store B" swap. The
+	// read path (get) stays lock-free on the atomic pointer.
+	rebuildMu sync.Mutex
+}
+
+// newAgentHolder builds the holder and resolves the initial Runner from whatever
+// the settings + environment currently say. A resolution failure is logged and
+// left as "no assistant", never fatal: a garden planner must still boot.
+func newAgentHolder(ctx context.Context, svc *service.Service) *agentHolder {
+	h := &agentHolder{svc: svc}
+	h.rebuild(ctx)
+	return h
+}
+
+// get returns the current Runner, or nil if the assistant is off.
+func (h *agentHolder) get() *agent.Runner { return h.ptr.Load() }
+
+// rebuild resolves the effective agent configuration and swaps the Runner to
+// match: a new one when the assistant should be on, nil when it shouldn't. It is
+// safe to call at boot and from a settings save; concurrent calls serialize.
+//
+// It logs what it did rather than returning an error, because every caller wants
+// the same thing — best-effort apply, keep serving either way — and a settings
+// save must not fail just because the new model won't resolve. The save already
+// validated the spec; a rebuild failure here means the environment changed under
+// it, and "assistant off, with a reason in the log" is the right outcome.
+func (h *agentHolder) rebuild(ctx context.Context) {
+	h.rebuildMu.Lock()
+	defer h.rebuildMu.Unlock()
+
+	eff, err := h.svc.EffectiveAgent(ctx)
+	if err != nil {
+		// EffectiveAgent errors only if the settings row can't be read — a DB fault,
+		// and a very transient one when it happens right after a settings write. We
+		// keep the current Runner rather than tear down a working assistant on a
+		// blip: the state is already persisted, so the next rebuild (any later save,
+		// or a restart) reconciles it. Loud, because a persistent failure here means
+		// the live assistant no longer matches stored settings.
+		slog.Error("api: could not resolve agent settings; leaving the assistant as-is", "error", err)
+		return
+	}
+	if !eff.Ready() {
+		if h.ptr.Swap(nil) != nil {
+			slog.Info("api: garden assistant turned off",
+				"enabled", eff.Enabled, "hasKey", eff.APIKey != "", "hasModel", eff.Model != "")
+		}
+		return
+	}
+	runner, err := agent.NewRunner(h.svc, eff.APIKey, eff.Model)
+	if err != nil {
+		// Configured but unusable. Turn the assistant off rather than leaving a
+		// stale Runner on the old model — an admin who just pointed it at a broken
+		// spec should see it stop, not silently keep answering on the previous one.
+		slog.Error("api: garden assistant disabled (model won't resolve)", "error", err, "model", eff.Model)
+		h.ptr.Store(nil)
+		return
+	}
+	h.ptr.Store(runner)
+	slog.Info("api: garden assistant ready", "model", eff.Model)
+}

internal/api/agent_test.go

@@ -0,0 +1,49 @@
+package api
+
+import (
+	"net/http"
+	"strconv"
+	"testing"
+)
+
+// TestAgentDisabledWithoutAKey — an instance with no API key must start, serve
+// the app, and not offer the assistant.
+//
+// The contract CHANGED with #79: the chat route is now always registered (so a
+// settings change can turn the assistant on without a restart), so "off" is a
+// runtime 503 rather than a missing route. capabilities reports agent:false, and
+// the frontend keys the chat tab off that — so a user never reaches the 503.
+func TestAgentDisabledWithoutAKey(t *testing.T) {
+	r := authEngine(t, localCfg()) // localCfg has no agent configuration
+	cookie := registerAndCookie(t, r, "[email protected]")
+	gid := createGardenAPI(t, r, cookie, "G")
+
+	// Chat is refused, plainly, because there is no Runner to run.
+	w := doJSON(t, r, http.MethodPost, "/api/v1/agent/chat",
+		map[string]any{"gardenId": gid, "message": "plant garlic"}, cookie)
+	if w.Code != http.StatusServiceUnavailable {
+		t.Errorf("chat without a key: status %d, want 503", w.Code)
+	}
+
+	// Capabilities advertises the assistant as unavailable, which is what the UI
+	// actually consults.
+	w = doJSON(t, r, http.MethodGet, "/api/v1/capabilities", nil, cookie)
+	if w.Code != http.StatusOK {
+		t.Fatalf("capabilities: status %d", w.Code)
+	}
+	if agent, _ := decodeMap(t, w.Body.Bytes())["agent"].(bool); agent {
+		t.Error("capabilities reported agent:true with no key")
+	}
+
+	// History is just stored data behind the ordinary garden-role check, so it
+	// reads fine (empty) whether or not a Runner exists — it isn't gated on one.
+	path := "/api/v1/gardens/" + strconv.FormatInt(gid, 10) + "/agent/history"
+	if w := doJSON(t, r, http.MethodGet, path, nil, cookie); w.Code != http.StatusOK {
+		t.Errorf("history without a key: status %d, want 200 (it's data, not the model)", w.Code)
+	}
+
+	// And the rest of the app is entirely unaffected.
+	if w := doJSON(t, r, http.MethodGet, fullPath(gid), nil, cookie); w.Code != http.StatusOK {
+		t.Errorf("editor load: status %d, want 200 — an unconfigured agent must not break the app", w.Code)
+	}
+}

internal/api/api.go

@@ -6,6 +6,7 @@
 package api
 
 import (
+	"context"
 	"log/slog"
 	"net/http"
 
@@ -22,6 +23,12 @@ type handlers struct {
 	cfg  *config.Config
 	svc  *service.Service
 	oidc *oidcClient // nil unless OIDC is configured (see config.OIDCReady)
+	// agent holds the live Runner behind an atomic pointer. Unlike oidc it is
+	// never nil — the holder is always present and its Runner may be nil when the
+	// assistant is off. The chat routes are registered unconditionally and
+	// nil-check agent.get(), so a settings change can turn the assistant on or off
+	// at runtime (#79) rather than only at boot.
+	agent *agentHolder
 }
 
 // New builds the gin engine with the standard middleware stack and registers the
@@ -49,6 +56,10 @@ func New(cfg *config.Config, svc *service.Service) *gin.Engine {
 	// is set; see csrfGuard).
 	v1.Use(h.csrfGuard())
 	v1.GET("/healthz", healthz)
+	// What this instance can actually do, so the UI offers only what works.
+	// Registered after the agent below, because it reports whether the runner
+	// actually built — not merely whether it was configured to.
+	v1.GET("/capabilities", h.capabilities)
 
 	// Auth endpoints are exempt from requireAuth (you can't be logged in yet);
 	// /me is the one that needs a session. Feature routers in later issues attach
@@ -111,6 +122,11 @@ func New(cfg *config.Config, svc *service.Service) *gin.Engine {
 	objects.PATCH("/:id", h.updateObject)
 	objects.DELETE("/:id", h.deleteObject)
 	objects.POST("/:id/plantings", h.createPlanting) // place a plop in this object
+	// Bulk ops. These wrap the same service methods the agent tools call, so an
+	// instance with no model configured still gets the most valuable operation in
+	// the app — and so "clear bed" is ONE change set rather than one per plop.
+	objects.POST("/:id/fill", h.fillObject)
+	objects.POST("/:id/clear", h.clearObject)
 
 	// Plantings ("plops") are addressed by their own id; the service resolves the
 	// owning object/garden for the permission check.
@@ -118,6 +134,31 @@ func New(cfg *config.Config, svc *service.Service) *gin.Engine {
 	plantings.PATCH("/:id", h.updatePlanting)
 	plantings.DELETE("/:id", h.deletePlanting)
 
+	// The garden assistant. Its routes are registered UNCONDITIONALLY and the live
+	// Runner sits behind an atomic pointer in the holder, so a settings change can
+	// turn the assistant on or off at runtime (#79). Each handler nil-checks
+	// agent.get(); a chat request while the assistant is off gets a clean 503
+	// (AGENT_DISABLED), not a panic and not a missing route.
+	//
+	// The holder resolves its initial Runner from settings + environment at
+	// construction. If the key never reaches the container, the assistant is off
+	// and the reason is logged below — the same operability need #72 added.
+	h.agent = newAgentHolder(context.Background(), svc)
+	if cfg.Agent.OllamaCloudAPIKey == "" {
+		slog.Info("api: garden assistant has no API key",
+			"hint", "set OLLAMA_CLOUD_API_KEY in the container's environment (not just the stack's); the model can be chosen in Settings")
+	}
+	agentGroup := v1.Group("/agent", h.requireAuth())
+	agentGroup.POST("/chat", h.agentChat)
+	gardens.GET("/:id/agent/history", h.getAgentHistory)
+	gardens.DELETE("/:id/agent/history", h.deleteAgentHistory)
+
+	// Instance settings: admin-only, and the first thing to enforce is_admin.
+	// requireAdmin runs after requireAuth (it reads the actor requireAuth stored).
+	settings := v1.Group("/settings", h.requireAuth(), h.requireAdmin())
+	settings.GET("", h.getSettings)
+	settings.PATCH("", h.updateSettings)
+
 	// Undo. A change set is addressed by its own id; the service resolves the
 	// owning garden for the permission check, same as objects and plantings.
 	changeSets := v1.Group("/change-sets", h.requireAuth())
@@ -155,6 +196,17 @@ func New(cfg *config.Config, svc *service.Service) *gin.Engine {
 	return r
 }
 
+// capabilities reports what this instance can actually do, so the UI offers only
+// what works.
+//
+// It reports whether the assistant is live RIGHT NOW, not merely configured:
+// the chat routes always exist, but a request while the Runner is nil is refused,
+// so offering the tab must track the live Runner. Reading agent.get() (an atomic
+// load) means this reflects a settings-driven swap on the very next poll.
+func (h *handlers) capabilities(c *gin.Context) {
+	c.JSON(http.StatusOK, gin.H{"agent": h.agent.get() != nil})
+}
+
 // healthz is a liveness probe: always returns {"ok": true} when the server is up.
 func healthz(c *gin.Context) {
 	c.JSON(http.StatusOK, gin.H{"ok": true})

internal/api/ops.go

@@ -0,0 +1,116 @@
+package api
+
+import (
+	"net/http"
+
+	"github.com/gin-gonic/gin"
+
+	"gitea.stevedudenhoeffer.com/steve/pansy/internal/domain"
+	"gitea.stevedudenhoeffer.com/steve/pansy/internal/service"
+)
+
+// Bulk operations on a plantable object (#82): fill a region with one plant, and
+// clear everything out of it.
+//
+// These were reachable only through the agent toolbox until now, which meant the
+// most valuable bulk operation in a garden planner — and the one carrying the
+// most carefully reasoned geometry in the codebase — did not exist at all on an
+// instance with no model configured. They are thin adapters over the same
+// service methods `internal/agent/tools.go` calls, so the permission checks and
+// the one-change-set-per-operation guarantee come along unchanged.
+
+// fillRect is an explicit rectangle in the object's local frame, the alternative
+// to a compass name. A named type (not an inline anonymous struct) to match the
+// rest of internal/api and so it can carry its own validity check.
+type fillRect struct {
+	MinX float64 `json:"minXCm"`
+	MinY float64 `json:"minYCm"`
+	MaxX float64 `json:"maxXCm"`
+	MaxY float64 `json:"maxYCm"`
+}
+
+// degenerate reports whether the rect encloses no area. Such a rect (including
+// the all-zeros an empty `"rect": {}` decodes to) would otherwise slip through
+// and plant a single plop at the object's centre — a surprising result for what
+// is really malformed input.
+func (r fillRect) degenerate() bool {
+	return r.MaxX <= r.MinX || r.MaxY <= r.MinY
+}
+
+// objectFillRequest is the body for POST /objects/:id/fill.
+//
+// A region is given EITHER by compass name ("ne", "south half", "all") or as an
+// explicit rect in the object's local frame. The named form is what a person
+// means and what the agent uses; the rect is for a future drag-a-box affordance.
+// Exactly one must be supplied — accepting both and silently preferring one
+// would make a client bug look like a geometry bug.
+type objectFillRequest struct {
+	PlantID int64     `json:"plantId" binding:"required"`
+	Region  string    `json:"region"`
+	Rect    *fillRect `json:"rect"`
+	// SpacingOverrideCM plants tighter or looser than the plant's mature spacing
+	// without editing the catalog entry.
+	SpacingOverrideCM *float64 `json:"spacingOverrideCm"`
+}
+
+func (h *handlers) fillObject(c *gin.Context) {
+	id, ok := parseIDParam(c, "id")
+	if !ok {
+		return
+	}
+	var req objectFillRequest
+	if err := c.ShouldBindJSON(&req); err != nil {
+		writeAPIError(c, http.StatusBadRequest, "INVALID_INPUT", "a plantId and a region are required")
+		return
+	}
+	named, hasRect := req.Region != "", req.Rect != nil
+	if named == hasRect {
+		writeAPIError(c, http.StatusBadRequest, "INVALID_INPUT",
+			`supply exactly one of "region" (e.g. "all", "ne", "south half") or "rect"`)
+		return
+	}
+
+	actor := mustActor(c).ID
+	var (
+		created []domain.Planting
+		err     error
+	)
+	if rect := req.Rect; rect != nil {
+		// Reject a zero-area rect here rather than let it plant one stray plop.
+		// (Binding the pointer to `rect` also keeps the deref visibly guarded,
+		// instead of reading req.Rect.MinX under an invariant from a line above.)
+		if rect.degenerate() {
+			writeAPIError(c, http.StatusBadRequest, "INVALID_INPUT", "rect must enclose a positive area")
+			return
+		}
+		region := service.Region{MinX: rect.MinX, MinY: rect.MinY, MaxX: rect.MaxX, MaxY: rect.MaxY}
+		created, err = h.svc.FillRegion(c.Request.Context(), actor, id, region, req.PlantID, req.SpacingOverrideCM)
+	} else {
+		created, err = h.svc.FillNamedRegion(c.Request.Context(), actor, id, req.Region, req.PlantID, req.SpacingOverrideCM)
+	}
+	if err != nil {
+		writeServiceError(c, err)
+		return
+	}
+	// 200, not 201: a fill can legitimately create nothing (the region is already
+	// planted), and there is no single resource to point a Location at.
+	c.JSON(http.StatusOK, gin.H{"plantings": created, "created": len(created)})
+}
+
+// clearObject soft-removes every active plop in an object.
+//
+// Distinct from deleting the object, and — unlike the client-side loop this
+// replaces — it lands as ONE change set, so undoing a cleared bed is one click
+// rather than one per plop.
+func (h *handlers) clearObject(c *gin.Context) {
+	id, ok := parseIDParam(c, "id")
+	if !ok {
+		return
+	}
+	n, err := h.svc.ClearObject(c.Request.Context(), mustActor(c).ID, id)
+	if err != nil {
+		writeServiceError(c, err)
+		return
+	}
+	c.JSON(http.StatusOK, gin.H{"cleared": n})
+}

internal/api/ops_test.go

@@ -0,0 +1,226 @@
+package api
+
+import (
+	"net/http"
+	"testing"
+
+	"github.com/gin-gonic/gin"
+)
+
+func fillPath(id int64) string  { return objectPath(id) + "/fill" }
+func clearPath(id int64) string { return objectPath(id) + "/clear" }
+
+// makeFillPlant creates a custom plant and returns its id. (A near-identical
+// createPlantAPI landed alongside the seed-lot tests; consolidating the two into
+// one shared helper is a fine follow-up, kept separate here only to avoid a
+// merge collision on the shared symbol.)
+func makeFillPlant(t *testing.T, r *gin.Engine, cookie *http.Cookie, name string, spacing float64) int64 {
+	t.Helper()
+	w := doJSON(t, r, http.MethodPost, "/api/v1/plants", map[string]any{
+		"name": name, "category": "vegetable", "spacingCm": spacing, "color": "#4a7c3f", "icon": "🌱",
+	}, cookie)
+	if w.Code != http.StatusCreated {
+		t.Fatalf("create plant %q: status %d, body %s", name, w.Code, w.Body.String())
+	}
+	return int64(decodeMap(t, w.Body.Bytes())["id"].(float64))
+}
+
+// seedFillableBed makes a garden with one plantable bed and a custom plant,
+// returning (gardenID, objectID, plantID).
+func seedFillableBed(t *testing.T, r *gin.Engine, cookie *http.Cookie, w, h, spacing float64) (int64, int64, int64) {
+	t.Helper()
+	gid := createGardenAPI(t, r, cookie, "G")
+	rec := doJSON(t, r, http.MethodPost, objectsPath(gid), map[string]any{
+		"kind": "bed", "widthCm": w, "heightCm": h, "plantable": true,
+	}, cookie)
+	if rec.Code != http.StatusCreated {
+		t.Fatalf("create bed: status %d, body %s", rec.Code, rec.Body.String())
+	}
+	objID := int64(decodeMap(t, rec.Body.Bytes())["id"].(float64))
+	plantID := makeFillPlant(t, r, cookie, "Fillable", spacing)
+	return gid, objID, plantID
+}
+
+// countChangeSets reads the history page and reports how many change sets exist.
+func countChangeSets(t *testing.T, r *gin.Engine, cookie *http.Cookie, gardenID int64) int {
+	t.Helper()
+	w := doJSON(t, r, http.MethodGet, historyPath(gardenID), nil, cookie)
+	if w.Code != http.StatusOK {
+		t.Fatalf("history: status %d, body %s", w.Code, w.Body.String())
+	}
+	sets, _ := decodeMap(t, w.Body.Bytes())["changeSets"].([]any)
+	return len(sets)
+}
+
+// TestFillAndClearAPI covers the two routes end to end through the router.
+//
+// These exist because both operations were previously reachable ONLY through the
+// agent toolbox, so on an instance with no model configured the most valuable
+// bulk operation in the app did not exist at all.
+func TestFillAndClearAPI(t *testing.T) {
+	r := authEngine(t, localCfg())
+	cookie := registerAndCookie(t, r, "[email protected]")
+	_, objID, plantID := seedFillableBed(t, r, cookie, 200, 200, 20)
+
+	// Fill by compass name.
+	w := doJSON(t, r, http.MethodPost, fillPath(objID), map[string]any{
+		"plantId": plantID, "region": "all",
+	}, cookie)
+	if w.Code != http.StatusOK {
+		t.Fatalf("fill: status %d, body %s", w.Code, w.Body.String())
+	}
+	body := decodeMap(t, w.Body.Bytes())
+	created := int(body["created"].(float64))
+	if created == 0 {
+		t.Fatalf("fill created nothing: %s", w.Body.String())
+	}
+	if plops, _ := body["plantings"].([]any); len(plops) != created {
+		t.Errorf("created=%d but returned %d plantings", created, len(plops))
+	}
+
+	// Clear it: one call, and it reports what it removed.
+	w = doJSON(t, r, http.MethodPost, clearPath(objID), nil, cookie)
+	if w.Code != http.StatusOK {
+		t.Fatalf("clear: status %d, body %s", w.Code, w.Body.String())
+	}
+	if n := int(decodeMap(t, w.Body.Bytes())["cleared"].(float64)); n != created {
+		t.Errorf("cleared %d, want %d (everything the fill made)", n, created)
+	}
+
+	// Clearing an already-empty bed is a no-op, not an error.
+	w = doJSON(t, r, http.MethodPost, clearPath(objID), nil, cookie)
+	if w.Code != http.StatusOK {
+		t.Fatalf("second clear: status %d", w.Code)
+	}
+	if n := int(decodeMap(t, w.Body.Bytes())["cleared"].(float64)); n != 0 {
+		t.Errorf("second clear removed %d, want 0", n)
+	}
+}
+
+// TestFillRegionSelectionAPI: exactly one of region/rect, and a rect fills only
+// its own corner of the bed.
+func TestFillRegionSelectionAPI(t *testing.T) {
+	r := authEngine(t, localCfg())
+	cookie := registerAndCookie(t, r, "[email protected]")
+	_, objID, plantID := seedFillableBed(t, r, cookie, 400, 400, 20)
+
+	// Neither → 400. Both → 400. Accepting both and silently preferring one
+	// would make a client bug look like a geometry bug.
+	if w := doJSON(t, r, http.MethodPost, fillPath(objID), map[string]any{"plantId": plantID}, cookie); w.Code != http.StatusBadRequest {
+		t.Errorf("no region = %d, want 400", w.Code)
+	}
+	both := map[string]any{
+		"plantId": plantID, "region": "all",
+		"rect": map[string]any{"minXCm": -50, "minYCm": -50, "maxXCm": 50, "maxYCm": 50},
+	}
+	if w := doJSON(t, r, http.MethodPost, fillPath(objID), both, cookie); w.Code != http.StatusBadRequest {
+		t.Errorf("both region and rect = %d, want 400", w.Code)
+	}
+	// An unknown compass name is rejected rather than silently filling nothing.
+	if w := doJSON(t, r, http.MethodPost, fillPath(objID), map[string]any{
+		"plantId": plantID, "region": "middle-ish",
+	}, cookie); w.Code != http.StatusBadRequest {
+		t.Errorf("bad region name = %d, want 400", w.Code)
+	}
+
+	// A zero-area rect is malformed input, not "plant one at the centre". An empty
+	// `"rect": {}` decodes to all-zeros and must be caught the same way.
+	for _, rect := range []map[string]any{
+		{}, // {} → 0,0,0,0
+		{"minXCm": 10, "minYCm": 10, "maxXCm": 10, "maxYCm": 50}, // zero width
+		{"minXCm": 10, "minYCm": 50, "maxXCm": 50, "maxYCm": 50}, // zero height
+		{"minXCm": 50, "minYCm": 50, "maxXCm": 10, "maxYCm": 10}, // inverted
+	} {
+		if w := doJSON(t, r, http.MethodPost, fillPath(objID),
+			map[string]any{"plantId": plantID, "rect": rect}, cookie); w.Code != http.StatusBadRequest {
+			t.Errorf("degenerate rect %v = %d, want 400", rect, w.Code)
+		}
+	}
+
+	// A rect confined to the NE corner produces plops only there. Local frame:
+	// +x east, -y north.
+	w := doJSON(t, r, http.MethodPost, fillPath(objID), map[string]any{
+		"plantId": plantID,
+		"rect":    map[string]any{"minXCm": 0, "minYCm": -200, "maxXCm": 200, "maxYCm": 0},
+	}, cookie)
+	if w.Code != http.StatusOK {
+		t.Fatalf("rect fill: status %d, body %s", w.Code, w.Body.String())
+	}
+	plops, _ := decodeMap(t, w.Body.Bytes())["plantings"].([]any)
+	if len(plops) == 0 {
+		t.Fatal("rect fill created nothing")
+	}
+	for _, raw := range plops {
+		p := raw.(map[string]any)
+		if x, y := p["xCm"].(float64), p["yCm"].(float64); x < 0 || y > 0 {
+			t.Errorf("plop at (%v,%v) outside the NE rect", x, y)
+		}
+	}
+}
+
+// TestClearObjectIsOneChangeSetAPI is the regression test for the behaviour this
+// endpoint exists to restore.
+//
+// The UI used to clear a bed with a loop of PATCHes, and since every service
+// mutation auto-scopes its own change set, clearing a 40-plop bed wrote 40 of
+// them — 40 presses of Undo to put the bed back. CLAUDE.md states the rule
+// directly: multi-row operations record together so they undo as one unit.
+func TestClearObjectIsOneChangeSetAPI(t *testing.T) {
+	r := authEngine(t, localCfg())
+	cookie := registerAndCookie(t, r, "[email protected]")
+	gid, objID, plantID := seedFillableBed(t, r, cookie, 300, 300, 20)
+
+	w := doJSON(t, r, http.MethodPost, fillPath(objID), map[string]any{
+		"plantId": plantID, "region": "all",
+	}, cookie)
+	if w.Code != http.StatusOK {
+		t.Fatalf("fill: status %d, body %s", w.Code, w.Body.String())
+	}
+	created := int(decodeMap(t, w.Body.Bytes())["created"].(float64))
+	if created < 4 {
+		t.Fatalf("need several plops to make this meaningful, got %d", created)
+	}
+
+	before := countChangeSets(t, r, cookie, gid)
+	if w := doJSON(t, r, http.MethodPost, clearPath(objID), nil, cookie); w.Code != http.StatusOK {
+		t.Fatalf("clear: status %d", w.Code)
+	}
+	if after := countChangeSets(t, r, cookie, gid); after != before+1 {
+		t.Errorf("clearing %d plops added %d change sets, want exactly 1", created, after-before)
+	}
+}
+
+// TestFillClearPermissionsAPI: a viewer may look but not fill or clear, and a
+// stranger gets 404 because existence is masked.
+func TestFillClearPermissionsAPI(t *testing.T) {
+	r := authEngine(t, localCfg())
+	owner := registerAndCookie(t, r, "[email protected]")
+	viewer := registerAndCookie(t, r, "[email protected]")
+	stranger := registerAndCookie(t, r, "[email protected]")
+
+	gid, objID, plantID := seedFillableBed(t, r, owner, 200, 200, 20)
+	if w := doJSON(t, r, http.MethodPost, sharesPath(gid),
+		map[string]any{"email": "[email protected]", "role": "viewer"}, owner); w.Code != http.StatusCreated {
+		t.Fatalf("share as viewer: status %d, body %s", w.Code, w.Body.String())
+	}
+
+	fillBody := map[string]any{"plantId": plantID, "region": "all"}
+	if w := doJSON(t, r, http.MethodPost, fillPath(objID), fillBody, viewer); w.Code != http.StatusForbidden {
+		t.Errorf("viewer fill = %d, want 403 (they can see it but may not do that)", w.Code)
+	}
+	if w := doJSON(t, r, http.MethodPost, clearPath(objID), nil, viewer); w.Code != http.StatusForbidden {
+		t.Errorf("viewer clear = %d, want 403", w.Code)
+	}
+	if w := doJSON(t, r, http.MethodPost, fillPath(objID), fillBody, stranger); w.Code != http.StatusNotFound {
+		t.Errorf("stranger fill = %d, want 404 (existence masked)", w.Code)
+	}
+	if w := doJSON(t, r, http.MethodPost, clearPath(objID), nil, stranger); w.Code != http.StatusNotFound {
+		t.Errorf("stranger clear = %d, want 404", w.Code)
+	}
+	if w := doJSON(t, r, http.MethodPost, fillPath(objID), fillBody, nil); w.Code != http.StatusUnauthorized {
+		t.Errorf("anonymous fill = %d, want 401", w.Code)
+	}
+	if w := doJSON(t, r, http.MethodPost, clearPath(objID), nil, nil); w.Code != http.StatusUnauthorized {
+		t.Errorf("anonymous clear = %d, want 401", w.Code)
+	}
+}

internal/api/seed_lots_test.go

@@ -0,0 +1,264 @@
+package api
+
+import (
+	"encoding/json"
+	"net/http"
+	"strconv"
+	"testing"
+
+	"github.com/gin-gonic/gin"
+)
+
+func seedLotPath(id int64) string {
+	return "/api/v1/seed-lots/" + strconv.FormatInt(id, 10)
+}
+
+// decodeList decodes a bare JSON array body. Seed lot listing returns the array
+// directly rather than wrapping it (unlike /journal's {"entries": …}), so a
+// helper that assumed an object would quietly read nothing.
+func decodeList(t *testing.T, body []byte) []any {
+	t.Helper()
+	var out []any
+	if err := json.Unmarshal(body, &out); err != nil {
+		t.Fatalf("decode list: %v (%s)", err, body)
+	}
+	return out
+}
+
+// createPlantAPI makes a custom plant and returns its id.
+func createPlantAPI(t *testing.T, r *gin.Engine, cookie *http.Cookie, name string, spacing float64) int64 {
+	t.Helper()
+	w := doJSON(t, r, http.MethodPost, "/api/v1/plants", map[string]any{
+		"name": name, "category": "vegetable", "spacingCm": spacing, "color": "#4a7c3f", "icon": "🌱",
+	}, cookie)
+	if w.Code != http.StatusCreated {
+		t.Fatalf("create plant %q: status %d, body %s", name, w.Code, w.Body.String())
+	}
+	return int64(decodeMap(t, w.Body.Bytes())["id"].(float64))
+}
+
+// TestSeedLotCrudAPI walks the whole seed lot lifecycle over HTTP.
+//
+// It exists for the reason CLAUDE.md gives: service tests cannot see a route
+// that was never registered, or one registered with the wrong :param name. Every
+// other handler file had a sibling API test; this group did not, which is the
+// state PATCH/DELETE /journal/:id shipped in — implemented, unit-tested, and
+// completely unreachable.
+func TestSeedLotCrudAPI(t *testing.T) {
+	r := authEngine(t, localCfg())
+	cookie := registerAndCookie(t, r, "[email protected]")
+	plantID := createPlantAPI(t, r, cookie, "Music Garlic", 15)
+
+	// Create.
+	w := doJSON(t, r, http.MethodPost, "/api/v1/seed-lots", map[string]any{
+		"plantId": plantID, "vendor": "Johnny's", "sku": "2761",
+		"quantity": 100, "unit": "seeds", "packedForYear": 2026, "costCents": 495,
+	}, cookie)
+	if w.Code != http.StatusCreated {
+		t.Fatalf("create: status %d, body %s", w.Code, w.Body.String())
+	}
+	lot := decodeMap(t, w.Body.Bytes())
+	id := int64(lot["id"].(float64))
+	if lot["vendor"] != "Johnny's" || lot["unit"] != "seeds" {
+		t.Errorf("unexpected lot: %+v", lot)
+	}
+
+	// GET by id — the route most likely to be missing or mis-registered.
+	w = doJSON(t, r, http.MethodGet, seedLotPath(id), nil, cookie)
+	if w.Code != http.StatusOK {
+		t.Fatalf("get: status %d, body %s", w.Code, w.Body.String())
+	}
+	if got := decodeMap(t, w.Body.Bytes()); int64(got["id"].(float64)) != id {
+		t.Errorf("get returned id %v, want %d", got["id"], id)
+	}
+
+	// List, and the ?plantId= filter.
+	w = doJSON(t, r, http.MethodGet, "/api/v1/seed-lots", nil, cookie)
+	if w.Code != http.StatusOK {
+		t.Fatalf("list: status %d, body %s", w.Code, w.Body.String())
+	}
+	if n := len(decodeList(t, w.Body.Bytes())); n != 1 {
+		t.Fatalf("list returned %d lots, want 1: %s", n, w.Body.String())
+	}
+
+	other := createPlantAPI(t, r, cookie, "Cherokee Purple", 45)
+	w = doJSON(t, r, http.MethodGet, "/api/v1/seed-lots?plantId="+strconv.FormatInt(other, 10), nil, cookie)
+	if w.Code != http.StatusOK {
+		t.Fatalf("filtered list: status %d, body %s", w.Code, w.Body.String())
+	}
+	if n := len(decodeList(t, w.Body.Bytes())); n != 0 {
+		t.Errorf("filter by a plant with no lots returned %d, want 0", n)
+	}
+	// A bad plantId filter is 400, whether non-numeric or out of range — the
+	// handler rejects id < 1, not just unparseable strings.
+	for _, bad := range []string{"nope", "0", "-1"} {
+		if w := doJSON(t, r, http.MethodGet, "/api/v1/seed-lots?plantId="+bad, nil, cookie); w.Code != http.StatusBadRequest {
+			t.Errorf("plantId=%q filter = %d, want 400", bad, w.Code)
+		}
+	}
+
+	// PATCH with the current version.
+	w = doJSON(t, r, http.MethodPatch, seedLotPath(id), map[string]any{
+		"vendor": "Fedco", "version": lot["version"],
+	}, cookie)
+	if w.Code != http.StatusOK {
+		t.Fatalf("patch: status %d, body %s", w.Code, w.Body.String())
+	}
+	updated := decodeMap(t, w.Body.Bytes())
+	if updated["vendor"] != "Fedco" {
+		t.Errorf("vendor = %v, want Fedco", updated["vendor"])
+	}
+	if updated["version"].(float64) != lot["version"].(float64)+1 {
+		t.Errorf("patch didn't bump version: %v -> %v", lot["version"], updated["version"])
+	}
+
+	// A stale version conflicts and carries the current row back, so the client
+	// can rebase without a second request.
+	w = doJSON(t, r, http.MethodPatch, seedLotPath(id), map[string]any{
+		"vendor": "stale", "version": lot["version"],
+	}, cookie)
+	if w.Code != http.StatusConflict {
+		t.Fatalf("stale patch: status %d, want 409", w.Code)
+	}
+	if cur, ok := decodeMap(t, w.Body.Bytes())["current"].(map[string]any); !ok || cur["vendor"] != "Fedco" {
+		t.Errorf("409 body missing the current row: %s", w.Body.String())
+	}
+
+	// DELETE.
+	if w := doJSON(t, r, http.MethodDelete, seedLotPath(id), nil, cookie); w.Code != http.StatusNoContent {
+		t.Fatalf("delete: status %d, body %s", w.Code, w.Body.String())
+	}
+	if w := doJSON(t, r, http.MethodGet, seedLotPath(id), nil, cookie); w.Code != http.StatusNotFound {
+		t.Errorf("get after delete = %d, want 404", w.Code)
+	}
+}
+
+// TestSeedLotRemainingIsDerivedAPI checks that `remaining` reflects plantings
+// through the HTTP surface, not just in the service.
+//
+// DESIGN.md makes derivation load-bearing — "a decremented column drifts the
+// moment a planting is edited behind its back" — so a route that returned a
+// stored or stale figure would break the invariant silently, and the number is
+// the whole reason anyone opens the seed shelf.
+func TestSeedLotRemainingIsDerivedAPI(t *testing.T) {
+	r := authEngine(t, localCfg())
+	cookie := registerAndCookie(t, r, "[email protected]")
+	plantID := createPlantAPI(t, r, cookie, "Beans", 10)
+
+	w := doJSON(t, r, http.MethodPost, "/api/v1/seed-lots", map[string]any{
+		"plantId": plantID, "quantity": 50, "unit": "seeds",
+	}, cookie)
+	if w.Code != http.StatusCreated {
+		t.Fatalf("create lot: status %d, body %s", w.Code, w.Body.String())
+	}
+	lotID := int64(decodeMap(t, w.Body.Bytes())["id"].(float64))
+
+	gid := createGardenAPI(t, r, cookie, "G")
+	w = doJSON(t, r, http.MethodPost, objectsPath(gid), map[string]any{
+		"kind": "bed", "widthCm": 100, "heightCm": 100, "plantable": true,
+	}, cookie)
+	if w.Code != http.StatusCreated {
+		t.Fatalf("create object: status %d, body %s", w.Code, w.Body.String())
+	}
+	oid := int64(decodeMap(t, w.Body.Bytes())["id"].(float64))
+
+	// Plant 12 of them against the lot.
+	w = doJSON(t, r, http.MethodPost, objectPlantingsPath(oid), map[string]any{
+		"plantId": plantID, "xCm": 0, "yCm": 0, "radiusCm": 20, "count": 12, "seedLotId": lotID,
+	}, cookie)
+	if w.Code != http.StatusCreated {
+		t.Fatalf("create planting: status %d, body %s", w.Code, w.Body.String())
+	}
+
+	w = doJSON(t, r, http.MethodGet, seedLotPath(lotID), nil, cookie)
+	if w.Code != http.StatusOK {
+		t.Fatalf("get lot: status %d, body %s", w.Code, w.Body.String())
+	}
+	got := decodeMap(t, w.Body.Bytes())
+	if rem, ok := got["remaining"].(float64); !ok || rem != 38 {
+		t.Errorf("remaining = %v, want 38 (50 bought - 12 planted): %s", got["remaining"], w.Body.String())
+	}
+
+	// Over-plant it: 45 more (57 total against 50 bought) drives remaining
+	// NEGATIVE. That's deliberate — the number is a derived truth about what
+	// you've committed, not a floor clamped at zero, and "you've planted more than
+	// you bought" is exactly the signal a gardener wants rather than a hidden -7.
+	w = doJSON(t, r, http.MethodPost, objectPlantingsPath(oid), map[string]any{
+		"plantId": plantID, "xCm": 40, "yCm": 40, "radiusCm": 20, "count": 45, "seedLotId": lotID,
+	}, cookie)
+	if w.Code != http.StatusCreated {
+		t.Fatalf("over-plant: status %d, body %s", w.Code, w.Body.String())
+	}
+	w = doJSON(t, r, http.MethodGet, seedLotPath(lotID), nil, cookie)
+	if rem, ok := decodeMap(t, w.Body.Bytes())["remaining"].(float64); !ok || rem != -7 {
+		t.Errorf("remaining after over-planting = %v, want -7 (50 - 57)", rem)
+	}
+}
+
+// TestSeedLotsArePrivateAPI checks the ACL through the router.
+//
+// Lots are private to the buyer and deliberately never travel with a shared
+// garden, so another user must not be able to read or edit one. Per the
+// project's convention, no-access is ErrNotFound rather than ErrForbidden —
+// existence is masked — so every one of these is a 404, not a 403.
+func TestSeedLotsArePrivateAPI(t *testing.T) {
+	r := authEngine(t, localCfg())
+	alice := registerAndCookie(t, r, "[email protected]")
+	bob := registerAndCookie(t, r, "[email protected]")
+
+	plantID := createPlantAPI(t, r, alice, "Alice's garlic", 15)
+	w := doJSON(t, r, http.MethodPost, "/api/v1/seed-lots", map[string]any{
+		"plantId": plantID, "vendor": "Secret Vendor", "quantity": 10, "unit": "bulbs",
+	}, alice)
+	if w.Code != http.StatusCreated {
+		t.Fatalf("alice create: status %d, body %s", w.Code, w.Body.String())
+	}
+	lotID := int64(decodeMap(t, w.Body.Bytes())["id"].(float64))
+
+	for _, tc := range []struct {
+		name   string
+		method string
+		body   any
+	}{
+		{"get", http.MethodGet, nil},
+		{"patch", http.MethodPatch, map[string]any{"vendor": "hijacked", "version": 1}},
+		{"delete", http.MethodDelete, nil},
+	} {
+		if w := doJSON(t, r, tc.method, seedLotPath(lotID), tc.body, bob); w.Code != http.StatusNotFound {
+			t.Errorf("bob %s = %d, want 404 (existence is masked)", tc.name, w.Code)
+		}
+	}
+
+	// Bob's own listing must not include it either.
+	w = doJSON(t, r, http.MethodGet, "/api/v1/seed-lots", nil, bob)
+	if w.Code != http.StatusOK {
+		t.Fatalf("bob list: status %d", w.Code)
+	}
+	if n := len(decodeList(t, w.Body.Bytes())); n != 0 {
+		t.Errorf("bob sees %d of alice's lots, want 0: %s", n, w.Body.String())
+	}
+
+	// And it's still intact for alice.
+	if w := doJSON(t, r, http.MethodGet, seedLotPath(lotID), nil, alice); w.Code != http.StatusOK {
+		t.Errorf("alice lost access to her own lot: %d", w.Code)
+	}
+}
+
+// TestSeedLotsRequireAuthAPI: the group is behind requireAuth, and an
+// unauthenticated caller gets 401 rather than an empty list.
+func TestSeedLotsRequireAuthAPI(t *testing.T) {
+	r := authEngine(t, localCfg())
+	for _, tc := range []struct {
+		method, path string
+	}{
+		{http.MethodGet, "/api/v1/seed-lots"},
+		{http.MethodPost, "/api/v1/seed-lots"},
+		{http.MethodGet, seedLotPath(1)},
+		{http.MethodPatch, seedLotPath(1)},
+		{http.MethodDelete, seedLotPath(1)},
+	} {
+		if w := doJSON(t, r, tc.method, tc.path, nil, nil); w.Code != http.StatusUnauthorized {
+			t.Errorf("%s %s = %d, want 401", tc.method, tc.path, w.Code)
+		}
+	}
+}

internal/api/settings.go

@@ -0,0 +1,143 @@
+package api
+
+import (
+	"context"
+	"encoding/json"
+	"errors"
+	"net/http"
+
+	"github.com/gin-gonic/gin"
+
+	"gitea.stevedudenhoeffer.com/steve/pansy/internal/domain"
+	"gitea.stevedudenhoeffer.com/steve/pansy/internal/service"
+)
+
+// Instance settings (#79): admin-only, instance-wide. The authoritative admin
+// check is in the service; requireAdmin here is a cheap early 403 that also
+// keeps the route group readable.
+
+// requireAdmin rejects a non-admin actor. It runs after requireAuth, so the
+// actor is already resolved and carries IsAdmin — no extra query. Returns 403
+// (not 404): a logged-in user knows settings exist, they just may not touch them.
+func (h *handlers) requireAdmin() gin.HandlerFunc {
+	return func(c *gin.Context) {
+		if !mustActor(c).IsAdmin {
+			writeAPIError(c, http.StatusForbidden, "FORBIDDEN", "admin access required")
+			c.Abort()
+			return
+		}
+		c.Next()
+	}
+}
+
+// settingsResponse is what GET/PATCH /settings return. It carries the stored
+// settings plus a read-only view of what's resolved and live, so the UI can show
+// "inheriting ollama-cloud/glm-5.2:cloud from the environment" and whether a key
+// is present — without ever exposing the key itself.
+type settingsResponse struct {
+	Settings *domain.InstanceSettings `json:"settings"`
+	// Effective is the configuration actually in force after layering settings
+	// over the environment.
+	Effective effectiveView `json:"effective"`
+}
+
+type effectiveView struct {
+	Model   string `json:"model"`
+	Enabled bool   `json:"enabled"`
+	// HasApiKey reports whether OLLAMA_CLOUD_API_KEY is set. The key itself is
+	// never serialized — an admin may know one exists, not what it is.
+	HasApiKey bool `json:"hasApiKey"`
+	// AgentLive is whether the assistant Runner is actually built right now. It
+	// can be false even when Enabled+HasApiKey are true (an unresolvable model),
+	// which is exactly the case the UI needs to surface.
+	AgentLive bool `json:"agentLive"`
+}
+
+// settingsPayload builds the response, or an error. It does NOT swallow an
+// EffectiveAgent failure into a misleading empty "effective" view — an empty
+// view would report no model and no key, which reads as "nothing configured"
+// rather than "we couldn't read it". Since EffectiveAgent re-reads the same row
+// GetInstanceSettings just returned, a failure here is a genuine DB fault worth
+// surfacing as a 500, not papering over.
+func (h *handlers) settingsPayload(c *gin.Context, st *domain.InstanceSettings) (settingsResponse, error) {
+	eff, err := h.svc.EffectiveAgent(c.Request.Context())
+	if err != nil {
+		return settingsResponse{}, err
+	}
+	return settingsResponse{
+		Settings: st,
+		Effective: effectiveView{
+			Model:     eff.Model,
+			Enabled:   eff.Enabled,
+			HasApiKey: eff.APIKey != "",
+			AgentLive: h.agent.get() != nil,
+		},
+	}, nil
+}
+
+func (h *handlers) getSettings(c *gin.Context) {
+	st, err := h.svc.GetInstanceSettings(c.Request.Context(), mustActor(c).ID)
+	if err != nil {
+		writeServiceError(c, err)
+		return
+	}
+	payload, err := h.settingsPayload(c, st)
+	if err != nil {
+		writeServiceError(c, err)
+		return
+	}
+	c.JSON(http.StatusOK, payload)
+}
+
+// settingsUpdateRequest is the PATCH body. agentModel "" means inherit the env
+// var. agentEnabled is json.RawMessage so an explicit null (inherit) is
+// distinguishable from an absent field and from true/false.
+type settingsUpdateRequest struct {
+	AgentModel   string          `json:"agentModel"`
+	AgentEnabled json.RawMessage `json:"agentEnabled"`
+	Version      int64           `json:"version" binding:"required"`
+}
+
+func (h *handlers) updateSettings(c *gin.Context) {
+	var req settingsUpdateRequest
+	if err := c.ShouldBindJSON(&req); err != nil {
+		writeAPIError(c, http.StatusBadRequest, "INVALID_INPUT", "a current version is required")
+		return
+	}
+
+	// agentEnabled: absent or null → inherit (nil); true/false → explicit override.
+	// The shared parseNullable does exactly this three-way decode; present is
+	// irrelevant here because absent and null both mean "inherit".
+	enabled, _, err := parseNullable[bool](req.AgentEnabled)
+	if err != nil {
+		writeAPIError(c, http.StatusBadRequest, "INVALID_INPUT", "agentEnabled must be true, false, or null")
+		return
+	}
+
+	st, err := h.svc.UpdateInstanceSettings(c.Request.Context(), mustActor(c).ID, service.InstanceSettingsPatch{
+		AgentModel:   req.AgentModel,
+		AgentEnabled: enabled,
+		Version:      req.Version,
+	})
+	if err != nil {
+		if errors.Is(err, domain.ErrVersionConflict) {
+			writeVersionConflict(c, st)
+			return
+		}
+		writeServiceError(c, err)
+		return
+	}
+
+	// Apply the change to the LIVE assistant. Detached from the request context:
+	// the write is committed and the rebuild describes it, so a client that hangs
+	// up now must not leave the running Runner out of step with the stored
+	// settings. Mirrors the same reasoning as the history-write detachment.
+	h.agent.rebuild(context.WithoutCancel(c.Request.Context()))
+
+	payload, err := h.settingsPayload(c, st)
+	if err != nil {
+		writeServiceError(c, err)
+		return
+	}
+	c.JSON(http.StatusOK, payload)
+}

internal/api/settings_test.go

@@ -0,0 +1,208 @@
+package api
+
+import (
+	"net/http"
+	"testing"
+
+	"github.com/gin-gonic/gin"
+
+	"gitea.stevedudenhoeffer.com/steve/pansy/internal/config"
+)
+
+// agentCfg is a config with the assistant configured — a (fake) key, on, and a
+// resolvable model. The key isn't real, but ValidateAgentModel/NewRunner only
+// PARSE the spec (no live call), so a Runner still builds and capabilities
+// reports it live. That's enough to exercise the runtime on/off swap.
+func agentCfg() *config.Config {
+	c := localCfg()
+	c.Agent = config.AgentConfig{
+		Model:             "ollama-cloud/glm-5.2:cloud",
+		OllamaCloudAPIKey: "test-key",
+		Enabled:           true,
+	}
+	return c
+}
+
+func settingsVersion(t *testing.T, r *gin.Engine, cookie *http.Cookie) int64 {
+	t.Helper()
+	w := doJSON(t, r, http.MethodGet, "/api/v1/settings", nil, cookie)
+	if w.Code != http.StatusOK {
+		t.Fatalf("get settings: status %d, body %s", w.Code, w.Body.String())
+	}
+	st, _ := decodeMap(t, w.Body.Bytes())["settings"].(map[string]any)
+	return int64(st["version"].(float64))
+}
+
+// TestSettingsAdminOnly: the first registered user is admin and can read/write
+// settings; a second user is not and gets 403 (not 404 — settings aren't a
+// masked resource).
+func TestSettingsAdminOnly(t *testing.T) {
+	r := authEngine(t, localCfg())
+	admin := registerAndCookie(t, r, "[email protected]") // first user → admin
+	member := registerAndCookie(t, r, "[email protected]")
+
+	if w := doJSON(t, r, http.MethodGet, "/api/v1/settings", nil, admin); w.Code != http.StatusOK {
+		t.Fatalf("admin GET settings: status %d, body %s", w.Code, w.Body.String())
+	}
+
+	if w := doJSON(t, r, http.MethodGet, "/api/v1/settings", nil, member); w.Code != http.StatusForbidden {
+		t.Errorf("member GET settings: status %d, want 403", w.Code)
+	}
+	if w := doJSON(t, r, http.MethodPatch, "/api/v1/settings",
+		map[string]any{"agentModel": "x", "version": 1}, member); w.Code != http.StatusForbidden {
+		t.Errorf("member PATCH settings: status %d, want 403", w.Code)
+	}
+	// Unauthenticated is 401, before the admin check.
+	if w := doJSON(t, r, http.MethodGet, "/api/v1/settings", nil, nil); w.Code != http.StatusUnauthorized {
+		t.Errorf("anonymous GET settings: status %d, want 401", w.Code)
+	}
+}
+
+// TestSettingsInheritFromEnv: an untouched instance reports the env model as
+// effective, and an empty stored model keeps inheriting it.
+func TestSettingsInheritFromEnv(t *testing.T) {
+	r := authEngine(t, agentCfg())
+	admin := registerAndCookie(t, r, "[email protected]")
+
+	w := doJSON(t, r, http.MethodGet, "/api/v1/settings", nil, admin)
+	if w.Code != http.StatusOK {
+		t.Fatalf("get: status %d, body %s", w.Code, w.Body.String())
+	}
+	body := decodeMap(t, w.Body.Bytes())
+	st := body["settings"].(map[string]any)
+	eff := body["effective"].(map[string]any)
+
+	if st["agentModel"] != "" {
+		t.Errorf("stored model = %v, want empty (inherit)", st["agentModel"])
+	}
+	if eff["model"] != "ollama-cloud/glm-5.2:cloud" {
+		t.Errorf("effective model = %v, want the env value", eff["model"])
+	}
+	if eff["hasApiKey"] != true || eff["agentLive"] != true {
+		t.Errorf("effective = %+v, want a key present and the agent live", eff)
+	}
+}
+
+// TestSettingsUpdateSwapsTheRunner is the core of #79: changing settings takes
+// effect on the LIVE assistant, with no restart. Driven entirely through HTTP.
+func TestSettingsUpdateSwapsTheRunner(t *testing.T) {
+	r := authEngine(t, agentCfg())
+	admin := registerAndCookie(t, r, "[email protected]")
+
+	capsAgent := func() bool {
+		w := doJSON(t, r, http.MethodGet, "/api/v1/capabilities", nil, admin)
+		return decodeMap(t, w.Body.Bytes())["agent"] == true
+	}
+
+	// Configured and on out of the box.
+	if !capsAgent() {
+		t.Fatal("assistant should be live at boot with a key + enabled")
+	}
+
+	// Turn it OFF via settings → capabilities flips immediately.
+	v := settingsVersion(t, r, admin)
+	w := doJSON(t, r, http.MethodPatch, "/api/v1/settings",
+		map[string]any{"agentModel": "", "agentEnabled": false, "version": v}, admin)
+	if w.Code != http.StatusOK {
+		t.Fatalf("disable: status %d, body %s", w.Code, w.Body.String())
+	}
+	if capsAgent() {
+		t.Error("assistant still live after being disabled — the runner wasn't swapped")
+	}
+	// Chat now refuses, at runtime, on a route that still exists.
+	gid := createGardenAPI(t, r, admin, "G")
+	if w := doJSON(t, r, http.MethodPost, "/api/v1/agent/chat",
+		map[string]any{"gardenId": gid, "message": "hi"}, admin); w.Code != http.StatusServiceUnavailable {
+		t.Errorf("chat while disabled: status %d, want 503", w.Code)
+	}
+
+	// Turn it back ON, with an explicit model, and confirm it's live again and the
+	// effective model reflects the change.
+	v = settingsVersion(t, r, admin)
+	w = doJSON(t, r, http.MethodPatch, "/api/v1/settings", map[string]any{
+		"agentModel": "ollama-cloud/kimi-k2.6:cloud", "agentEnabled": true, "version": v,
+	}, admin)
+	if w.Code != http.StatusOK {
+		t.Fatalf("re-enable: status %d, body %s", w.Code, w.Body.String())
+	}
+	if eff := decodeMap(t, w.Body.Bytes())["effective"].(map[string]any); eff["model"] != "ollama-cloud/kimi-k2.6:cloud" {
+		t.Errorf("effective model = %v after change, want the new one", eff["model"])
+	}
+	if !capsAgent() {
+		t.Error("assistant not live after being re-enabled")
+	}
+}
+
+// TestSettingsRejectsBadModel: a spec that won't resolve is a 400 at save time,
+// not a broken assistant on the next turn.
+func TestSettingsRejectsBadModel(t *testing.T) {
+	r := authEngine(t, agentCfg())
+	admin := registerAndCookie(t, r, "[email protected]")
+	v := settingsVersion(t, r, admin)
+
+	if w := doJSON(t, r, http.MethodPatch, "/api/v1/settings",
+		map[string]any{"agentModel": "nonesuch/model", "version": v}, admin); w.Code != http.StatusBadRequest {
+		t.Errorf("bad model: status %d, want 400", w.Code)
+	}
+	// agentEnabled must be a bool or null, not a string.
+	if w := doJSON(t, r, http.MethodPatch, "/api/v1/settings",
+		map[string]any{"agentModel": "", "agentEnabled": "yes", "version": v}, admin); w.Code != http.StatusBadRequest {
+		t.Errorf("string agentEnabled: status %d, want 400", w.Code)
+	}
+}
+
+// TestSettingsVersionConflict: a stale version 409s and carries the current row.
+func TestSettingsVersionConflict(t *testing.T) {
+	r := authEngine(t, agentCfg())
+	admin := registerAndCookie(t, r, "[email protected]")
+	v := settingsVersion(t, r, admin)
+
+	// First write succeeds and bumps the version.
+	if w := doJSON(t, r, http.MethodPatch, "/api/v1/settings",
+		map[string]any{"agentModel": "", "agentEnabled": false, "version": v}, admin); w.Code != http.StatusOK {
+		t.Fatalf("first update: status %d, body %s", w.Code, w.Body.String())
+	}
+	// Reusing the old version conflicts.
+	w := doJSON(t, r, http.MethodPatch, "/api/v1/settings",
+		map[string]any{"agentModel": "", "agentEnabled": true, "version": v}, admin)
+	if w.Code != http.StatusConflict {
+		t.Fatalf("stale update: status %d, want 409", w.Code)
+	}
+	if cur, ok := decodeMap(t, w.Body.Bytes())["current"].(map[string]any); !ok || cur["version"].(float64) != float64(v+1) {
+		t.Errorf("409 body missing the current row at the bumped version: %s", w.Body.String())
+	}
+}
+
+// TestSettingsSwapUnderRace runs settings saves concurrently with chat requests,
+// so `go test -race` proves the atomic swap of the live Runner is safe against
+// in-flight readers. The whole point of the atomic.Pointer is this: without it,
+// toggling the assistant while a request reads it is a data race.
+func TestSettingsSwapUnderRace(t *testing.T) {
+	r := authEngine(t, agentCfg())
+	admin := registerAndCookie(t, r, "[email protected]")
+
+	done := make(chan struct{})
+	// Readers hammer capabilities, whose h.agent.get() is the SAME atomic Load the
+	// chat handler does — so this races the pointer read against the writer's swap
+	// without ever invoking the model (a real Run would hit the network on a fake
+	// key). If get() is race-clean here it is race-clean in chat.
+	for i := 0; i < 4; i++ {
+		go func() {
+			for {
+				select {
+				case <-done:
+					return
+				default:
+					doJSON(t, r, http.MethodGet, "/api/v1/capabilities", nil, admin)
+				}
+			}
+		}()
+	}
+	// Writer: flip the assistant on and off, swapping the pointer each time.
+	for i := 0; i < 12; i++ {
+		v := settingsVersion(t, r, admin)
+		doJSON(t, r, http.MethodPatch, "/api/v1/settings",
+			map[string]any{"agentModel": "", "agentEnabled": i%2 == 0, "version": v}, admin)
+	}
+	close(done)
+}

internal/api/sse_deadline_test.go

@@ -0,0 +1,101 @@
+package api
+
+import (
+	"bufio"
+	"net/http/httptest"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/gin-gonic/gin"
+)
+
+// streamFrames spins up a real http.Server with the given WriteTimeout and an
+// SSE handler that emits `frames` data frames, one every `tick`, then returns.
+// It reports how many frames the client actually received and any read error —
+// the only vantage point from which the deadline failures in #78/#87 are
+// visible, since the writes themselves return nil when the bytes are dropped.
+func streamFrames(t *testing.T, serverWriteTimeout, tick time.Duration, frames int) (int, error) {
+	t.Helper()
+	gin.SetMode(gin.TestMode)
+	r := gin.New()
+	r.GET("/stream", func(c *gin.Context) {
+		s := openEventStream(c)
+		for i := 0; i < frames; i++ {
+			time.Sleep(tick)
+			s.send(chatEvent{Error: "frame"})
+		}
+	})
+
+	srv := httptest.NewUnstartedServer(r)
+	srv.Config.WriteTimeout = serverWriteTimeout
+	srv.Start()
+	defer srv.Close()
+
+	resp, err := srv.Client().Get(srv.URL + "/stream")
+	if err != nil {
+		t.Fatalf("get: %v", err)
+	}
+	defer resp.Body.Close()
+
+	got := 0
+	sc := bufio.NewScanner(resp.Body)
+	for sc.Scan() {
+		if strings.HasPrefix(sc.Text(), "data: ") {
+			got++
+		}
+	}
+	return got, sc.Err()
+}
+
+// TestEventStreamOutlivesServerWriteTimeout is the regression test for #78.
+//
+// http.Server.WriteTimeout is an ABSOLUTE deadline measured from when the
+// request header was read — not an idle timeout — so a streaming response is cut
+// once it passes, however recently the handler wrote. pansy sets it to 30s while
+// an agent turn may run for minutes. openEventStream must override it.
+//
+// This has to be asserted from the CLIENT side because the failure cannot be
+// observed from the handler: writes made after the deadline return err == nil
+// and their bytes are silently discarded. A test that checked the return of
+// io.WriteString would pass against the bug.
+func TestEventStreamOutlivesServerWriteTimeout(t *testing.T) {
+	// sseWriteTimeout stays at its 30s default here, so the per-frame refresh
+	// keeps the stream alive with a huge margin — CI slowness only ever makes
+	// this pass more surely. The server's 300ms WriteTimeout is the thing being
+	// overridden; frames straddle it (300ms/600ms/900ms).
+	got, err := streamFrames(t, 300*time.Millisecond, 300*time.Millisecond, 3)
+	if err != nil {
+		t.Errorf("client read error after %d/3 frames: %v", got, err)
+	}
+	if got != 3 {
+		t.Errorf("client received %d frames, want 3 — the stream was cut at the server WriteTimeout", got)
+	}
+}
+
+// TestEventStreamRefreshesDeadlinePerFrame guards the #87 fix specifically: the
+// write deadline is refreshed on EVERY frame, not set once.
+//
+// A set-once deadline is a plausible "simplification" and it reintroduces the
+// unbounded-block risk the per-frame refresh exists to prevent — yet it would
+// sail through the test above, whose whole run is far under sseWriteTimeout. So
+// shrink sseWriteTimeout below the stream's total duration and send frames whose
+// gap stays comfortably under it: per-frame refresh delivers them all, while a
+// deadline set once at open would expire mid-stream and cut it short.
+func TestEventStreamRefreshesDeadlinePerFrame(t *testing.T) {
+	orig := sseWriteTimeout
+	sseWriteTimeout = 400 * time.Millisecond
+	t.Cleanup(func() { sseWriteTimeout = orig })
+
+	// The server WriteTimeout is generous (5s), so it isn't the limiter — the
+	// per-frame sseWriteTimeout is. 8 frames at a 100ms tick span 800ms, well past
+	// the 400ms deadline, but each 100ms gap is a 4× margin under it.
+	got, err := streamFrames(t, 5*time.Second, 100*time.Millisecond, 8)
+	if err != nil {
+		t.Errorf("client read error after %d/8 frames: %v", got, err)
+	}
+	if got != 8 {
+		t.Errorf("client received %d frames, want 8 — a set-once deadline would cut the stream at ~%v; the refresh must be per-frame",
+			got, sseWriteTimeout)
+	}
+}

internal/config/config.go

@@ -19,6 +19,9 @@ const (
 	RegistrationClosed = "closed"
 )
 
+// DefaultAgentModel is the assistant's model when PANSY_AGENT_MODEL is unset.
+const DefaultAgentModel = "ollama-cloud/glm-5.2:cloud"
+
 // Config is the fully-resolved runtime configuration.
 type Config struct {
 	// Port is the TCP port the HTTP server listens on (PANSY_PORT, default 8080).
@@ -37,6 +40,8 @@ type Config struct {
 	LocalAuth bool
 	// OIDC holds the optional OpenID Connect provider settings.
 	OIDC OIDCConfig
+	// Agent holds the garden-assistant settings.
+	Agent AgentConfig
 	// TrustedProxies is the set of proxy CIDRs/IPs gin trusts for client IP
 	// resolution (PANSY_TRUSTED_PROXIES, comma-separated). Empty trusts none.
 	TrustedProxies []string
@@ -51,6 +56,26 @@ type OIDCConfig struct {
 	ButtonLabel  string // PANSY_OIDC_BUTTON_LABEL — login-page button text
 }
 
+// AgentConfig holds the garden assistant's settings.
+type AgentConfig struct {
+	// Model is the majordomo model spec, passed VERBATIM to majordomo.Parse
+	// (PANSY_AGENT_MODEL). The grammar is majordomo's, not pansy's — parsing or
+	// validating it here would only mean two places to update when it grows. A
+	// comma-separated spec is a failover chain, so
+	// "ollama-cloud/glm-5.2:cloud,ollama-cloud/kimi-k2.6:cloud" gets a fallback
+	// for free.
+	Model string
+	// OllamaCloudAPIKey authenticates against Ollama Cloud
+	// (OLLAMA_CLOUD_API_KEY — the same secret name gadfly uses, which is why
+	// it isn't majordomo's own OLLAMA_API_KEY; pansy passes it explicitly rather
+	// than relying on ambient environment).
+	OllamaCloudAPIKey string
+	// Enabled turns the assistant on (PANSY_AGENT_ENABLED). Defaults to on when
+	// a key is present, so an instance with no key starts cleanly and simply
+	// doesn't offer the agent — the same shape as OIDC 404ing when unconfigured.
+	Enabled bool
+}
+
 // Enabled reports whether enough OIDC config is present to attempt discovery.
 func (o OIDCConfig) Enabled() bool {
 	return o.Issuer != "" && o.ClientID != ""
@@ -87,6 +112,16 @@ func Load() *Config {
 		TrustedProxies: envList("PANSY_TRUSTED_PROXIES"),
 	}
 
+	agentKey := envStr("OLLAMA_CLOUD_API_KEY", "")
+	cfg.Agent = AgentConfig{
+		Model:             envStr("PANSY_AGENT_MODEL", DefaultAgentModel),
+		OllamaCloudAPIKey: agentKey,
+		// Default on when a key is present: having configured the key IS the
+		// opt-in, and making people set a second flag to use what they just
+		// configured is a papercut with no upside.
+		Enabled: envBool("PANSY_AGENT_ENABLED", agentKey != ""),
+	}
+
 	if cfg.Registration != RegistrationOpen && cfg.Registration != RegistrationClosed {
 		slog.Warn("config: invalid PANSY_REGISTRATION, defaulting to open", "value", cfg.Registration)
 		cfg.Registration = RegistrationOpen

internal/domain/domain.go

@@ -191,6 +191,28 @@ type JournalEntry struct {
 	AuthorName string `json:"authorName,omitempty"`
 }
 
+// Roles a stored agent message can have.
+const (
+	AgentRoleUser      = "user"
+	AgentRoleAssistant = "assistant"
+)
+
+// AgentMessage is one side of a chat exchange with the garden assistant. Only
+// the text is kept, not the model's full transcript with its tool calls:
+// continuity needs what was said and what came back, and replaying a stored tool
+// call would be replaying a decision made against a garden that has since moved
+// on.
+type AgentMessage struct {
+	ID             int64  `json:"id"`
+	ConversationID int64  `json:"conversationId"`
+	Role           string `json:"role"`
+	Body           string `json:"body"`
+	// ChangeSetID is what this turn changed, if anything — the handle the chat
+	// panel needs to offer Undo on the message that caused it.
+	ChangeSetID *int64 `json:"changeSetId,omitempty"`
+	CreatedAt   string `json:"createdAt"`
+}
+
 // RevertConflict reports one entity a revert deliberately left alone, because
 // reverting it would have discarded a change made after the change set being
 // reverted. The rest of the change set still reverts.
@@ -220,6 +242,22 @@ const (
 	ConflictUnsupported = "unsupported"
 )
 
+// InstanceSettings is the single row of instance-wide, admin-editable
+// configuration (#79). Secrets are deliberately absent — see migration 0010.
+//
+// Both agent fields express "inherit from the environment unless set":
+// AgentModel == "" falls back to PANSY_AGENT_MODEL; AgentEnabled == nil inherits
+// the env default. EffectiveAgent resolves them against the environment.
+type InstanceSettings struct {
+	// AgentModel overrides PANSY_AGENT_MODEL when non-empty. Passed verbatim to
+	// majordomo.Parse, exactly like the env var it shadows.
+	AgentModel string `json:"agentModel"`
+	// AgentEnabled overrides PANSY_AGENT_ENABLED when non-nil. nil = inherit.
+	AgentEnabled *bool  `json:"agentEnabled"`
+	Version      int64  `json:"version"`
+	UpdatedAt    string `json:"updatedAt"`
+}
+
 // User is a pansy account. It may have a local password, OIDC identity, or both.
 type User struct {
 	ID           int64   `json:"id"`

internal/imagenorm/imagenorm.go

@@ -0,0 +1,207 @@
+// Package imagenorm normalizes an uploaded image to a JPEG the rest of pansy
+// (and majordomo's vision path) can rely on, decoding the formats a phone
+// actually produces.
+//
+// # Why this exists
+//
+// majordomo's own media pipeline is stdlib-based, so it cannot decode HEIC or
+// WebP — and HEIC is the iPhone camera default. The seed-packet feature's very
+// first input is "a photo from my phone", so without this the feature fails on
+// the exact device that motivates it. Normalizing at the upload boundary means
+// everything downstream only ever sees JPEG.
+//
+// # The CGO constraint
+//
+// pansy is CGO_ENABLED=0 (a single static binary — the reason modernc/sqlite was
+// chosen over the C one), so a libheif *binding* is out. github.com/gen2brain/heic
+// runs libheif as WebAssembly via wazero: pure Go, no cgo, and it registers with
+// image.Decode like any other format. golang.org/x/image/webp is pure Go too.
+//
+// # Import-driven registry
+//
+// Go's image decoders register via blank imports, and the failure mode is
+// backwards from intuition: forget "image/png" and PNG uploads fail with
+// "unknown format" while the exotic HEIC still works. So the blank imports below
+// are load-bearing, and TestNormalizeAllFormats exercises all four formats to
+// keep them so.
+package imagenorm
+
+import (
+	"bytes"
+	"errors"
+	"fmt"
+	"image"
+	"image/jpeg"
+	"io"
+
+	"golang.org/x/image/draw"
+
+	// Decoders, registered with image.Decode by side effect. All four matter:
+	// jpeg/png are the common cases, heic is the iPhone default, webp is common
+	// on the web. Dropping any one silently breaks that format's uploads.
+	_ "image/jpeg"
+	_ "image/png"
+
+	_ "github.com/gen2brain/heic"
+	_ "golang.org/x/image/webp"
+)
+
+// Defaults chosen for the seed-packet path against ollama-cloud's limits (8
+// images, 20 MiB, 2048px, jpeg+png). We re-encode to JPEG well under all of them.
+const (
+	// DefaultMaxDim is the longest-edge ceiling. 2048 matches ollama-cloud's
+	// MaxDim; anything larger is downscaled. A packet photo has plenty of detail
+	// left at 2048.
+	DefaultMaxDim = 2048
+	// DefaultMaxBytes caps the *input* we will read. A phone photo is 3–8 MB; 25
+	// MiB leaves headroom for a large HEIC without inviting a decompression bomb
+	// as an unbounded read. The re-encoded output is far smaller.
+	DefaultMaxBytes = 25 << 20
+	// maxDecodePixels bounds the DECODED bitmap regardless of input byte size, so
+	// a small file claiming enormous dimensions (a decompression bomb) is refused
+	// before its ~4-bytes/px bitmap is allocated. 50 MP ≈ 200 MB peak — above any
+	// current phone camera (a 48 MP sensor is 48 MP) while capping the
+	// amplification a hostile header can force. maxDecodePixels and maxDimension
+	// are internal safety floors, not knobs — unlike MaxDim/MaxBytes there's no
+	// reason for a caller to raise them.
+	maxDecodePixels = 50_000_000
+	// maxDimension caps EACH side independently. It exists to make the pixel-count
+	// check overflow-safe: without it, a header claiming ~2^32 on a side could
+	// wrap int64(w)*int64(h) negative and slip past maxDecodePixels. No real image
+	// is 50k px on a side.
+	maxDimension = 50_000
+	// jpegQuality for the normalized output. 85 is visually clean and keeps the
+	// file small; the model reads text off it, not fine gradients.
+	jpegQuality = 85
+)
+
+// Options tunes Normalize. The zero value uses the Default* constants.
+type Options struct {
+	MaxDim   int // longest edge; 0 → DefaultMaxDim
+	MaxBytes int // input read cap; 0 → DefaultMaxBytes
+}
+
+func (o Options) maxDim() int {
+	if o.MaxDim > 0 {
+		return o.MaxDim
+	}
+	return DefaultMaxDim
+}
+
+func (o Options) maxBytes() int {
+	if o.MaxBytes > 0 {
+		return o.MaxBytes
+	}
+	return DefaultMaxBytes
+}
+
+// ErrTooLarge means the input exceeded the byte cap or decoded to an absurd
+// pixel count. ErrUnsupported means the bytes weren't a decodable image format —
+// or a decoder panicked on them (see the recover in Normalize).
+var (
+	ErrTooLarge    = errors.New("imagenorm: image too large")
+	ErrUnsupported = errors.New("imagenorm: unsupported or corrupt image")
+)
+
+// Normalize reads an image of any supported format (JPEG, PNG, HEIC, WebP),
+// downscales it to fit opts.MaxDim on its longest edge, and returns it re-encoded
+// as JPEG, plus the decoded format name (e.g. "heic") — handy for logging what a
+// phone actually sent. On any error the returned bytes are nil and format is "".
+//
+// Errors, by cause:
+//   - input over opts.MaxBytes, or a decoded canvas over maxDecodePixels /
+//     maxDimension → ErrTooLarge, refused before the bitmap is allocated;
+//   - bytes that aren't a decodable image, or a decoder that panics on them →
+//     ErrUnsupported;
+//   - a genuine read or JPEG-encode I/O failure → a wrapped error (not a
+//     sentinel), since those are the caller's stream/environment, not the image.
+//
+// It bounds work against a hostile upload three ways: the byte cap, the
+// pre-decode pixel/dimension check, and a recover around the third-party decoders
+// (a malformed HEIC/WebP shouldn't take the process down).
+//
+// Two known gaps, both deferred to the upload handler that wires this in (#81):
+//   - EXIF ORIENTATION is not applied, so a portrait phone photo tagged
+//     "rotate 90°" comes out sideways. That's best fixed and tested with a real
+//     oriented photo end-to-end, which the library has no consumer for yet.
+//   - There is no context: image.Decode is CPU-bound and not cancellable
+//     mid-decode, so a caller that needs a hard deadline should run Normalize
+//     under its own timeout. The size guards keep the work finite regardless.
+func Normalize(r io.Reader, opts Options) (out []byte, format string, err error) {
+	// Cap the read at MaxBytes+1 so we can tell "exactly at the cap" from "over".
+	// maxBytes() is always a sane positive (default 25 MiB); guard the +1 anyway.
+	byteCap := opts.maxBytes()
+	limit := int64(byteCap) + 1
+	if limit < 1 {
+		limit = int64(DefaultMaxBytes) + 1
+	}
+	raw, err := io.ReadAll(io.LimitReader(r, limit))
+	if err != nil {
+		return nil, "", fmt.Errorf("imagenorm: read: %w", err)
+	}
+	if len(raw) > byteCap {
+		return nil, "", ErrTooLarge
+	}
+
+	// Check dimensions BEFORE a full decode, so a decompression bomb is refused
+	// before it allocates its bitmap. The per-side maxDimension check runs first
+	// so the pixel-count multiply below can't overflow.
+	cfg, _, err := image.DecodeConfig(bytes.NewReader(raw))
+	if err != nil {
+		return nil, "", ErrUnsupported
+	}
+	if cfg.Width <= 0 || cfg.Height <= 0 ||
+		cfg.Width > maxDimension || cfg.Height > maxDimension ||
+		int64(cfg.Width)*int64(cfg.Height) > maxDecodePixels {
+		return nil, "", ErrTooLarge
+	}
+
+	img, format, err := decodeSafely(raw)
+	if err != nil {
+		return nil, "", err
+	}
+
+	img = downscale(img, opts.maxDim())
+
+	var buf bytes.Buffer
+	if err := jpeg.Encode(&buf, img, &jpeg.Options{Quality: jpegQuality}); err != nil {
+		return nil, "", fmt.Errorf("imagenorm: encode jpeg: %w", err)
+	}
+	return buf.Bytes(), format, nil
+}
+
+// decodeSafely decodes raw, converting both a decode error and a decoder PANIC
+// into ErrUnsupported. The recover matters because the image comes from an
+// untrusted upload and the HEIC/WebP decoders are third-party (libheif via WASM,
+// x/image/webp): a malformed file that panics one of them must fail this one
+// request, not crash the process.
+func decodeSafely(raw []byte) (img image.Image, format string, err error) {
+	defer func() {
+		if r := recover(); r != nil {
+			img, format, err = nil, "", ErrUnsupported
+		}
+	}()
+	img, format, err = image.Decode(bytes.NewReader(raw))
+	if err != nil {
+		return nil, "", ErrUnsupported
+	}
+	return img, format, nil
+}
+
+// downscale returns img shrunk so its longest edge is at most maxDim, preserving
+// aspect ratio. An image already within bounds is returned unchanged (no
+// re-sampling, no quality loss beyond the JPEG round-trip). Uses Catmull-Rom for
+// a sharp result on text, which is what a packet photo is mostly made of.
+func downscale(img image.Image, maxDim int) image.Image {
+	b := img.Bounds()
+	w, h := b.Dx(), b.Dy()
+	longest := max(w, h)
+	if longest <= maxDim || longest == 0 {
+		return img
+	}
+	scale := float64(maxDim) / float64(longest)
+	nw, nh := max(int(float64(w)*scale), 1), max(int(float64(h)*scale), 1)
+	dst := image.NewRGBA(image.Rect(0, 0, nw, nh))
+	draw.CatmullRom.Scale(dst, dst.Bounds(), img, b, draw.Over, nil)
+	return dst
+}

internal/imagenorm/imagenorm_test.go

@@ -0,0 +1,202 @@
+package imagenorm
+
+import (
+	"bytes"
+	"encoding/binary"
+	"hash/crc32"
+	"image"
+	"image/jpeg"
+	"image/png"
+	"os"
+	"testing"
+)
+
+// pngBytes and jpegBytes generate in-memory fixtures for the two formats Go can
+// encode; heic/webp come from testdata (Go has no encoder for them).
+func pngBytes(t *testing.T, w, h int) []byte {
+	t.Helper()
+	m := image.NewRGBA(image.Rect(0, 0, w, h))
+	for y := range h {
+		for x := range w {
+			m.Pix[m.PixOffset(x, y)+0] = uint8(x)
+			m.Pix[m.PixOffset(x, y)+3] = 255
+		}
+	}
+	var b bytes.Buffer
+	if err := png.Encode(&b, m); err != nil {
+		t.Fatalf("encode png: %v", err)
+	}
+	return b.Bytes()
+}
+
+func jpegBytes(t *testing.T, w, h int) []byte {
+	t.Helper()
+	m := image.NewRGBA(image.Rect(0, 0, w, h))
+	var b bytes.Buffer
+	if err := jpeg.Encode(&b, m, nil); err != nil {
+		t.Fatalf("encode jpeg: %v", err)
+	}
+	return b.Bytes()
+}
+
+func readTestdata(t *testing.T, name string) []byte {
+	t.Helper()
+	b, err := os.ReadFile("testdata/" + name)
+	if err != nil {
+		t.Fatalf("read %s: %v", name, err)
+	}
+	return b
+}
+
+// TestNormalizeAllFormats is the load-bearing test: every format pansy claims to
+// accept must round-trip to a valid JPEG. It exists specifically to catch a
+// dropped blank import — the failure mode where the common format (PNG) breaks
+// while the exotic one (HEIC) works, because someone deleted `_ "image/png"`.
+func TestNormalizeAllFormats(t *testing.T) {
+	cases := []struct {
+		name       string
+		input      []byte
+		wantFormat string
+	}{
+		{"png", pngBytes(t, 120, 90), "png"},
+		{"jpeg", jpegBytes(t, 120, 90), "jpeg"},
+		{"heic", readTestdata(t, "sample.heic"), "heic"},
+		{"webp", readTestdata(t, "sample.webp"), "webp"},
+	}
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			out, format, err := Normalize(bytes.NewReader(tc.input), Options{})
+			if err != nil {
+				t.Fatalf("Normalize(%s): %v", tc.name, err)
+			}
+			if format != tc.wantFormat {
+				t.Errorf("format = %q, want %q", format, tc.wantFormat)
+			}
+			// The output must itself be a decodable JPEG.
+			_, outFormat, err := image.Decode(bytes.NewReader(out))
+			if err != nil {
+				t.Fatalf("output isn't a valid image: %v", err)
+			}
+			if outFormat != "jpeg" {
+				t.Errorf("output format = %q, want jpeg", outFormat)
+			}
+		})
+	}
+}
+
+// TestNormalizeDownscales checks a large image is shrunk to fit MaxDim on its
+// longest edge with aspect ratio preserved, and a small one is left alone.
+func TestNormalizeDownscales(t *testing.T) {
+	// A 2:1 image twice as wide as DefaultMaxDim → clamped to DefaultMaxDim on the
+	// long edge with aspect preserved. Derived from the constant, not hard-coded,
+	// so the test tracks the default rather than silently asserting a magic number.
+	longEdge := DefaultMaxDim * 2
+	big := pngBytes(t, longEdge, longEdge/2)
+	out, _, err := Normalize(bytes.NewReader(big), Options{})
+	if err != nil {
+		t.Fatalf("Normalize: %v", err)
+	}
+	cfg, _, err := image.DecodeConfig(bytes.NewReader(out))
+	if err != nil {
+		t.Fatalf("decode out: %v", err)
+	}
+	if cfg.Width != DefaultMaxDim {
+		t.Errorf("width = %d, want %d (longest edge clamped)", cfg.Width, DefaultMaxDim)
+	}
+	if cfg.Height != DefaultMaxDim/2 {
+		t.Errorf("height = %d, want %d (aspect preserved)", cfg.Height, DefaultMaxDim/2)
+	}
+
+	// A small image within bounds keeps its dimensions.
+	small := pngBytes(t, 100, 80)
+	out, _, err = Normalize(bytes.NewReader(small), Options{})
+	if err != nil {
+		t.Fatalf("Normalize small: %v", err)
+	}
+	cfg, _, err = image.DecodeConfig(bytes.NewReader(out))
+	if err != nil {
+		t.Fatalf("decode small out: %v", err)
+	}
+	if cfg.Width != 100 || cfg.Height != 80 {
+		t.Errorf("small image resized to %dx%d, want 100x80", cfg.Width, cfg.Height)
+	}
+}
+
+// TestNormalizeRejectsOversizeInput: an input past the byte cap is ErrTooLarge,
+// refused without a full decode.
+func TestNormalizeRejectsOversizeInput(t *testing.T) {
+	big := pngBytes(t, 500, 500)
+	_, _, err := Normalize(bytes.NewReader(big), Options{MaxBytes: 100})
+	if err != ErrTooLarge {
+		t.Errorf("over-cap input err = %v, want ErrTooLarge", err)
+	}
+}
+
+// TestNormalizeRejectsGarbage: unreadable-as-image bytes and a truncated image
+// both fail cleanly with ErrUnsupported, not a panic.
+func TestNormalizeRejectsGarbage(t *testing.T) {
+	_, _, err := Normalize(bytes.NewReader([]byte("not an image at all")), Options{})
+	if err != ErrUnsupported {
+		t.Errorf("garbage err = %v, want ErrUnsupported", err)
+	}
+	// A truncated image (valid header, cut body) also fails cleanly, not a panic.
+	png := pngBytes(t, 100, 100)
+	_, _, err = Normalize(bytes.NewReader(png[:len(png)/2]), Options{})
+	if err != ErrUnsupported {
+		t.Errorf("truncated image err = %v, want ErrUnsupported", err)
+	}
+}
+
+// pngHeader builds a valid PNG signature + IHDR chunk (with a correct CRC, which
+// DecodeConfig verifies) for the given dimensions, and nothing else. It's enough
+// for image.DecodeConfig to report width/height without a real bitmap — exactly
+// what's needed to exercise the pre-decode size guard with a tiny input.
+func pngHeader(w, h uint32) []byte {
+	var buf bytes.Buffer
+	buf.Write([]byte{0x89, 'P', 'N', 'G', 0x0d, 0x0a, 0x1a, 0x0a})
+	ihdr := make([]byte, 13)
+	binary.BigEndian.PutUint32(ihdr[0:], w)
+	binary.BigEndian.PutUint32(ihdr[4:], h)
+	ihdr[8] = 8 // bit depth
+	ihdr[9] = 6 // colour type: RGBA
+	// compression/filter/interlace already 0.
+	binary.Write(&buf, binary.BigEndian, uint32(len(ihdr)))
+	chunk := append([]byte("IHDR"), ihdr...)
+	buf.Write(chunk)
+	binary.Write(&buf, binary.BigEndian, crc32.ChecksumIEEE(chunk))
+	return buf.Bytes()
+}
+
+// TestNormalizeRejectsPixelBomb is the guard the review found untested: a small
+// input (a bare ~40-byte PNG header) claiming an enormous canvas is refused with
+// ErrTooLarge from DecodeConfig alone, before image.Decode allocates anything.
+// Covers both the per-side maxDimension trip and the pixel-count trip — and, via
+// the near-2^16-per-side case, that the count math doesn't overflow.
+func TestNormalizeRejectsPixelBomb(t *testing.T) {
+	cases := []struct {
+		name string
+		w, h uint32
+	}{
+		{"huge single side", 60000, 10},             // > maxDimension on width
+		{"huge area within side cap", 40000, 40000}, // sides < cap, area 1.6 GP > maxDecodePixels
+	}
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			hdr := pngHeader(tc.w, tc.h)
+			if len(hdr) > 100 {
+				t.Fatalf("header unexpectedly large (%d bytes) — not a bomb test", len(hdr))
+			}
+			// Sanity: the header really does decode to those dimensions.
+			cfg, _, err := image.DecodeConfig(bytes.NewReader(hdr))
+			if err != nil {
+				t.Fatalf("crafted PNG header didn't parse: %v", err)
+			}
+			if uint32(cfg.Width) != tc.w || uint32(cfg.Height) != tc.h {
+				t.Fatalf("header reports %dx%d, want %dx%d", cfg.Width, cfg.Height, tc.w, tc.h)
+			}
+			if _, _, err := Normalize(bytes.NewReader(hdr), Options{}); err != ErrTooLarge {
+				t.Errorf("pixel bomb %dx%d err = %v, want ErrTooLarge", tc.w, tc.h, err)
+			}
+		})
+	}
+}

internal/imagenorm/testdata/README.md

@@ -0,0 +1,14 @@
+# imagenorm test fixtures
+
+Go has no encoder for HEIC or WebP, so these small samples are committed rather
+than generated at test time. They are used by `TestNormalizeAllFormats` to prove
+every accepted format round-trips to JPEG (and, mainly, to catch a dropped blank
+import — see the package doc).
+
+- `sample.heic` — a 240×160 gradient, created from a Go-generated PNG with macOS
+  `sips -s format heic`. HEVC still-image profile.
+- `sample.webp` — a 16×16 image copied from CPython's stdlib test corpus
+  (`Lib/test/test_email/data/python.webp`), verified to decode with
+  `golang.org/x/image/webp` before committing. Used only as a decode fixture.
+
+Neither contains anything meaningful; they exist purely to be decoded.

internal/service/agent.go

@@ -0,0 +1,62 @@
+package service
+
+import (
+	"context"
+
+	"gitea.stevedudenhoeffer.com/steve/pansy/internal/domain"
+)
+
+// Chat persistence for the garden assistant (#56). The run loop itself lives in
+// internal/agent; this is the part that needs pansy's permission checks.
+
+// maxRetainedTurns caps how much of a thread is kept in context. Retained turns
+// are a working memory, not an archive — a season of chat replayed into every
+// prompt would cost more than it informs.
+const maxRetainedTurns = 20
+
+// AgentHistory returns the recent messages of the actor's own thread for a
+// garden they can edit, oldest first.
+//
+// Requires EDITOR, not viewer: the assistant acts, so being able to talk to it
+// about a garden is the same permission as being able to change it. A viewer
+// asking it to plant something would get a refusal from every tool anyway, and
+// offering the conversation would be offering something that cannot work.
+func (s *Service) AgentHistory(ctx context.Context, actorID, gardenID int64) ([]domain.AgentMessage, error) {
+	if _, err := s.requireGardenRole(ctx, actorID, gardenID, roleEditor); err != nil {
+		return nil, err
+	}
+	conv, err := s.store.EnsureConversation(ctx, gardenID, actorID)
+	if err != nil {
+		return nil, err
+	}
+	return s.store.ListAgentMessages(ctx, conv, maxRetainedTurns*2)
+}
+
+// RecordAgentExchange stores one user message and the assistant's reply, and
+// returns the reply row. Called by the runtime after a turn completes.
+func (s *Service) RecordAgentExchange(ctx context.Context, actorID, gardenID int64, userMessage, reply string, changeSetID *int64) (*domain.AgentMessage, error) {
+	if _, err := s.requireGardenRole(ctx, actorID, gardenID, roleEditor); err != nil {
+		return nil, err
+	}
+	conv, err := s.store.EnsureConversation(ctx, gardenID, actorID)
+	if err != nil {
+		return nil, err
+	}
+	if _, err := s.store.AppendAgentMessage(ctx, &domain.AgentMessage{
+		ConversationID: conv, Role: domain.AgentRoleUser, Body: userMessage,
+	}); err != nil {
+		return nil, err
+	}
+	return s.store.AppendAgentMessage(ctx, &domain.AgentMessage{
+		ConversationID: conv, Role: domain.AgentRoleAssistant, Body: reply, ChangeSetID: changeSetID,
+	})
+}
+
+// ClearAgentHistory drops the actor's thread for a garden — the "start over"
+// escape hatch when a conversation has gone somewhere unhelpful.
+func (s *Service) ClearAgentHistory(ctx context.Context, actorID, gardenID int64) error {
+	if _, err := s.requireGardenRole(ctx, actorID, gardenID, roleEditor); err != nil {
+		return err
+	}
+	return s.store.DeleteConversation(ctx, gardenID, actorID)
+}

internal/service/instance_settings.go

@@ -0,0 +1,110 @@
+package service
+
+import (
+	"context"
+	"strings"
+
+	"gitea.stevedudenhoeffer.com/steve/pansy/internal/agentmodel"
+	"gitea.stevedudenhoeffer.com/steve/pansy/internal/domain"
+)
+
+// Instance settings (#79): admin-editable, instance-wide configuration. The
+// admin gate lives HERE, in the seam, not in the handler — the same rule every
+// other permission follows. The API's requireAdmin middleware is a cheap early
+// 403, not the authority.
+
+// requireAdmin returns nil iff the actor is an admin, else ErrForbidden.
+//
+// ErrForbidden, not ErrNotFound: settings are not a resource whose existence is
+// masked. A logged-in non-admin knows the instance has settings; they simply may
+// not touch them. (Contrast objects/lots, where no-access masks existence.)
+func (s *Service) requireAdmin(ctx context.Context, actorID int64) error {
+	u, err := s.store.GetUserByID(ctx, actorID)
+	if err != nil {
+		return err
+	}
+	if !u.IsAdmin {
+		return domain.ErrForbidden
+	}
+	return nil
+}
+
+// GetInstanceSettings returns the instance settings for an admin.
+func (s *Service) GetInstanceSettings(ctx context.Context, actorID int64) (*domain.InstanceSettings, error) {
+	if err := s.requireAdmin(ctx, actorID); err != nil {
+		return nil, err
+	}
+	return s.store.GetInstanceSettings(ctx)
+}
+
+// InstanceSettingsPatch is a full replacement of the editable fields plus the
+// current version. Both agent fields carry their "inherit" sentinel: an empty
+// AgentModel means fall back to env, a nil AgentEnabled means inherit.
+type InstanceSettingsPatch struct {
+	AgentModel   string
+	AgentEnabled *bool
+	Version      int64
+}
+
+// UpdateInstanceSettings applies an admin's change, version-guarded. It returns
+// (current row, ErrVersionConflict) on a stale version, like every mutable
+// resource. The model spec is validated before it is stored, so a typo is a 400
+// now rather than a broken assistant on the next turn.
+//
+// It does NOT rebuild the running agent — that is the API layer's job, because
+// the live Runner lives there. The caller rebuilds on success.
+func (s *Service) UpdateInstanceSettings(ctx context.Context, actorID int64, patch InstanceSettingsPatch) (*domain.InstanceSettings, error) {
+	if err := s.requireAdmin(ctx, actorID); err != nil {
+		return nil, err
+	}
+	model := strings.TrimSpace(patch.AgentModel)
+	// Validate a non-empty spec up front. An empty one is the "inherit env"
+	// sentinel and needs no check — the env value was validated at boot.
+	if model != "" {
+		if err := agentmodel.Validate(s.cfg.Agent.OllamaCloudAPIKey, model); err != nil {
+			return nil, domain.ErrInvalidInput
+		}
+	}
+	return s.store.UpdateInstanceSettings(ctx, &domain.InstanceSettings{
+		AgentModel:   model,
+		AgentEnabled: patch.AgentEnabled,
+		Version:      patch.Version,
+	})
+}
+
+// EffectiveAgent resolves the agent configuration actually in force: DB settings
+// override the environment, and the API key always comes from the environment.
+//
+// This is internal plumbing (the API layer calls it to build the Runner), NOT an
+// admin-gated operation — resolving what's configured is not the same as editing
+// it, and the agent bootstrap must work before any user is even authenticated.
+type EffectiveAgent struct {
+	Model   string
+	Enabled bool
+	APIKey  string
+}
+
+// Ready mirrors config.AgentConfig.Ready: enabled, with a key and a model.
+func (e EffectiveAgent) Ready() bool {
+	return e.Enabled && e.APIKey != "" && e.Model != ""
+}
+
+// EffectiveAgent reads the settings row and layers it over the environment.
+func (s *Service) EffectiveAgent(ctx context.Context) (EffectiveAgent, error) {
+	st, err := s.store.GetInstanceSettings(ctx)
+	if err != nil {
+		return EffectiveAgent{}, err
+	}
+	eff := EffectiveAgent{
+		Model:   s.cfg.Agent.Model,
+		Enabled: s.cfg.Agent.Enabled,
+		APIKey:  s.cfg.Agent.OllamaCloudAPIKey,
+	}
+	if st.AgentModel != "" {
+		eff.Model = st.AgentModel
+	}
+	if st.AgentEnabled != nil {
+		eff.Enabled = *st.AgentEnabled
+	}
+	return eff, nil
+}

internal/service/instance_settings_test.go

@@ -0,0 +1,117 @@
+package service
+
+import (
+	"context"
+	"errors"
+	"testing"
+
+	"gitea.stevedudenhoeffer.com/steve/pansy/internal/config"
+	"gitea.stevedudenhoeffer.com/steve/pansy/internal/domain"
+)
+
+// settingsTestService builds a service whose env config carries the given agent
+// model/enabled/key, so EffectiveAgent's env fallback can be exercised.
+func settingsTestService(t *testing.T, envModel string, envEnabled bool, key string) (*Service, int64) {
+	t.Helper()
+	cfg := openConfig()
+	cfg.Agent = config.AgentConfig{Model: envModel, Enabled: envEnabled, OllamaCloudAPIKey: key}
+	s := newTestService(t, cfg)
+	admin := seedUser(t, s, "[email protected]") // first user is admin
+	return s, admin
+}
+
+// TestRequireAdmin: the first user is admin; a second is not and gets
+// ErrForbidden (not ErrNotFound — settings existence isn't masked).
+func TestRequireAdmin(t *testing.T) {
+	s, admin := settingsTestService(t, "ollama-cloud/x", true, "k")
+	member := seedUser(t, s, "[email protected]")
+
+	if err := s.requireAdmin(context.Background(), admin); err != nil {
+		t.Errorf("admin rejected: %v", err)
+	}
+	if err := s.requireAdmin(context.Background(), member); !errors.Is(err, domain.ErrForbidden) {
+		t.Errorf("member requireAdmin = %v, want ErrForbidden", err)
+	}
+}
+
+// TestEffectiveAgentLayering: DB settings override env; the API key always comes
+// from env; the "inherit" sentinels fall back.
+func TestEffectiveAgentLayering(t *testing.T) {
+	ctx := context.Background()
+	s, admin := settingsTestService(t, "ollama-cloud/env-model", true, "envkey")
+
+	// Untouched: everything inherits env.
+	eff, err := s.EffectiveAgent(ctx)
+	if err != nil {
+		t.Fatalf("effective: %v", err)
+	}
+	if eff.Model != "ollama-cloud/env-model" || !eff.Enabled || eff.APIKey != "envkey" {
+		t.Errorf("inherited effective = %+v, want the env values", eff)
+	}
+
+	// Override the model only; enabled still inherits env (true).
+	cur, _ := s.GetInstanceSettings(ctx, admin)
+	if _, err := s.UpdateInstanceSettings(ctx, admin, InstanceSettingsPatch{
+		AgentModel: "ollama-cloud/glm-5.2:cloud", Version: cur.Version,
+	}); err != nil {
+		t.Fatalf("update model: %v", err)
+	}
+	eff, _ = s.EffectiveAgent(ctx)
+	if eff.Model != "ollama-cloud/glm-5.2:cloud" {
+		t.Errorf("model = %q, want the DB override", eff.Model)
+	}
+	if !eff.Enabled {
+		t.Error("enabled should still inherit env (true) when unset")
+	}
+
+	// Now override enabled to false explicitly.
+	cur, _ = s.GetInstanceSettings(ctx, admin)
+	no := false
+	if _, err := s.UpdateInstanceSettings(ctx, admin, InstanceSettingsPatch{
+		AgentModel: "ollama-cloud/glm-5.2:cloud", AgentEnabled: &no, Version: cur.Version,
+	}); err != nil {
+		t.Fatalf("update enabled: %v", err)
+	}
+	eff, _ = s.EffectiveAgent(ctx)
+	if eff.Enabled {
+		t.Error("enabled should be the explicit false override now")
+	}
+	if eff.Ready() {
+		t.Error("Ready() should be false when disabled")
+	}
+}
+
+// TestUpdateInstanceSettingsRejectsBadModel: a spec that won't resolve is
+// ErrInvalidInput, before it is stored.
+func TestUpdateInstanceSettingsRejectsBadModel(t *testing.T) {
+	ctx := context.Background()
+	s, admin := settingsTestService(t, "ollama-cloud/x", true, "k")
+	cur, _ := s.GetInstanceSettings(ctx, admin)
+
+	if _, err := s.UpdateInstanceSettings(ctx, admin, InstanceSettingsPatch{
+		AgentModel: "nonesuch/model", Version: cur.Version,
+	}); !errors.Is(err, domain.ErrInvalidInput) {
+		t.Errorf("bad model = %v, want ErrInvalidInput", err)
+	}
+
+	// The rejected write didn't touch the row.
+	after, _ := s.GetInstanceSettings(ctx, admin)
+	if after.Version != cur.Version || after.AgentModel != "" {
+		t.Errorf("a rejected update changed the row: %+v", after)
+	}
+}
+
+// TestInstanceSettingsAdminGate: the read/write operations are admin-gated at the
+// service seam, not just in the handler.
+func TestInstanceSettingsAdminGate(t *testing.T) {
+	ctx := context.Background()
+	s, _ := settingsTestService(t, "ollama-cloud/x", true, "k")
+	member := seedUser(t, s, "[email protected]")
+
+	if _, err := s.GetInstanceSettings(ctx, member); !errors.Is(err, domain.ErrForbidden) {
+		t.Errorf("member GetInstanceSettings = %v, want ErrForbidden", err)
+	}
+	if _, err := s.UpdateInstanceSettings(ctx, member, InstanceSettingsPatch{Version: 1}); !errors.Is(err, domain.ErrForbidden) {
+		t.Errorf("member UpdateInstanceSettings = %v, want ErrForbidden", err)
+	}
+}

internal/service/ops.go

@@ -28,13 +28,9 @@ type Region struct {
 	MinX, MinY, MaxX, MaxY float64
 }
 
-// contains reports whether a local point lies in the region.
-func (r Region) contains(x, y float64) bool {
-	return x >= r.MinX && x <= r.MaxX && y >= r.MinY && y <= r.MaxY
-}
-
 // clampTo intersects the region with an object's local bounds (±halfW, ±halfH),
-// so an oversized caller-supplied region can't make hexCenters loop forever.
+// so a fill can't plant outside the object it was aimed at. A region that misses
+// the object entirely comes back empty — see empty().
 func (r Region) clampTo(halfW, halfH float64) Region {
 	return Region{
 		MinX: math.Max(r.MinX, -halfW), MinY: math.Max(r.MinY, -halfH),
@@ -42,6 +38,16 @@ func (r Region) clampTo(halfW, halfH float64) Region {
 	}
 }
 
+// empty reports whether the region encloses nothing.
+//
+// This exists because clampTo expresses "no overlap" by INVERTING the region —
+// Max clamps below Min — rather than by zeroing it, which is not something a
+// reader guesses. Naming it once here beats a bare `MaxX < MinX` at each place
+// that has to care.
+func (r Region) empty() bool {
+	return r.MaxX < r.MinX || r.MaxY < r.MinY
+}
+
 // rect builds a rectangular region.
 func rect(minX, minY, maxX, maxY float64) Region {
 	return Region{MinX: minX, MinY: minY, MaxX: maxX, MaxY: maxY}
@@ -94,9 +100,11 @@ func defaultPlopRadius(spacingCM float64) float64 {
 // FillRegion lays a hex-packed field of plops of one plant across a region of a
 // plantable object the actor can edit. Plop radius comes from the plant's spacing
 // (or spacingOverride) via defaultPlopRadius; centers sit on a hex lattice at 2×
-// radius pitch, kept where the center is inside the region. A candidate is
+// radius pitch, centered in the region, and set in from each edge by the plop's
-// skipped when its plop would sit entirely inside an existing active plop (so
+// radius less half a spacing — see hexCenters for why that half-spacing is what
-// re-filling doesn't stack duplicates). Returns the plops it created.
+// the edge is owed. A candidate is skipped when its plop would sit entirely
+// inside an existing active plop (so re-filling doesn't stack duplicates).
+// Returns the plops it created.
 func (s *Service) FillRegion(ctx context.Context, actorID, objectID int64, region Region, plantID int64, spacingOverride *float64) ([]domain.Planting, error) {
 	o, _, err := s.objectForRole(ctx, actorID, objectID, roleEditor)
 	if err != nil {
@@ -106,9 +114,9 @@ func (s *Service) FillRegion(ctx context.Context, actorID, objectID int64, regio
 }
 
 // fillLoaded is the shared body of FillRegion/FillNamedRegion given an object
-// already loaded and authorized (roleEditor). It clamps the region to the
+// already loaded and authorized (roleEditor). It rejects a non-finite region,
-// object's bounds, refuses fills over maxFillPlops, and inserts the whole batch
+// clamps the region to the object's bounds, refuses fills over maxFillPlops, and
-// in one transaction rather than one round-trip per plop.
+// inserts the whole batch in one transaction rather than one round-trip per plop.
 func (s *Service) fillLoaded(ctx context.Context, actorID int64, o *domain.GardenObject, region Region, plantID int64, spacingOverride *float64) ([]domain.Planting, error) {
 	if !o.Plantable {
 		return nil, domain.ErrInvalidInput
@@ -129,9 +137,21 @@ func (s *Service) fillLoaded(ctx context.Context, actorID int64, o *domain.Garde
 		return nil, domain.ErrInvalidInput
 	}
 
+	// A caller-supplied region is arbitrary floats, and non-finite ones survive
+	// everything downstream: clamping keeps them, the inverted-region guard can't
+	// see NaN (it compares false both ways), and fitAxis centres on them happily.
+	// Nothing corrupt reaches the table — SQLite stores NaN as NULL and the NOT
+	// NULL constraint refuses it — but the caller gets an opaque store error for
+	// NaN, and for +Inf a silent zero-plop success. Both are lies about what went
+	// wrong; say "bad input" here instead.
+	if !isFinite(region.MinX) || !isFinite(region.MinY) ||
+		!isFinite(region.MaxX) || !isFinite(region.MaxY) {
+		return nil, domain.ErrInvalidInput
+	}
+
 	region = region.clampTo(o.WidthCM/2, o.HeightCM/2)
-	centers := hexCenters(region, radius)
+	centers, total := hexCenters(region, radius, spacing, maxFillPlops)
-	if len(centers) > maxFillPlops {
+	if total > maxFillPlops {
 		return nil, domain.ErrInvalidInput // region too large for this spacing; ask for less
 	}
 
@@ -169,32 +189,112 @@ func (s *Service) fillLoaded(ctx context.Context, actorID int64, o *domain.Garde
 
 type localPoint struct{ x, y float64 }
 
-// hexCenters returns hex-packed lattice centers whose center lies in the region.
+// hexCenters returns hex-packed lattice centers filling a region: rows radius·√3
-// Rows are spaced radius·√3 apart and every other row is offset by radius, the
+// apart, alternate rows offset by half a pitch, at a 2×radius pitch. The lattice
-// standard hexagonal packing at a 2×radius pitch. The lattice is anchored one
+// is CENTERED, so the leftover is shared between opposite edges instead of piling
-// radius inside the region's min corner so the first plop sits inside it.
+// up against the far one.
-func hexCenters(r Region, radius float64) []localPoint {
+//
+// # How close to the edge the outer row goes
+//
+// Spacing is a constraint BETWEEN NEIGHBOURING PLANTS competing for the same
+// soil, light and water. A bed edge is not a competitor, so the outer row only
+// owes it HALF the spacing — the half it would otherwise share with a neighbour.
+// That is the arithmetic inside every square-foot-gardening chart: 4 per square
+// is 6" apart and 3" from the square's edge; 9 per square is 4" apart and 2"
+// from the edge. Garlic at 9 per square goes in 2" from the frame, not 6".
+//
+// A plop is a CLUMP, not a plant — defaultPlopRadius makes it 1.5×spacing, so
+// three spacings across — and its plants sit out to its rim. So keeping the whole
+// circle inside the bed would inset the outer row by a full 1.5 spacings, three
+// times what the rule allows. Instead the clump may hang over the edge by up to
+// half a spacing, which puts its outermost plants exactly the half-spacing from
+// the edge that the rule asks for. Overhang is capped there and nowhere near the
+// full radius: a clump mostly outside the bed is a drawing of plants in the path.
+//
+// Do not "simplify" this back to anchoring at the region's min corner. That is
+// what #75 was: staggered rows start a full pitch in, and the leftover all lands
+// on the far edge, where clumps hang outside a bed that nothing clips them to.
+//
+// # Counting before building
+//
+// hexCenters returns the total alongside the points, and works that total out
+// BEFORE building anything: a fill large enough to be refused shouldn't allocate
+// its whole lattice first just to be counted and thrown away. Over `limit` it
+// returns (nil, total), so the caller can still refuse with the real number.
+func hexCenters(r Region, radius, spacing float64, limit int) ([]localPoint, int) {
 	if radius <= 0 {
-		return nil
+		return nil, 0
+	}
+	// An empty region has no inside to plant. The old loop-until-past-MaxX form
+	// got this for free by never entering the loop; counting positions up front
+	// does not, and would site a plop off the bed.
+	if r.empty() {
+		return nil, 0
 	}
 	pitch := 2 * radius
 	rowH := pitch * math.Sqrt(3) / 2
-	const eps = 1e-6
+
-	var pts []localPoint
+	// How far a clump's centre must stay inside the edge: its own radius, less the
-	row := 0
+	// half-spacing of overhang the rule allows. Never negative, and never past the
-	for y := r.MinY + radius; y <= r.MaxY+eps; y += rowH {
+	// centre of the clump.
-		xStart := r.MinX + radius
+	inset := math.Max(0, radius-math.Max(0, spacing)/2)
-		if row%2 == 1 {
+
-			xStart += radius
+	rows, y0 := fitAxis(r.MaxY-r.MinY, rowH, inset)
+	cols, x0 := fitAxis(r.MaxX-r.MinX, pitch, inset)
+
+	// Exact, not an upper bound: staggered rows hold one fewer, so rows*cols would
+	// over-reserve by ~12% — and, more to the point, allocating it is the thing we
+	// are trying to avoid when the answer is "too many".
+	staggered := cols
+	if cols > 1 {
+		staggered = cols - 1
 	}
-		for x := xStart; x <= r.MaxX+eps; x += pitch {
+	total := (rows+1)/2*cols + rows/2*staggered
-			if r.contains(x, y) {
+	if total > limit {
-				pts = append(pts, localPoint{x, y})
+		return nil, total
+	}
+
+	pts := make([]localPoint, 0, total)
+	for row := 0; row < rows; row++ {
+		y := r.MinY + y0 + float64(row)*rowH
+		n, x := cols, r.MinX+x0
+		// The stagger falls out of centering: an offset row holds one fewer plop,
+		// and centering THAT run puts it exactly half a pitch off its neighbours.
+		// A single-column region has nothing to stagger against.
+		if row%2 == 1 && cols > 1 {
+			n, x = staggered, r.MinX+x0+pitch/2
+		}
+		for i := 0; i < n; i++ {
+			pts = append(pts, localPoint{x + float64(i)*pitch, y})
 		}
 	}
-		row++
+	return pts, total
 }
-	return pts
+
+// fitAxis returns how many lattice positions fit along a span at `step`, keeping
+// at least `inset` from each end, and the offset from the span's start that
+// centers them — so the leftover is split between the two edges rather than all
+// landing on the far one.
+//
+// A span too small to hold even one position at that inset still gets one, in the
+// middle: filling a bed narrower than a single plop with one plop is a better
+// answer than refusing to plant it.
+//
+// The step<=0 half of that guard is currently unreachable — hexCenters, the only
+// caller, returns early unless radius > 0, which makes both steps it passes
+// positive. It stays because dividing by a non-positive step yields ±Inf and then
+// a garbage int conversion, and a helper this small should not require reading
+// its caller to know it is safe. Deliberate, not an oversight.
+func fitAxis(length, step, inset float64) (n int, start float64) {
+	if step <= 0 || length < 2*inset {
+		return 1, length / 2
+	}
+	// The epsilon keeps an exact fit from being lost to floating point — a 60cm
+	// span at a 30cm step should give 2 positions, not 1 because the division
+	// landed on 0.9999999.
+	const eps = 1e-9
+	n = int(math.Floor((length-2*inset)/step+eps)) + 1
+	return n, (length - float64(n-1)*step) / 2
 }
 
 // coveredByExisting reports whether a new plop (center, radius) would sit

internal/service/ops_test.go

@@ -3,6 +3,8 @@ package service
 import (
 	"context"
 	"errors"
+	"math"
+	"sort"
 	"testing"
 
 	"gitea.stevedudenhoeffer.com/steve/pansy/internal/domain"
@@ -71,6 +73,165 @@ func TestDefaultPlopRadius(t *testing.T) {
 	}
 }
 
+// TestHexCentersEdgeInset pins the spacing rule the packing exists to honour:
+// spacing is a constraint between neighbouring plants, so a bed edge — which is
+// nobody's neighbour — is owed half a pitch, not a whole one.
+//
+// The bug this guards against was visible to anyone who filled a bed: staggered
+// rows began a full pitch in, leaving a bare strip a whole plop wide down one
+// side of every other row, while the far edge had plops hanging off it.
+func TestHexCentersEdgeInset(t *testing.T) {
+	for _, tc := range []struct {
+		name                  string
+		w, h, radius, spacing float64
+		wantRowStarts         []float64 // x of the first plop in rows 0 and 1
+	}{
+		// 4ft × 8ft bed, garlic at 15cm spacing → radius 22.5, pitch 45. Three
+		// columns, the outer ones overhanging by 6.5cm — under the 7.5cm the rule
+		// allows. Anchored at the corner this row started at -38.5 and its
+		// staggered neighbour a full 45 further in still.
+		{"4ft bed of garlic", 122, 244, 22.5, 15, []float64{-45, -22.5}},
+		// An exact fit: 90 wide at pitch 30 → 3 columns, no overhang needed.
+		{"exact fit", 90, 90, 15, 10, []float64{-30, -15}},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			r := rect(-tc.w/2, -tc.h/2, tc.w/2, tc.h/2)
+			pts, total := hexCenters(r, tc.radius, tc.spacing, maxFillPlops)
+			if len(pts) == 0 {
+				t.Fatal("no centers")
+			}
+			// The count is derived up front so an oversized fill is refused without
+			// building its lattice — which only works if it matches what gets built.
+			if total != len(pts) {
+				t.Errorf("reported total %d, built %d", total, len(pts))
+			}
+
+			// A clump may cross the edge, but only by the half-spacing the rule
+			// allows — never enough to be mostly out in the path.
+			budget := tc.spacing / 2
+			for _, p := range pts {
+				over := math.Max(
+					math.Max(r.MinX-(p.x-tc.radius), (p.x+tc.radius)-r.MaxX),
+					math.Max(r.MinY-(p.y-tc.radius), (p.y+tc.radius)-r.MaxY),
+				)
+				if over > budget+1e-6 {
+					t.Errorf("plop at (%.1f,%.1f) overhangs by %.2f, budget %.2f", p.x, p.y, over, budget)
+				}
+			}
+
+			// The margins match on opposite edges: the leftover is shared, not piled
+			// against the far side.
+			minX, maxX, minY, maxY := pts[0].x, pts[0].x, pts[0].y, pts[0].y
+			for _, p := range pts {
+				minX, maxX = math.Min(minX, p.x), math.Max(maxX, p.x)
+				minY, maxY = math.Min(minY, p.y), math.Max(maxY, p.y)
+			}
+			if w, e := minX-r.MinX, r.MaxX-maxX; math.Abs(w-e) > 1e-6 {
+				t.Errorf("lopsided horizontally: west margin %.2f, east %.2f", w, e)
+			}
+			if n, s := minY-r.MinY, r.MaxY-maxY; math.Abs(n-s) > 1e-6 {
+				t.Errorf("lopsided vertically: north margin %.2f, south %.2f", n, s)
+			}
+
+			// The staggered row is offset by HALF a pitch, not a whole one.
+			starts := map[float64]float64{}
+			for _, p := range pts {
+				if x, ok := starts[p.y]; !ok || p.x < x {
+					starts[p.y] = p.x
+				}
+			}
+			ys := make([]float64, 0, len(starts))
+			for y := range starts {
+				ys = append(ys, y)
+			}
+			sort.Float64s(ys)
+			for i, want := range tc.wantRowStarts {
+				if i >= len(ys) {
+					t.Fatalf("only %d rows, want at least %d", len(ys), len(tc.wantRowStarts))
+				}
+				if got := starts[ys[i]]; math.Abs(got-want) > 1e-6 {
+					t.Errorf("row %d starts at x=%.2f, want %.2f", i, got, want)
+				}
+			}
+		})
+	}
+}
+
+// TestHexCentersTinyRegion covers a region too small to hold a plop at the
+// half-pitch inset: planting one in the middle beats refusing to plant at all.
+//
+// The off-centre case earns its place — a region symmetric about the origin
+// can't tell "the middle of the region" from "the origin", so on its own it
+// would pass for an implementation that just returned (0,0).
+func TestHexCentersTinyRegion(t *testing.T) {
+	for _, tc := range []struct {
+		name         string
+		r            Region
+		wantX, wantY float64
+	}{
+		{"centred on the origin", rect(-5, -5, 5, 5), 0, 0},
+		{"off in a corner", rect(20, -40, 30, -30), 25, -35},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			pts, _ := hexCenters(tc.r, 15, 10, maxFillPlops)
+			if len(pts) != 1 || pts[0].x != tc.wantX || pts[0].y != tc.wantY {
+				t.Errorf("got %+v, want one plop at (%v,%v)", pts, tc.wantX, tc.wantY)
+			}
+		})
+	}
+}
+
+// TestFillRegionRejectsNonFiniteRegion: non-finite bounds survive clamping and
+// the inverted-region guard (NaN compares false both ways). Without the explicit
+// check, NaN surfaced as a raw store error ("NOT NULL constraint failed") and
+// +Inf as a silent success that planted nothing — neither of which tells the
+// caller what it actually did wrong.
+func TestFillRegionRejectsNonFiniteRegion(t *testing.T) {
+	ctx := context.Background()
+	s := newTestService(t, openConfig())
+	owner := seedUser(t, s, "[email protected]")
+	g, _ := s.CreateGarden(ctx, owner, GardenInput{Name: "Big", WidthCM: 2000, HeightCM: 2000})
+	bed := seedFillBed(t, s, owner, g.ID, 100, 100)
+	plant := seedOwnPlant(t, s, owner, 10)
+
+	nan := math.NaN()
+	for _, r := range []Region{
+		{MinX: nan, MinY: -50, MaxX: 50, MaxY: 50},
+		{MinX: -50, MinY: -50, MaxX: 50, MaxY: math.Inf(1)},
+	} {
+		created, err := s.FillRegion(ctx, owner, bed.ID, r, plant.ID, nil)
+		if !errors.Is(err, domain.ErrInvalidInput) {
+			t.Errorf("FillRegion(%+v) err = %v, want ErrInvalidInput", r, err)
+		}
+		for _, p := range created {
+			if !isFinite(p.XCM) || !isFinite(p.YCM) {
+				t.Errorf("persisted a plop with non-finite coordinates: %+v", p)
+			}
+		}
+	}
+}
+
+// TestFillRegionOutsideObjectPlantsNothing covers a region that misses the object
+// entirely. clampTo inverts such a region rather than emptying it, and an
+// inverted region must plant nothing — not one plop at some point off the bed.
+func TestFillRegionOutsideObjectPlantsNothing(t *testing.T) {
+	ctx := context.Background()
+	s := newTestService(t, openConfig())
+	owner := seedUser(t, s, "[email protected]")
+	g, _ := s.CreateGarden(ctx, owner, GardenInput{Name: "Big", WidthCM: 2000, HeightCM: 2000})
+	bed := seedFillBed(t, s, owner, g.ID, 100, 100) // local bounds ±50
+	plant := seedOwnPlant(t, s, owner, 10)
+
+	// Wholly east of the bed: clampTo gives MinX=500, MaxX=50.
+	created, err := s.FillRegion(ctx, owner, bed.ID, rect(500, -50, 600, 50), plant.ID, nil)
+	if err != nil {
+		t.Fatalf("FillRegion: %v", err)
+	}
+	if len(created) != 0 {
+		t.Errorf("filled %d plops for a region outside the bed, want 0: %+v", len(created), created)
+	}
+}
+
 // seedFillBed makes a plantable bed of the given size centered in a big garden.
 func seedFillBed(t *testing.T, s *Service, owner, gardenID int64, w, h float64) *domain.GardenObject {
 	t.Helper()
@@ -99,16 +260,24 @@ func TestFillRegionDeterministicPacking(t *testing.T) {
 	if err != nil {
 		t.Fatalf("FillRegion: %v", err)
 	}
-	// Hex lattice on [-30,30]² at pitch 30, rows ~26 apart → 4 plops (2 rows × 2).
+	// Hex lattice on [-30,30]² at pitch 30, rows ~26 apart, centered: a row of 2
-	if len(created) != 4 {
+	// (x=±15), then a staggered row of 1 (x=0) → 3 plops.
-		t.Fatalf("filled %d plops, want 4 (60×60 bed, radius 15)", len(created))
+	//
+	// This was 4 while the lattice was anchored at the min corner, and the fourth
+	// sat at x=30 — centred ON the east edge, so half of it lay outside the bed,
+	// well past the half-spacing (5cm here) the rule allows. Packing one fewer
+	// plop is the point of the fix, not a regression in it.
+	if len(created) != 3 {
+		t.Fatalf("filled %d plops, want 3 (60×60 bed, radius 15)", len(created))
 	}
 	for _, p := range created {
 		if p.RadiusCM != 15 || p.PlantedAt == nil || p.DerivedCount < 1 {
 			t.Errorf("unexpected created plop: %+v", p)
 		}
-		if p.XCM < -30 || p.XCM > 30 || p.YCM < -30 || p.YCM > 30 {
+		// This bed fits its lattice exactly, so nothing should need to overhang.
-			t.Errorf("plop center out of bed bounds: %+v", p)
+		if p.XCM-p.RadiusCM < -30 || p.XCM+p.RadiusCM > 30 ||
+			p.YCM-p.RadiusCM < -30 || p.YCM+p.RadiusCM > 30 {
+			t.Errorf("plop overhangs a bed it fits inside: %+v", p)
 		}
 	}
 

internal/service/revisions.go

@@ -6,6 +6,7 @@ import (
 	"errors"
 	"fmt"
 	"log/slog"
+	"sort"
 	"sync"
 
 	"gitea.stevedudenhoeffer.com/steve/pansy/internal/domain"
@@ -99,7 +100,8 @@ type ChangeSetOptions struct {
 	Source string
 	// Summary is the one-line description shown in the history list.
 	Summary string
-	// AgentRunID joins this change set back to the executus run that produced it.
+	// AgentRunID joins this change set back to the agent run that produced it,
+	// so a change in the history list can be correlated with the run's log lines.
 	AgentRunID *string
 }
 
@@ -148,11 +150,19 @@ func partialSummary(summary string) string {
 // commitScope writes a scope's buffered revisions as one change set. revertsID is
 // set only by RevertChangeSet. A scope with no revisions writes nothing — an
 // operation that changed nothing doesn't belong in history.
+//
+// The write is DETACHED FROM CANCELLATION, always, and that belongs here rather
+// than at each call site so no caller can be the one that forgets. By the time a
+// commit runs, the data changes it describes have already been written — so
+// cancelling it cannot undo anything. It can only lose the record of what
+// happened and leave real changes with no way to undo them, which is the one
+// thing the whole change-set design exists to prevent.
 func (s *Service) commitScope(ctx context.Context, sc *changeScope, revertsID *int64) (*domain.ChangeSet, error) {
 	revs := sc.taken()
 	if len(revs) == 0 {
 		return nil, nil
 	}
+	ctx = context.WithoutCancel(ctx)
 	return s.store.WriteChangeSet(ctx, &domain.ChangeSet{
 		GardenID:   sc.gardenID,
 		ActorID:    sc.actorID,
@@ -193,9 +203,13 @@ func (s *Service) record(ctx context.Context, gardenID, actorID int64, summary s
 		sc.append(revs)
 		return
 	}
-	if _, err := s.store.WriteChangeSet(ctx, &domain.ChangeSet{
+	// Auto-scope: one operation, its own change set. Written through the same
-		GardenID: gardenID, ActorID: actorID, Source: domain.SourceUI, Summary: summary,
+	// detached path as everything else — a REST client that hangs up right after
-	}, revs); err != nil {
+	// its PATCH landed must not leave that change without history, and this is
+	// the path virtually every mutation takes.
+	auto := &changeScope{gardenID: gardenID, actorID: actorID, source: domain.SourceUI, summary: summary}
+	auto.append(revs)
+	if _, err := s.commitScope(ctx, auto, nil); err != nil {
 		slog.Error("service: record change set", "error", err, "garden", gardenID, "summary", summary)
 	}
 }
@@ -307,6 +321,7 @@ func (s *Service) RevertChangeSet(ctx context.Context, actorID, changeSetID int6
 		if err != nil {
 			// Record what actually landed before surfacing the failure, so the
 			// partial revert is visible and undoable rather than orphaned.
+			// (commitScope detaches from cancellation itself.)
 			if _, cerr := s.commitScope(ctx, sc, &target.ID); cerr != nil {
 				slog.Error("service: partial revert could not be recorded", "error", cerr, "changeSet", changeSetID)
 			}
@@ -327,9 +342,48 @@ func (s *Service) RevertChangeSet(ctx context.Context, actorID, changeSetID int6
 	if err != nil {
 		return nil, nil, err
 	}
+	// Fill in the per-op counts. commitScope returns the freshly inserted row,
+	// which carries no tally — and a caller receiving a change set with an empty
+	// Counts cannot tell "nothing was reverted" from "the tally wasn't loaded".
+	// That ambiguity is not theoretical: it made the chat panel report a
+	// successful undo as "nothing left to undo".
+	if cs != nil {
+		cs.Counts = countRevisions(sc.taken())
+	}
 	return cs, conflicts, nil
 }
 
+// countRevisions tallies revisions by (entity type, op).
+//
+// This deliberately reproduces in Go what ListChangeSets does in SQL, because a
+// change set that has just been written has no rows to GROUP BY yet — the
+// alternative is a second round trip to count what we are already holding.
+// The parity is a real cross-layer contract, and TestRevertResultCarriesItsCounts
+// compares the two breakdowns row for row so it can't drift silently.
+func countRevisions(revs []domain.Revision) []domain.ChangeCount {
+	type key struct{ entityType, op string }
+	seen := map[key]int{}
+	order := []key{}
+	for _, r := range revs {
+		k := key{r.EntityType, r.Op}
+		if _, ok := seen[k]; !ok {
+			order = append(order, k)
+		}
+		seen[k]++
+	}
+	sort.Slice(order, func(i, j int) bool {
+		if order[i].entityType != order[j].entityType {
+			return order[i].entityType < order[j].entityType
+		}
+		return order[i].op < order[j].op
+	})
+	counts := make([]domain.ChangeCount, 0, len(order))
+	for _, k := range order {
+		counts = append(counts, domain.ChangeCount{EntityType: k.entityType, Op: k.op, N: seen[k]})
+	}
+	return counts
+}
+
 // entityKey identifies a row across the entity types a revision can name.
 type entityKey struct {
 	entityType string

internal/service/revisions_test.go

@@ -704,3 +704,192 @@ func TestClearObjectOnlyClearsWhatItSnapshotted(t *testing.T) {
 		t.Errorf("cleared %d of %d with %d revisions — all three should match", n, len(before), len(revs))
 	}
 }
+
+// TestRevertResultCarriesItsCounts — found by using the thing.
+//
+// commitScope returns the freshly inserted row, which carries no tally. A caller
+// receiving a change set with empty Counts cannot tell "nothing was reverted"
+// from "the tally wasn't loaded", and the chat panel read that ambiguity the
+// wrong way: a successful undo reported "nothing left to undo", which is the
+// worst possible thing to say about an action that just worked.
+//
+// It also pins the cross-layer contract that fix created. countRevisions tallies
+// in Go what ListChangeSets tallies in SQL, and nothing but this test stops the
+// two drifting — so it compares the FULL per-(entity, op) breakdown, not just
+// the totals, which would agree even if the groupings had diverged.
+func TestRevertResultCarriesItsCounts(t *testing.T) {
+	s := newTestService(t, openConfig())
+	owner := seedUser(t, s, "[email protected]")
+	g := seedGarden(t, s, owner)
+	bed := seedBed(t, s, owner, g.ID)
+	plant := seedOwnPlant(t, s, owner, 15)
+	ctx := context.Background()
+
+	if _, err := s.FillNamedRegion(ctx, owner, bed.ID, "all", plant.ID, nil); err != nil {
+		t.Fatalf("fill: %v", err)
+	}
+	// A second, different kind of change, so the breakdown has more than one row
+	// to get wrong.
+	if _, err := s.UpdateObject(ctx, owner, bed.ID, ObjectPatch{Name: strPtr("North Bed")}, bed.Version); err != nil {
+		t.Fatalf("rename: %v", err)
+	}
+	sets := history(t, s, owner, g.ID)
+	rename, fill := sets[0], sets[1]
+
+	undoRename, conflicts, err := s.RevertChangeSet(ctx, owner, rename.ID, domain.SourceUI)
+	if err != nil || len(conflicts) != 0 {
+		t.Fatalf("revert rename: err=%v conflicts=%+v", err, conflicts)
+	}
+	undoFill, conflicts, err := s.RevertChangeSet(ctx, owner, fill.ID, domain.SourceUI)
+	if err != nil || len(conflicts) != 0 {
+		t.Fatalf("revert fill: err=%v conflicts=%+v", err, conflicts)
+	}
+
+	for _, undo := range []*domain.ChangeSet{undoRename, undoFill} {
+		if undo == nil {
+			t.Fatal("a revert that did work returned no change set")
+		}
+		if len(undo.Counts) == 0 {
+			t.Fatalf("change set %d came back with no tally — an empty tally reads as 'nothing happened'", undo.ID)
+		}
+	}
+	// Undoing a fill deletes every plop it created, so the tallies must match.
+	if got, want := countsTotal(undoFill.Counts), countsTotal(fill.Counts); got != want {
+		t.Errorf("undo of the fill reports %d changes, want %d", got, want)
+	}
+
+	// The SQL tally and the Go tally must agree, row for row.
+	listed := history(t, s, owner, g.ID)
+	byID := map[int64][]domain.ChangeCount{}
+	for _, cs := range listed {
+		byID[cs.ID] = cs.Counts
+	}
+	for _, undo := range []*domain.ChangeSet{undoRename, undoFill} {
+		fromSQL, ok := byID[undo.ID]
+		if !ok {
+			t.Fatalf("revert %d is missing from the history list", undo.ID)
+		}
+		if !sameCounts(undo.Counts, fromSQL) {
+			t.Errorf("change set %d: revert returned %+v, the list read says %+v — countRevisions and the SQL grouping have drifted",
+				undo.ID, undo.Counts, fromSQL)
+		}
+	}
+}
+
+// sameCounts compares two tallies regardless of order.
+func sameCounts(a, b []domain.ChangeCount) bool {
+	if len(a) != len(b) {
+		return false
+	}
+	index := func(cs []domain.ChangeCount) map[string]int {
+		m := map[string]int{}
+		for _, c := range cs {
+			m[c.EntityType+"/"+c.Op] = c.N
+		}
+		return m
+	}
+	ia, ib := index(a), index(b)
+	for k, v := range ia {
+		if ib[k] != v {
+			return false
+		}
+	}
+	return true
+}
+
+// countsTotal sums a tally — the server-side twin of the client's totalChanges.
+func countsTotal(counts []domain.ChangeCount) int {
+	n := 0
+	for _, c := range counts {
+		n += c.N
+	}
+	return n
+}
+
+// TestSucceededTurnRecordsEvenIfTheCallerWentAway is the production bug.
+//
+// The failure path was detached from cancellation; the SUCCESS path was not. An
+// agent turn whose client disconnects can still COMPLETE — and then the success
+// path committed with a dead context, the write failed, and real changes were
+// left with no change set and no way to undo them. Found live with 18 plantings
+// behind no history at all.
+//
+// Nothing about a commit needs the caller to still be there: by the time it
+// runs, the data it describes has already been written.
+func TestSucceededTurnRecordsEvenIfTheCallerWentAway(t *testing.T) {
+	s := newTestService(t, openConfig())
+	owner := seedUser(t, s, "[email protected]")
+	g := seedGarden(t, s, owner)
+	bed := seedBed(t, s, owner, g.ID)
+	plant := seedOwnPlant(t, s, owner, 15)
+
+	ctx, cancel := context.WithCancel(context.Background())
+	before := len(history(t, s, owner, g.ID))
+
+	// fn does its work and SUCCEEDS, but the caller goes away before it returns.
+	cs, err := s.WithChangeSet(ctx, owner, g.ID, ChangeSetOptions{
+		Source: domain.SourceAgent, Summary: "plant beans in the second bed",
+	}, func(ctx context.Context) error {
+		if _, err := s.FillNamedRegion(ctx, owner, bed.ID, "all", plant.ID, nil); err != nil {
+			return err
+		}
+		cancel() // the client disconnects, mid-turn, after the work landed
+		return nil
+	})
+	if err != nil {
+		t.Fatalf("WithChangeSet: %v", err)
+	}
+	if cs == nil {
+		t.Fatal("the turn changed things but produced no change set")
+	}
+
+	after := history(t, s, owner, g.ID)
+	if len(after) != before+1 {
+		t.Fatalf("recorded %d change sets, want 1", len(after)-before)
+	}
+	if after[0].Summary != "plant beans in the second bed" {
+		t.Errorf("summary = %q; a completed turn shouldn't be marked partial", after[0].Summary)
+	}
+	// And it's undoable, which is the entire point.
+	if _, conflicts, err := s.RevertChangeSet(context.Background(), owner, cs.ID, domain.SourceUI); err != nil || len(conflicts) != 0 {
+		t.Fatalf("the recorded turn should be undoable: err=%v conflicts=%+v", err, conflicts)
+	}
+	active, _ := s.store.ListActivePlantingsForObject(context.Background(), bed.ID)
+	if len(active) != 0 {
+		t.Errorf("%d plantings survived the undo", len(active))
+	}
+}
+
+// TestAutoScopedMutationRecordsEvenIfTheCallerWentAway — the same rule for the
+// path virtually every mutation takes.
+//
+// A plain REST PATCH auto-scopes into its own change set. If the client hangs up
+// between the row landing and the change set being written, that change is
+// orphaned exactly as an agent turn's was — and this path is used far more.
+func TestAutoScopedMutationRecordsEvenIfTheCallerWentAway(t *testing.T) {
+	s := newTestService(t, openConfig())
+	owner := seedUser(t, s, "[email protected]")
+	g := seedGarden(t, s, owner)
+	bed := seedBed(t, s, owner, g.ID)
+	before := len(history(t, s, owner, g.ID))
+
+	// The row write and the history write share this context; cancelling after
+	// the mutation returns is the client hanging up mid-request.
+	ctx, cancel := context.WithCancel(context.Background())
+	if _, err := s.UpdateObject(ctx, owner, bed.ID, ObjectPatch{XCM: f64Ptr(300)}, bed.Version); err != nil {
+		t.Fatalf("UpdateObject: %v", err)
+	}
+	cancel()
+
+	after := history(t, s, owner, g.ID)
+	if len(after) != before+1 {
+		t.Fatalf("recorded %d change sets, want 1 — the move is otherwise un-undoable", len(after)-before)
+	}
+	if _, conflicts, err := s.RevertChangeSet(context.Background(), owner, after[0].ID, domain.SourceUI); err != nil || len(conflicts) != 0 {
+		t.Fatalf("the recorded move should be undoable: err=%v conflicts=%+v", err, conflicts)
+	}
+	back, _ := s.store.GetObject(context.Background(), bed.ID)
+	if back.XCM != bed.XCM {
+		t.Errorf("undo left x at %v, want %v", back.XCM, bed.XCM)
+	}
+}

internal/store/agent.go

@@ -0,0 +1,95 @@
+package store
+
+import (
+	"context"
+	"fmt"
+
+	"gitea.stevedudenhoeffer.com/steve/pansy/internal/domain"
+)
+
+// agentMessageColumns lists agent_messages columns in scan order.
+const agentMessageColumns = `id, conversation_id, role, body, change_set_id, created_at`
+
+// EnsureConversation returns the (garden, user) conversation id, creating it on
+// first use. One thread per person per garden: two people editing a shared
+// garden hold separate dialogues, and merging them would put words in someone's
+// mouth.
+func (d *DB) EnsureConversation(ctx context.Context, gardenID, userID int64) (int64, error) {
+	var id int64
+	err := d.sql.QueryRowContext(ctx,
+		`SELECT id FROM agent_conversations WHERE garden_id = ? AND user_id = ?`,
+		gardenID, userID).Scan(&id)
+	if err == nil {
+		return id, nil
+	}
+	if err := d.sql.QueryRowContext(ctx,
+		`INSERT INTO agent_conversations (garden_id, user_id) VALUES (?, ?)
+		 ON CONFLICT (garden_id, user_id) DO UPDATE SET updated_at = updated_at
+		 RETURNING id`,
+		gardenID, userID).Scan(&id); err != nil {
+		return 0, fmt.Errorf("store: ensure conversation: %w", err)
+	}
+	return id, nil
+}
+
+// AppendAgentMessage records one side of an exchange.
+func (d *DB) AppendAgentMessage(ctx context.Context, m *domain.AgentMessage) (*domain.AgentMessage, error) {
+	var out domain.AgentMessage
+	if err := d.sql.QueryRowContext(ctx,
+		`INSERT INTO agent_messages (conversation_id, role, body, change_set_id)
+		 VALUES (?, ?, ?, ?)
+		 RETURNING `+agentMessageColumns,
+		m.ConversationID, m.Role, m.Body, m.ChangeSetID,
+	).Scan(&out.ID, &out.ConversationID, &out.Role, &out.Body, &out.ChangeSetID, &out.CreatedAt); err != nil {
+		return nil, fmt.Errorf("store: append agent message: %w", err)
+	}
+	// Touch the conversation so a "most recent thread" read is possible later.
+	if _, err := d.sql.ExecContext(ctx,
+		`UPDATE agent_conversations SET updated_at = strftime('%Y-%m-%dT%H:%M:%SZ', 'now') WHERE id = ?`,
+		m.ConversationID); err != nil {
+		return nil, fmt.Errorf("store: touch conversation: %w", err)
+	}
+	return &out, nil
+}
+
+// ListAgentMessages returns the last `limit` messages of a conversation in
+// chronological order.
+//
+// Reading the TAIL rather than the head is the point: an old opening exchange is
+// the least useful context for "now do the same to the north bed", and an
+// uncapped read would grow without bound over a season.
+func (d *DB) ListAgentMessages(ctx context.Context, conversationID int64, limit int) ([]domain.AgentMessage, error) {
+	rows, err := d.sql.QueryContext(ctx,
+		`SELECT `+agentMessageColumns+` FROM (
+		   SELECT `+agentMessageColumns+` FROM agent_messages
+		    WHERE conversation_id = ? ORDER BY id DESC LIMIT ?
+		 ) ORDER BY id`,
+		conversationID, limit)
+	if err != nil {
+		return nil, fmt.Errorf("store: list agent messages: %w", err)
+	}
+	defer rows.Close()
+
+	msgs := []domain.AgentMessage{}
+	for rows.Next() {
+		var m domain.AgentMessage
+		if err := rows.Scan(&m.ID, &m.ConversationID, &m.Role, &m.Body, &m.ChangeSetID, &m.CreatedAt); err != nil {
+			return nil, fmt.Errorf("store: scan agent message: %w", err)
+		}
+		msgs = append(msgs, m)
+	}
+	if err := rows.Err(); err != nil {
+		return nil, fmt.Errorf("store: iterate agent messages: %w", err)
+	}
+	return msgs, nil
+}
+
+// DeleteConversation clears a thread (the messages cascade).
+func (d *DB) DeleteConversation(ctx context.Context, gardenID, userID int64) error {
+	if _, err := d.sql.ExecContext(ctx,
+		`DELETE FROM agent_conversations WHERE garden_id = ? AND user_id = ?`,
+		gardenID, userID); err != nil {
+		return fmt.Errorf("store: delete conversation: %w", err)
+	}
+	return nil
+}

internal/store/instance_settings.go

@@ -0,0 +1,81 @@
+package store
+
+import (
+	"context"
+	"database/sql"
+	"errors"
+	"fmt"
+
+	"gitea.stevedudenhoeffer.com/steve/pansy/internal/domain"
+)
+
+// The instance_settings row is seeded by migration 0010 and there is exactly one
+// (CHECK id = 1), so reads never branch on existence and writes never insert.
+
+const instanceSettingsColumns = `agent_model, agent_enabled, version, updated_at`
+
+// scanInstanceSettings reads the single settings row. agent_enabled is a nullable
+// INTEGER (NULL = inherit env), so it is scanned through sql.NullInt64.
+func scanInstanceSettings(s scanner) (*domain.InstanceSettings, error) {
+	var (
+		out     domain.InstanceSettings
+		enabled sql.NullInt64
+	)
+	if err := s.Scan(&out.AgentModel, &enabled, &out.Version, &out.UpdatedAt); err != nil {
+		return nil, err
+	}
+	if enabled.Valid {
+		b := enabled.Int64 != 0
+		out.AgentEnabled = &b
+	}
+	return &out, nil
+}
+
+// GetInstanceSettings returns the single settings row.
+func (d *DB) GetInstanceSettings(ctx context.Context) (*domain.InstanceSettings, error) {
+	s, err := scanInstanceSettings(d.sql.QueryRowContext(ctx,
+		`SELECT `+instanceSettingsColumns+` FROM instance_settings WHERE id = 1`))
+	if errors.Is(err, sql.ErrNoRows) {
+		// The migration seeds this row, so its absence is a broken database, not a
+		// normal "not found" the caller should paper over.
+		return nil, fmt.Errorf("store: instance_settings row missing (migration 0010 not applied?)")
+	}
+	if err != nil {
+		return nil, fmt.Errorf("store: get instance settings: %w", err)
+	}
+	return s, nil
+}
+
+// UpdateInstanceSettings applies a version-guarded update to the single row,
+// following the same optimistic-concurrency contract as every mutable resource:
+// the updated row on success, (current row, ErrVersionConflict) on a version
+// mismatch. There is no ErrNotFound path — the row always exists.
+//
+// agentEnabled is nil to store SQL NULL (inherit env), or a pointer to store an
+// explicit 0/1.
+func (d *DB) UpdateInstanceSettings(ctx context.Context, s *domain.InstanceSettings) (*domain.InstanceSettings, error) {
+	var enabled any
+	if s.AgentEnabled != nil {
+		enabled = boolToInt(*s.AgentEnabled)
+	}
+	updated, err := scanInstanceSettings(d.sql.QueryRowContext(ctx,
+		`UPDATE instance_settings
+		    SET agent_model = ?, agent_enabled = ?,
+		        version = version + 1,
+		        updated_at = strftime('%Y-%m-%dT%H:%M:%SZ', 'now')
+		  WHERE id = 1 AND version = ?
+		  RETURNING `+instanceSettingsColumns,
+		s.AgentModel, enabled, s.Version,
+	))
+	if errors.Is(err, sql.ErrNoRows) {
+		current, gerr := d.GetInstanceSettings(ctx)
+		if gerr != nil {
+			return nil, gerr
+		}
+		return current, domain.ErrVersionConflict
+	}
+	if err != nil {
+		return nil, fmt.Errorf("store: update instance settings: %w", err)
+	}
+	return updated, nil
+}

internal/store/migrations/0009_agent_chat.sql

@@ -0,0 +1,36 @@
+-- Agent conversations (#56): multi-turn chat with the garden assistant.
+--
+-- "Now do the same to the north bed" needs the previous exchange, and history
+-- held only by the client would be lost on a refresh — which is exactly when
+-- someone reloads to see whether the agent's change actually landed.
+--
+-- One conversation per (user, garden). Two people editing a shared garden get
+-- separate threads: a conversation is a dialogue with one person, and merging
+-- them would put words in someone's mouth.
+CREATE TABLE agent_conversations (
+    id         INTEGER PRIMARY KEY,
+    garden_id  INTEGER NOT NULL REFERENCES gardens (id) ON DELETE CASCADE,
+    user_id    INTEGER NOT NULL REFERENCES users (id)   ON DELETE CASCADE,
+    created_at TEXT    NOT NULL DEFAULT (strftime('%Y-%m-%dT%H:%M:%SZ', 'now')),
+    updated_at TEXT    NOT NULL DEFAULT (strftime('%Y-%m-%dT%H:%M:%SZ', 'now')),
+    UNIQUE (garden_id, user_id)
+);
+
+-- Only the user/assistant TEXT of each turn is kept, not the full model
+-- transcript with its tool calls. Continuity needs what was said and what came
+-- back; replaying a stored tool call would be replaying a decision made against
+-- a garden that has since moved on. It also keeps the schema free of majordomo's
+-- message shape, which is a dependency's business, not the database's.
+CREATE TABLE agent_messages (
+    conversation_id INTEGER NOT NULL REFERENCES agent_conversations (id) ON DELETE CASCADE,
+    id              INTEGER PRIMARY KEY,
+    role            TEXT    NOT NULL CHECK (role IN ('user', 'assistant')),
+    body            TEXT    NOT NULL,
+    -- The change set this turn produced, so the panel can offer Undo on the
+    -- message that caused it. SET NULL rather than CASCADE: losing the history
+    -- entry must not delete what was said about it.
+    change_set_id   INTEGER REFERENCES change_sets (id) ON DELETE SET NULL,
+    created_at      TEXT    NOT NULL DEFAULT (strftime('%Y-%m-%dT%H:%M:%SZ', 'now'))
+);
+
+CREATE INDEX idx_agent_messages_conversation ON agent_messages (conversation_id, id);

internal/store/migrations/0010_instance_settings.sql

@@ -0,0 +1,35 @@
+-- Instance settings (#79): the first configuration that lives in the database
+-- rather than the environment.
+--
+-- Until now every preference hung off a garden or object row; this is pansy's
+-- first INSTANCE-level state. A single-row table (CHECK id = 1) is the least
+-- surprising shape for "there is exactly one of these" — a key/value table would
+-- invite typo'd keys and lose the column types.
+--
+-- Only NON-SECRET agent settings live here. OLLAMA_CLOUD_API_KEY stays in the
+-- environment on purpose: a copy in SQLite would land in every backup and in the
+-- blast radius of the undo history. An admin can change WHICH model runs, not
+-- WHOSE account pays for it.
+--
+-- Both agent columns are "inherit from env unless set":
+--   * agent_model = '' means fall back to PANSY_AGENT_MODEL, then the built-in
+--     default. So an instance that never opens Settings behaves exactly as it
+--     did before this migration, and the documented env var keeps working.
+--   * agent_enabled is NULLABLE: NULL means inherit PANSY_AGENT_ENABLED's
+--     behaviour (on when a key is present), 0/1 is an explicit override. A plain
+--     boolean couldn't tell "admin hasn't touched this" from "admin turned it
+--     off", and those must deploy differently.
+--
+-- version drives the same optimistic-concurrency 409 every other mutable row
+-- uses, so two admins editing at once conflict rather than clobber.
+CREATE TABLE instance_settings (
+    id            INTEGER PRIMARY KEY CHECK (id = 1),
+    agent_model   TEXT    NOT NULL DEFAULT '',
+    agent_enabled INTEGER          CHECK (agent_enabled IN (0, 1)),
+    version       INTEGER NOT NULL DEFAULT 1,
+    updated_at    TEXT    NOT NULL DEFAULT (strftime('%Y-%m-%dT%H:%M:%SZ', 'now'))
+);
+
+-- Seed the single row so every read is a plain SELECT with no "does it exist
+-- yet" branch. Inherits everything from the environment out of the box.
+INSERT INTO instance_settings (id, agent_model, agent_enabled) VALUES (1, '', NULL);

internal/store/revisions.go

@@ -119,7 +119,9 @@ func (d *DB) ListChangeSets(ctx context.Context, gardenID int64, limit, offset i
 		return sets, nil
 	}
 
-	// One grouped query for the whole page rather than N per-row counts.
+	// One grouped query for the whole page rather than N per-row counts. The
+	// service's countRevisions mirrors this grouping for change sets it has just
+	// written; TestRevertResultCarriesItsCounts holds the two in step.
 	countRows, err := d.sql.QueryContext(ctx,
 		`SELECT change_set_id, entity_type, op, COUNT(*)
 		   FROM revisions

web/src/components/layout/AppShell.tsx

@@ -53,6 +53,19 @@ export function AppShell() {
                   {l.label}
                 </Link>
               ))}
+            {/* Settings is admin-only, matching the server's requireAdmin gate.
+                A non-admin who typed /settings still gets a 403 from the API — the
+                hidden link is convenience, not the security boundary. */}
+            {user?.isAdmin && (
+              <Link
+                to="/settings"
+                className={navLinkBase}
+                activeProps={{ className: navLinkActive }}
+                inactiveProps={{ className: navLinkInactive }}
+              >
+                Settings
+              </Link>
+            )}
           </div>
 
           {user ? (

web/src/components/ui/toast.tsx

@@ -17,9 +17,16 @@ interface ToastState {
 
 let nextId = 1
 
+// Error toasts no longer auto-dismiss (#85), so a burst of failures could grow
+// the stack without bound and push older ones off-screen. Cap it: keep the most
+// recent MAX_TOASTS and drop the oldest, so the newest — the one that just
+// happened — is always visible.
+const MAX_TOASTS = 4
+
 export const useToastStore = create<ToastState>((set) => ({
   toasts: [],
-  push: (message, tone = 'info') => set((s) => ({ toasts: [...s.toasts, { id: nextId++, message, tone }] })),
+  push: (message, tone = 'info') =>
+    set((s) => ({ toasts: [...s.toasts, { id: nextId++, message, tone }].slice(-MAX_TOASTS) })),
   dismiss: (id) => set((s) => ({ toasts: s.toasts.filter((t) => t.id !== id) })),
 }))
 
@@ -32,21 +39,34 @@ export const toast = {
 // Param is `item`, not `toast`, so it doesn't shadow the module's `toast` export.
 function ToastItem({ item }: { item: Toast }) {
   const dismiss = useToastStore((s) => s.dismiss)
+  const isError = item.tone === 'error'
   useEffect(() => {
+    // Error toasts are the primary report that a mutation failed, so they do NOT
+    // auto-dismiss — a user who looked away at second 4 would otherwise lose the
+    // only notice, with nothing to retrieve (#85). Info toasts still time out.
+    if (isError) return
     const t = setTimeout(() => dismiss(item.id), 4000)
     return () => clearTimeout(t)
-  }, [item.id, dismiss])
+  }, [item.id, dismiss, isError])
   return (
     <div
-      role={item.tone === 'error' ? 'alert' : 'status'}
+      role={isError ? 'alert' : 'status'}
       className={cn(
-        'pointer-events-auto rounded-md border px-3 py-2 text-sm shadow-md',
+        'pointer-events-auto flex items-start gap-2 rounded-md border px-3 py-2 text-sm shadow-md',
-        item.tone === 'error'
+        isError
           ? 'border-red-500/40 bg-red-500/10 text-red-700 dark:text-red-300'
           : 'border-border bg-surface text-fg',
       )}
     >
-      {item.message}
+      <span className="flex-1">{item.message}</span>
+      <button
+        type="button"
+        onClick={() => dismiss(item.id)}
+        aria-label="Dismiss"
+        className="-mr-1 shrink-0 rounded px-1 text-current opacity-60 outline-none hover:opacity-100 focus-visible:ring-2 focus-visible:ring-current/40"
+      >
+        ✕
+      </button>
     </div>
   )
 }

web/src/editor/ChatPanel.tsx

@@ -0,0 +1,239 @@
+import { useEffect, useRef, useState, type ReactNode } from 'react'
+import { Alert } from '@/components/ui/Alert'
+import { errorMessage } from '@/lib/api'
+import { Button } from '@/components/ui/Button'
+import { TextArea } from '@/components/ui/TextArea'
+import { cn } from '@/lib/cn'
+import {
+  describeStep,
+  streamChat,
+  useAgentHistory,
+  useAgentRefresh,
+  useClearAgentHistory,
+  type AgentStep,
+  type AgentTurn,
+} from '@/lib/agent'
+import { useUndo } from '@/lib/history'
+import { UndoButton } from './UndoButton'
+
+/**
+ * Talk to the garden assistant, in the editor beside the canvas.
+ *
+ * Here rather than on its own page because watching the garden change as the
+ * agent works IS the confirmation — which is what makes acting without asking
+ * first tolerable. It also means the agent never has to guess which garden you
+ * mean.
+ */
+export function ChatPanel({ gardenId, canEdit }: { gardenId: number; canEdit: boolean }) {
+  const history = useAgentHistory(gardenId, true)
+  const clear = useClearAgentHistory(gardenId)
+  const refresh = useAgentRefresh(gardenId)
+  const undo = useUndo(gardenId)
+
+  const [input, setInput] = useState('')
+  // The turn in flight: what we sent, the steps so far, and how it ended.
+  const [pending, setPending] = useState<{ message: string; steps: AgentStep[] } | null>(null)
+  const [error, setError] = useState<string | null>(null)
+  const [warning, setWarning] = useState<string | null>(null)
+  const abort = useRef<AbortController | null>(null)
+  const bottom = useRef<HTMLDivElement>(null)
+
+  // Deliberately NOT aborted on unmount. Selecting an object auto-switches the
+  // rail to the inspector, so aborting here would mean clicking the canvas
+  // mid-turn silently killed the turn — and the canvas is exactly what you're
+  // meant to be watching. The request continues, the exchange is persisted
+  // server-side, and coming back to this tab shows it. Only Stop aborts, and
+  // even that only stops us READING: the turn keeps running server-side, which
+  // is why its work still lands in History either way.
+
+  // Follow the conversation as it grows, including mid-turn as steps arrive.
+  useEffect(() => {
+    bottom.current?.scrollIntoView({ behavior: 'smooth', block: 'end' })
+  }, [history.data, pending])
+
+  const send = () => {
+    const message = input.trim()
+    if (!message || pending) return
+    setInput('')
+    setError(null)
+    setWarning(null)
+    setPending({ message, steps: [] })
+
+    const controller = new AbortController()
+    abort.current = controller
+
+    void streamChat(
+      gardenId,
+      message,
+      {
+        onStep: (step) => {
+          setPending((p) => (p ? { ...p, steps: [...p.steps, step] } : p))
+          // The canvas updating under the conversation is the whole point of
+          // putting the chat here, so refresh it as each step lands — but only
+          // it: nothing else can have changed until the turn commits.
+          refresh.canvas()
+        },
+        onDone: (turn: AgentTurn) => {
+          setPending(null)
+          refresh.everything()
+          if (turn.truncated) {
+            setError('That turned into more steps than I should take at once — check what changed before continuing.')
+          }
+        },
+        onWarning: setWarning,
+        onError: (message) => {
+          setPending(null)
+          setError(message)
+          // Something may still have landed before it failed.
+          refresh.everything()
+        },
+      },
+      controller.signal,
+    )
+  }
+
+  const messages = history.data ?? []
+
+  return (
+    <div className="flex h-full flex-col gap-3">
+      <div className="flex items-center justify-between gap-2">
+        <h2 className="text-sm font-semibold text-fg">Assistant</h2>
+        {messages.length > 0 && (
+          <button
+            type="button"
+            onClick={() =>
+              clear.mutate(undefined, {
+                onError: (err) => setError(errorMessage(err, "Couldn't clear the conversation.")),
+              })
+            }
+            disabled={clear.isPending}
+            className="rounded px-1.5 py-0.5 text-xs text-muted outline-none hover:text-fg focus-visible:ring-2 focus-visible:ring-accent/40"
+          >
+            Start over
+          </button>
+        )}
+      </div>
+
+      {!canEdit && (
+        <p className="rounded-md bg-border/40 px-2 py-1 text-xs text-muted">
+          You can only view this garden, so the assistant can't change anything in it.
+        </p>
+      )}
+
+      <div className="flex min-h-0 flex-1 flex-col gap-2 overflow-y-auto">
+        {history.isPending && <p className="text-sm text-muted">Loading the conversation…</p>}
+        {/* A failed load rendering as an empty thread would look like the
+            conversation had been lost, which is a much worse thing to believe. */}
+        {history.isError && (
+          <Alert>{errorMessage(history.error, "Couldn't load the conversation.")}</Alert>
+        )}
+
+        {history.isSuccess && messages.length === 0 && !pending && (
+          <p className="text-sm text-muted">
+            Ask for what you want and it'll do it — “change the garlic bed to cucumbers this year”, “fill the
+            north half of the west bed with beans”, “what's in here?”. Everything it does lands as one change you
+            can undo.
+          </p>
+        )}
+
+        {messages.map((m) => (
+          <Bubble key={m.id} role={m.role} body={m.body}>
+            {/* Undo on the turn itself, so the common case never involves
+                opening the History panel. Same hook #49 uses, not a second
+                implementation. */}
+            {m.role === 'assistant' && m.changeSetId != null && canEdit && (
+              <UndoButton changeSet={{ id: m.changeSetId }} undo={undo} className="mt-1 items-start" />
+            )}
+          </Bubble>
+        ))}
+
+        {pending && (
+          <>
+            <Bubble role="user" body={pending.message} />
+            <div className="rounded-lg border border-border px-2.5 py-2 text-sm">
+              <ol className="flex flex-col gap-0.5">
+                {pending.steps.map((s) => (
+                  <li key={s.index} className="text-xs text-muted">
+                    {describeStep(s)}…
+                  </li>
+                ))}
+              </ol>
+              <p className="mt-1 flex items-center gap-1.5 text-xs text-muted">
+                <span className="inline-block h-1.5 w-1.5 animate-pulse rounded-full bg-accent" />
+                {pending.steps.length === 0 ? 'Thinking…' : 'Working…'}
+              </p>
+            </div>
+          </>
+        )}
+
+        {warning && <Alert tone="info">{warning}</Alert>}
+        {error && <Alert>{error}</Alert>}
+        <div ref={bottom} />
+      </div>
+
+      <div className="flex flex-col gap-2 border-t border-border pt-2">
+        <TextArea
+          label="Message"
+          name="agentMessage"
+          rows={2}
+          placeholder="Change the garlic bed to cucumbers this year"
+          value={input}
+          disabled={!!pending}
+          onChange={(e) => setInput(e.target.value)}
+          onKeyDown={(e) => {
+            // Enter sends, Shift+Enter breaks the line — the convention every
+            // chat box uses, and typing a newline by accident mid-thought is a
+            // worse failure than the reverse.
+            if (e.key === 'Enter' && !e.shiftKey) {
+              e.preventDefault()
+              send()
+            }
+          }}
+        />
+        <div className="flex justify-end gap-2">
+          {pending && (
+            <Button
+              variant="ghost"
+              className="px-2 py-1 text-xs"
+              onClick={() => {
+                abort.current?.abort()
+                setPending(null)
+                refresh.everything()
+              }}
+            >
+              Stop
+            </Button>
+          )}
+          <Button className="px-3 py-1.5 text-sm" disabled={!!pending || input.trim() === ''} onClick={send}>
+            {pending ? 'Working…' : 'Send'}
+          </Button>
+        </div>
+      </div>
+    </div>
+  )
+}
+
+function Bubble({
+  role,
+  body,
+  children,
+}: {
+  role: 'user' | 'assistant'
+  body: string
+  children?: ReactNode
+}) {
+  const mine = role === 'user'
+  return (
+    <div className={cn('flex flex-col', mine ? 'items-end' : 'items-start')}>
+      <div
+        className={cn(
+          'max-w-[90%] whitespace-pre-wrap rounded-lg px-2.5 py-2 text-sm',
+          mine ? 'bg-accent/15 text-fg' : 'border border-border text-fg',
+        )}
+      >
+        {body}
+      </div>
+      {children}
+    </div>
+  )
+}

web/src/editor/ClearBedModal.tsx

@@ -5,23 +5,25 @@ import { useClearObject } from '@/lib/objects'
 /** Confirm clearing every active plop from a focused bed (soft-remove — the rows
  *  are kept with removed_at, so history survives). */
 export function ClearBedModal({
+  objectId,
   objectName,
-  plops,
+  plopCount,
   gardenId,
   onClose,
 }: {
+  objectId: number
   objectName: string
-  plops: { id: number; version: number }[]
+  plopCount: number
   gardenId: number
   onClose: () => void
 }) {
   const clear = useClearObject(gardenId)
-  const n = plops.length
   return (
     <Modal title="Clear bed" onClose={onClose} busy={clear.isPending}>
       <div className="flex flex-col gap-4">
         <p className="text-sm text-muted">
-          Remove all <span className="font-medium text-fg">{n}</span> {n === 1 ? 'plant' : 'plants'} from{' '}
+          Remove all <span className="font-medium text-fg">{plopCount}</span>{' '}
+          {plopCount === 1 ? 'plant' : 'plants'} from{' '}
           <span className="font-medium text-fg">{objectName}</span>? They're marked removed but kept in history.
         </p>
         <div className="flex justify-end gap-2">
@@ -31,8 +33,8 @@ export function ClearBedModal({
           <Button
             type="button"
             variant="danger"
-            disabled={clear.isPending || n === 0}
+            disabled={clear.isPending || plopCount === 0}
-            onClick={() => clear.mutate(plops, { onSuccess: onClose })}
+            onClick={() => clear.mutate(objectId, { onSuccess: onClose })}
           >
             {clear.isPending ? 'Clearing…' : 'Clear bed'}
           </Button>

web/src/editor/HistoryPanel.tsx

@@ -116,7 +116,7 @@ function HistoryEntry({
           <UndoButton
             changeSet={changeSet}
             undo={undo}
-            className="w-32 shrink-0"
+            className="w-32 shrink-0 text-right"
             label={reverted ? 'Undo again' : 'Undo'}
           />
         )}

web/src/editor/JournalPanel.tsx

@@ -16,6 +16,11 @@ import {
 import type { EditorObject } from './types'
 import { objectDisplayName } from './kinds'
 
+// Shared styling for the small From/To date inputs, so the two stay in step and
+// don't drift from each other.
+const dateInputClass =
+  'rounded-md border border-border bg-surface px-1.5 py-1 text-fg outline-none focus-visible:ring-2 focus-visible:ring-accent/40'
+
 /**
  * The garden's journal: write an entry, read the season back.
  *
@@ -42,7 +47,15 @@ export function JournalPanel({
   scopeObjectId: number | null
   onScopeChange: (id: number | null) => void
 }) {
-  const filter = scopeObjectId != null ? { objectId: scopeObjectId } : {}
+  // Date-range narrowing (#85): the backend and JournalFilter already supported
+  // from/to; they just had no UI. Empty inputs don't filter.
+  const [from, setFrom] = useState('')
+  const [to, setTo] = useState('')
+  const filter = {
+    ...(scopeObjectId != null ? { objectId: scopeObjectId } : {}),
+    ...(from ? { from } : {}),
+    ...(to ? { to } : {}),
+  }
   const journal = useJournal(gardenId, filter)
   const entries = journal.data?.pages.flatMap((p) => p.entries) ?? []
   const scopedObject = objects.find((o) => o.id === scopeObjectId) ?? null
@@ -70,6 +83,41 @@ export function JournalPanel({
         )}
       </div>
 
+      <div className="flex items-center gap-2 text-xs text-muted">
+        <label className="flex items-center gap-1">
+          <span>From</span>
+          <input
+            type="date"
+            value={from}
+            max={to || undefined}
+            onChange={(e) => setFrom(e.target.value)}
+            className={dateInputClass}
+          />
+        </label>
+        <label className="flex items-center gap-1">
+          <span>To</span>
+          <input
+            type="date"
+            value={to}
+            min={from || undefined}
+            onChange={(e) => setTo(e.target.value)}
+            className={dateInputClass}
+          />
+        </label>
+        {(from || to) && (
+          <button
+            type="button"
+            onClick={() => {
+              setFrom('')
+              setTo('')
+            }}
+            className="rounded px-1 text-muted outline-none hover:text-fg focus-visible:ring-2 focus-visible:ring-accent/40"
+          >
+            Clear
+          </button>
+        )}
+      </div>
+
       {canEdit && (
         <Composer
           gardenId={gardenId}

web/src/editor/UndoButton.tsx

@@ -1,6 +1,6 @@
 import { Button } from '@/components/ui/Button'
 import { cn } from '@/lib/cn'
-import type { ChangeSet, UndoOutcome, useUndo } from '@/lib/history'
+import type { UndoOutcome, UndoTarget, useUndo } from '@/lib/history'
 
 /**
  * Undo, plus what happened. Paired with useUndo so the history list and the
@@ -14,7 +14,7 @@ export function UndoButton({
   className,
   label = 'Undo',
 }: {
-  changeSet: ChangeSet
+  changeSet: UndoTarget
   undo: ReturnType<typeof useUndo>
   className?: string
   label?: string
@@ -43,8 +43,12 @@ const TONE_CLASS: Record<Exclude<UndoOutcome['tone'], 'pending'>, string> = {
 
 function OutcomeNote({ outcome }: { outcome: UndoOutcome }) {
   if (outcome.tone === 'pending') return null
+  // No text alignment of its own: the container decides. The history list stacks
+  // this to the right of an entry, the chat panel puts it under a left-aligned
+  // message, and a hard-coded text-right made the chat's copy read against its
+  // own column.
   return (
-    <p role="status" className={cn('text-right text-xs', TONE_CLASS[outcome.tone])}>
+    <p role="status" className={cn('text-xs', TONE_CLASS[outcome.tone])}>
       {outcome.message}
     </p>
   )

web/src/lib/agent.test.ts

@@ -0,0 +1,29 @@
+import { describe, expect, it } from 'vitest'
+import { describeStep } from './agent'
+
+describe('describeStep', () => {
+  // Raw tool names tell you the agent is busy; these tell you what it's busy
+  // DOING, which is the difference between the panel feeling alive and hung.
+  it("names tools in the app's own vocabulary", () => {
+    expect(describeStep({ index: 0, tools: ['clear_object'] })).toBe('Clearing a bed')
+    expect(describeStep({ index: 1, tools: ['find_plant'] })).toBe('Looking up a plant')
+  })
+
+  it('collapses repeats within one step', () => {
+    expect(describeStep({ index: 0, tools: ['fill_region', 'fill_region'] })).toBe('Filling a bed')
+  })
+
+  it('joins distinct tools', () => {
+    expect(describeStep({ index: 0, tools: ['clear_object', 'fill_region'] })).toBe('Clearing a bed, Filling a bed')
+  })
+
+  it('says something for a step with no tool calls', () => {
+    expect(describeStep({ index: 0, tools: [] })).toBe('Thinking')
+  })
+
+  // A tool added server-side before the client knows about it should degrade to
+  // something readable rather than showing snake_case at the user.
+  it('falls back readably for an unknown tool', () => {
+    expect(describeStep({ index: 0, tools: ['prune_orchard'] })).toBe('prune orchard')
+  })
+})

web/src/lib/agent.ts

@@ -0,0 +1,254 @@
+// The garden assistant's client (#57).
+//
+// The chat lives in the editor, not on its own page, because watching the canvas
+// change as the agent works IS the confirmation — which is what makes acting
+// without asking first tolerable. So this module's job is as much about
+// surfacing progress as it is about sending a message.
+
+import { useMutation, useQuery, useQueryClient } from '@tanstack/react-query'
+import { z } from 'zod'
+import { API_BASE, api } from './api'
+import { gardenFullKey } from './objects'
+import { historyKey } from './history'
+
+const capabilitiesSchema = z.object({ agent: z.boolean() })
+
+export const capabilitiesKey = ['capabilities'] as const
+
+/** Whether the assistant is live RIGHT NOW. Without it the panel isn't rendered
+ *  at all — a dead button is worse than no button.
+ *
+ *  Not `staleTime: Infinity` any more: an admin can turn the assistant on or off
+ *  in Settings (#79), so this must be able to change under a running page. The
+ *  settings save invalidates this key directly; the finite staleTime just means
+ *  another admin's change is picked up on the next focus/remount rather than
+ *  never. */
+export function useCapabilities() {
+  return useQuery({
+    queryKey: capabilitiesKey,
+    queryFn: async () => capabilitiesSchema.parse(await api.get('/capabilities')),
+    staleTime: 60_000,
+  })
+}
+
+export const agentMessageSchema = z.object({
+  id: z.number(),
+  conversationId: z.number(),
+  role: z.enum(['user', 'assistant']),
+  body: z.string(),
+  changeSetId: z.number().optional(),
+  createdAt: z.string(),
+})
+export type AgentMessage = z.infer<typeof agentMessageSchema>
+
+const historySchema = z.object({ messages: z.array(agentMessageSchema) })
+
+export function agentHistoryKey(gardenId: number) {
+  return ['gardens', gardenId, 'agent-history'] as const
+}
+
+/** The stored conversation, so a reload doesn't lose the thread — which is
+ *  exactly when someone reloads, to check whether a change actually landed. */
+export function useAgentHistory(gardenId: number, enabled: boolean) {
+  return useQuery({
+    queryKey: agentHistoryKey(gardenId),
+    enabled,
+    queryFn: async (): Promise<AgentMessage[]> =>
+      historySchema.parse(await api.get(`/gardens/${gardenId}/agent/history`)).messages,
+  })
+}
+
+export function useClearAgentHistory(gardenId: number) {
+  const qc = useQueryClient()
+  return useMutation({
+    mutationFn: async (): Promise<void> => {
+      await api.delete(`/gardens/${gardenId}/agent/history`)
+    },
+    onSuccess: () => qc.invalidateQueries({ queryKey: agentHistoryKey(gardenId) }),
+  })
+}
+
+/** One completed turn, as the server reports it. */
+export interface AgentTurn {
+  reply: string
+  changeSetId?: number
+  steps: number
+  truncated?: boolean
+}
+
+/** A step the model just finished, named in the app's own vocabulary. */
+export interface AgentStep {
+  index: number
+  tools: string[]
+}
+
+// What each tool is doing, in words. Raw tool names ("fill_region") tell you the
+// agent is busy; these tell you what it's busy DOING, which is the difference
+// between the panel feeling alive and feeling hung.
+const TOOL_LABELS: Record<string, string> = {
+  list_gardens: 'Looking at your gardens',
+  describe_garden: 'Reading the garden',
+  create_object: 'Adding a bed',
+  move_object: 'Moving a bed',
+  place_planting: 'Planting',
+  fill_region: 'Filling a bed',
+  clear_object: 'Clearing a bed',
+  find_plant: 'Looking up a plant',
+  create_plant: 'Adding a plant to your catalog',
+  add_journal_entry: 'Writing a journal note',
+}
+
+export function describeStep(step: AgentStep): string {
+  if (step.tools.length === 0) return 'Thinking'
+  const labels = step.tools.map((t) => TOOL_LABELS[t] ?? t.replace(/_/g, ' '))
+  // Repeated tools in one step read as one action, not a list of identical ones.
+  return [...new Set(labels)].join(', ')
+}
+
+const chatEventSchema = z.object({
+  step: z.object({ index: z.number(), tools: z.array(z.string()) }).optional(),
+  done: z
+    .object({
+      reply: z.string(),
+      changeSetId: z.number().optional(),
+      steps: z.number(),
+      truncated: z.boolean().optional(),
+    })
+    .optional(),
+  error: z.string().optional(),
+  // The turn worked but something adjacent to it didn't — currently, the
+  // exchange couldn't be saved. Dropping this on the floor would recreate
+  // exactly the silent swallow the server added it to avoid.
+  warning: z.string().optional(),
+})
+
+export interface StreamHandlers {
+  onStep: (step: AgentStep) => void
+  onDone: (turn: AgentTurn) => void
+  onError: (message: string) => void
+  /** The turn succeeded, but something alongside it didn't. */
+  onWarning?: (message: string) => void
+}
+
+/**
+ * Send a message and stream the reply.
+ *
+ * Hand-rolled rather than EventSource, which can only issue GETs — this needs a
+ * POST body. The wire format is still SSE so a proxy that understands it doesn't
+ * buffer, and so switching to EventSource later wouldn't change the server.
+ */
+export async function streamChat(
+  gardenId: number,
+  message: string,
+  handlers: StreamHandlers,
+  signal?: AbortSignal,
+): Promise<void> {
+  let res: Response
+  try {
+    res = await fetch(`${API_BASE}/agent/chat`, {
+      method: 'POST',
+      headers: { 'content-type': 'application/json' },
+      body: JSON.stringify({ gardenId, message }),
+      credentials: 'same-origin',
+      signal,
+    })
+  } catch {
+    // An abort here is the caller's own doing — Stop, or navigating away — not a
+    // failure to report back to them. The read loop below already knew this; the
+    // request path did not.
+    if (signal?.aborted) return
+    handlers.onError('Could not reach the server.')
+    return
+  }
+  if (res.status === 401) {
+    // Session expired mid-conversation (#85). Reporting this as "the assistant is
+    // unavailable" would send the user chasing a config problem that isn't there.
+    // Send them to sign in again, preserving where they were.
+    handlers.onError('Your session has expired — please sign in again.')
+    const back = encodeURIComponent(location.pathname + location.search)
+    window.location.assign(`/login?redirect=${back}`)
+    return
+  }
+  if (!res.ok || !res.body) {
+    // 503 is the assistant being turned off at runtime (#79) — the route exists,
+    // there's just no model behind it. Distinct from a 404, which would mean the
+    // whole endpoint is absent.
+    handlers.onError(
+      res.status === 503
+        ? "The assistant isn't enabled on this instance."
+        : res.status === 404
+          ? "This instance doesn't have the assistant configured."
+          : 'The assistant is not available right now.',
+    )
+    return
+  }
+
+  const reader = res.body.getReader()
+  const decoder = new TextDecoder()
+  let buffer = ''
+  for (;;) {
+    let chunk: ReadableStreamReadResult<Uint8Array>
+    try {
+      chunk = await reader.read()
+    } catch {
+      // An aborted read is the caller navigating away, not a failure worth
+      // reporting back to them.
+      if (signal?.aborted) return
+      handlers.onError('The connection dropped partway through.')
+      return
+    }
+    if (chunk.done) break
+    buffer += decoder.decode(chunk.value, { stream: true })
+
+    // SSE frames are separated by a blank line; anything after the last one is
+    // a partial frame to keep for the next chunk.
+    const frames = buffer.split('\n\n')
+    buffer = frames.pop() ?? ''
+    for (const frame of frames) {
+      const line = frame.split('\n').find((l) => l.startsWith('data:'))
+      if (!line) continue
+      // Parsed AND validated: a malformed or unexpected frame shouldn't kill a
+      // working stream, and shouldn't be trusted into the UI either.
+      const parsed = chatEventSchema.safeParse(safeJson(line.slice(5).trim()))
+      if (!parsed.success) continue
+      const e = parsed.data
+      if (e.warning) handlers.onWarning?.(e.warning)
+      if (e.error) handlers.onError(e.error)
+      else if (e.step) handlers.onStep(e.step)
+      else if (e.done) handlers.onDone(e.done)
+    }
+  }
+}
+
+function safeJson(raw: string): unknown {
+  try {
+    return JSON.parse(raw)
+  } catch {
+    return null
+  }
+}
+
+/**
+ * Refreshes for the two moments that need different amounts of work.
+ *
+ * Mid-turn, only the canvas can have changed: the change set isn't written until
+ * the turn commits, and the exchange isn't stored until it finishes. Refetching
+ * those on every step would be up to 2×(N−1) requests per turn for data that
+ * cannot have moved.
+ */
+export function useAgentRefresh(gardenId: number) {
+  const qc = useQueryClient()
+  const canvas = () => {
+    void qc.invalidateQueries({ queryKey: gardenFullKey(gardenId) })
+  }
+  return {
+    /** After a step: the garden may have changed under the conversation. */
+    canvas,
+    /** After a turn: the change set and the stored exchange exist now too. */
+    everything: () => {
+      canvas()
+      void qc.invalidateQueries({ queryKey: historyKey(gardenId) })
+      void qc.invalidateQueries({ queryKey: agentHistoryKey(gardenId) })
+    },
+  }
+}

web/src/lib/history.test.ts

@@ -49,6 +49,14 @@ describe('describeConflict', () => {
 describe('describeUndo', () => {
   const target = changeSet({ counts: [{ entityType: 'object', op: 'update', n: 3 }] })
 
+  // The bug this pair exists for: a real revert response carries a change set
+  // whose counts the server didn't populate. Reading that as "nothing happened"
+  // told the user a successful undo had done nothing.
+  it('trusts the change set, not the tally, for whether anything happened', () => {
+    const out = describeUndo({ id: 5 }, { changeSet: changeSet({ id: 6, counts: [] }), conflicts: [] })
+    expect(out).toEqual({ tone: 'ok', message: 'Undone.' })
+  })
+
   it('is a plain success when nothing conflicted', () => {
     const out = describeUndo(target, {
       changeSet: changeSet({ id: 2, counts: [{ entityType: 'object', op: 'update', n: 3 }] }),
@@ -61,6 +69,7 @@ describe('describeUndo', () => {
   // already a no-op — reachable by undoing a creation whose object is gone.
   // Claiming "Undone." there reports work that didn't happen.
   it("doesn't claim to have undone a no-op", () => {
+    // A NULL change set — not an empty tally — is the server's no-op signal.
     const out = describeUndo(target, { changeSet: null, conflicts: [] })
     expect(out.tone).toBe('ok')
     expect(out.message).toBe('Nothing left to undo — this was already reversed.')
@@ -113,3 +122,29 @@ describe('totalChanges', () => {
     ).toBe(13)
   })
 })
+
+describe('describeUndo with an unknown total', () => {
+  // The chat panel offers Undo on a turn knowing only its change set id — the
+  // agent's reply carries the id, not the tally. Inventing a denominator to
+  // fill the sentence would be worse than not having one.
+  it("doesn't invent a denominator it was never given", () => {
+    const out = describeUndo(
+      { id: 5 },
+      {
+        changeSet: changeSet({ id: 6, counts: [{ entityType: 'planting', op: 'delete', n: 2 }] }),
+        conflicts: [{ entityType: 'object', entityId: 7, reason: 'changed', name: 'North Bed' }],
+      },
+    )
+    expect(out.tone).toBe('partial')
+    expect(out.message).toBe('Partly undone — “North Bed” was edited since, so it was left alone.')
+    expect(out.message).not.toMatch(/\d+ of \d+/)
+  })
+
+  it('still reports a clean undo the same way', () => {
+    const out = describeUndo(
+      { id: 5 },
+      { changeSet: changeSet({ id: 6, counts: [{ entityType: 'object', op: 'update', n: 1 }] }), conflicts: [] },
+    )
+    expect(out).toEqual({ tone: 'ok', message: 'Undone.' })
+  })
+})

web/src/lib/history.ts

@@ -113,9 +113,10 @@ export function useRevertChangeSet(gardenId: number) {
   })
 }
 
-/** How many rows a change set touched, for "3 changes" in the list. */
+/** How many rows a change set touched, for "3 changes" in the list. Undefined
-export function totalChanges(cs: ChangeSet): number {
+ *  counts total zero, which callers read as "no denominator to quote". */
-  return cs.counts.reduce((sum, c) => sum + c.n, 0)
+export function totalChanges(cs: { counts?: ChangeCount[] }): number {
+  return (cs.counts ?? []).reduce((sum, c) => sum + c.n, 0)
 }
 
 const ENTITY_NOUNS: Record<ChangeCount['entityType'], [string, string]> = {
@@ -171,7 +172,7 @@ export function useUndo(gardenId: number) {
   const revert = useRevertChangeSet(gardenId)
   const [outcomes, setOutcomes] = useState<Record<number, UndoOutcome>>({})
 
-  const undo = (cs: ChangeSet) => {
+  const undo = (cs: UndoTarget) => {
     setOutcomes((prev) => ({ ...prev, [cs.id]: { tone: 'pending', message: 'Undoing…' } }))
     revert.mutate(cs.id, {
       onSuccess: (result) => {
@@ -197,6 +198,19 @@ export function useUndo(gardenId: number) {
   }
 }
 
+/**
+ * What undo needs to know about a change set.
+ *
+ * `counts` is optional because the chat panel offers Undo on a turn knowing only
+ * its change set id — the agent's reply carries the id, not the tally. Fabricating
+ * counts to satisfy a type would produce a confidently wrong "1 of 1 changes
+ * undone"; leaving them out lets describeUndo say what it actually knows.
+ */
+export interface UndoTarget {
+  id: number
+  counts?: ChangeCount[]
+}
+
 export interface UndoOutcome {
   tone: 'pending' | 'ok' | 'partial' | 'error'
   message: string
@@ -208,19 +222,25 @@ export interface UndoOutcome {
  * useful thing to report: a bare failure would be a lie about the two that did
  * apply, and a bare success would hide the one that didn't.
  */
-export function describeUndo(target: ChangeSet, result: RevertResult): UndoOutcome {
+export function describeUndo(target: UndoTarget, result: RevertResult): UndoOutcome {
   const skipped = result.conflicts.map(describeConflict).join('; ')
+  // A NULL change set is the server's signal that nothing needed doing. An empty
+  // `counts` is not the same thing and must not be read as one — that conflation
+  // made a successful undo report "nothing left to undo", which is the worst
+  // possible thing to tell someone about an action that just worked.
+  const didSomething = result.changeSet != null
   const applied = result.changeSet ? totalChanges(result.changeSet) : 0
   if (result.conflicts.length === 0) {
-    // The server answers 200 with a null change set when every revision was
+    // Reachable by undoing a creation whose object is already gone.
-    // already a no-op — reachable by undoing a creation whose object is gone.
+    if (!didSomething) return { tone: 'ok', message: 'Nothing left to undo — this was already reversed.' }
-    // Claiming "Undone." there would be reporting work that didn't happen.
-    if (applied === 0) return { tone: 'ok', message: 'Nothing left to undo — this was already reversed.' }
     return { tone: 'ok', message: 'Undone.' }
   }
-  if (applied === 0) {
+  if (!didSomething) {
     return { tone: 'error', message: `Nothing was undone — ${skipped}.` }
   }
+  // Only claim a denominator when we have one. "2 of 3" from a caller that
+  // never knew the total would be a number invented to fill a sentence.
   const total = totalChanges(target)
-  return { tone: 'partial', message: `${applied} of ${total} changes undone — ${skipped}.` }
+  const scale = total > 0 && applied > 0 ? `${applied} of ${total} changes undone` : 'Partly undone'
+  return { tone: 'partial', message: `${scale} — ${skipped}.` }
 }

web/src/lib/objects.ts

@@ -328,28 +328,28 @@ export function useUpdatePlanting(gardenId: number) {
   })
 }
 
-/** Clear a bed: soft-remove every active plop in an object (a loop of PATCHes;
+const clearResultSchema = z.object({ cleared: z.number() })
- *  a bulk ClearObject endpoint arrives with the agent seam, #19). Invalidates
+
- *  once at the end. Pass the object's active plops (id + current version). */
+/** Clear a bed: soft-remove every active plop in an object (#82).
+ *
+ *  ONE request, and so ONE change set. This used to be a loop of PATCHes, which
+ *  meant clearing a 40-plop bed wrote 40 change sets and took 40 presses of Undo
+ *  to put back — while the agent's clear_object, for the identical user-facing
+ *  action, undid in a single click. The rule it violated is stated in CLAUDE.md:
+ *  multi-row operations record all their changes together so they undo as one
+ *  unit. Doing it server-side also removes the partial-failure case the old loop
+ *  had to reconcile. */
 export function useClearObject(gardenId: number) {
   const qc = useQueryClient()
   return useMutation({
-    mutationFn: async (plops: { id: number; version: number }[]) => {
+    mutationFn: async (objectId: number): Promise<number> => {
-      const today = new Date().toISOString().slice(0, 10)
+      // No body — clear takes none; passing undefined sends none rather than an
-      // allSettled, not all: a partial failure still soft-removed some rows
+      // empty {}. The response is just a count; validate it rather than cast.
-      // server-side, so we must reconcile the cache rather than roll everything
+      const res = clearResultSchema.parse(await api.post(`/objects/${objectId}/clear`))
-      // back. Report how many failed.
+      return res.cleared
-      const results = await Promise.allSettled(
-        plops.map((p) => api.patch(`/plantings/${p.id}`, { removedAt: today, version: p.version })),
-      )
-      const failed = results.filter((r) => r.status === 'rejected').length
-      if (failed > 0) {
-        throw new Error(`${failed} of ${plops.length} plants couldn't be cleared — refresh and try again.`)
-      }
     },
-    // Reconcile on success OR partial failure, so the cache matches the server.
     onSettled: () => qc.invalidateQueries({ queryKey: fullKey(gardenId) }),
-    onError: (err) => toast.error(err instanceof Error ? err.message : 'Could not clear the bed.'),
+    onError: (err) => toast.error(objectErrorMessage(err, 'Could not clear the bed.')),
   })
 }
 

web/src/lib/settings.ts

@@ -0,0 +1,79 @@
+// Instance settings data layer (#79): admin-only, instance-wide configuration.
+//
+// The GET/PATCH return both the stored settings and a read-only "effective" view
+// — what's actually in force after layering the DB over the environment — so the
+// form can say "inheriting ollama-cloud/glm-5.2:cloud from the environment" and
+// whether the API key is present, without the key ever crossing the wire.
+
+import { queryOptions, useMutation, useQuery, useQueryClient } from '@tanstack/react-query'
+import { z } from 'zod'
+import { ApiError, api } from './api'
+import { capabilitiesKey } from './agent'
+
+export const instanceSettingsSchema = z.object({
+  // '' means "inherit the PANSY_AGENT_MODEL env var".
+  agentModel: z.string(),
+  // null means "inherit PANSY_AGENT_ENABLED"; true/false is an explicit override.
+  agentEnabled: z.boolean().nullable(),
+  version: z.number(),
+  updatedAt: z.string(),
+})
+export type InstanceSettings = z.infer<typeof instanceSettingsSchema>
+
+export const effectiveAgentSchema = z.object({
+  model: z.string(),
+  enabled: z.boolean(),
+  hasApiKey: z.boolean(),
+  agentLive: z.boolean(),
+})
+export type EffectiveAgent = z.infer<typeof effectiveAgentSchema>
+
+export const settingsResponseSchema = z.object({
+  settings: instanceSettingsSchema,
+  effective: effectiveAgentSchema,
+})
+export type SettingsResponse = z.infer<typeof settingsResponseSchema>
+
+export const settingsKey = ['settings'] as const
+
+export const settingsQueryOptions = queryOptions({
+  queryKey: settingsKey,
+  queryFn: async (): Promise<SettingsResponse> =>
+    settingsResponseSchema.parse(await api.get('/settings')),
+})
+
+export function useSettings() {
+  return useQuery(settingsQueryOptions)
+}
+
+export interface SettingsUpdate {
+  agentModel: string
+  agentEnabled: boolean | null
+  version: number
+}
+
+export function useUpdateSettings() {
+  const qc = useQueryClient()
+  return useMutation({
+    mutationFn: async (input: SettingsUpdate): Promise<SettingsResponse> =>
+      settingsResponseSchema.parse(await api.patch('/settings', input)),
+    onSuccess: (res) => {
+      qc.setQueryData(settingsKey, res)
+      // The save may have turned the assistant on or off; the editor keys its
+      // chat tab off /capabilities, so make it re-read rather than trust its
+      // cached answer.
+      qc.invalidateQueries({ queryKey: capabilitiesKey })
+    },
+  })
+}
+
+/** If err is a 409 version conflict, return the fresh settings it carries so a
+ *  form can rebase; otherwise null. */
+export function conflictSettings(err: unknown): InstanceSettings | null {
+  if (err instanceof ApiError && err.isConflict && err.body && typeof err.body === 'object') {
+    const current = (err.body as { current?: unknown }).current
+    const parsed = instanceSettingsSchema.safeParse(current)
+    if (parsed.success) return parsed.data
+  }
+  return null
+}

web/src/pages/GardenEditorPage.tsx

@@ -5,6 +5,7 @@ import { Button } from '@/components/ui/Button'
 import { GardenCanvas } from '@/editor/GardenCanvas'
 import { EditorRail, type RailTab } from '@/editor/EditorRail'
 import { HistoryPanel } from '@/editor/HistoryPanel'
+import { ChatPanel } from '@/editor/ChatPanel'
 import { JournalPanel } from '@/editor/JournalPanel'
 import { Inspector } from '@/editor/Inspector'
 import { PlopInspector } from '@/editor/PlopInspector'
@@ -30,6 +31,7 @@ import {
 } from '@/lib/objects'
 import { toEditorPlanting } from '@/lib/plantings'
 import type { Plant } from '@/lib/plants'
+import { useCapabilities } from '@/lib/agent'
 import { useJournalCounts } from '@/lib/journal'
 import { attributableLots, lotsByPlant, useSeedLots, type SeedLot } from '@/lib/seedLots'
 import { useSeedTray } from '@/lib/seedTray'
@@ -64,6 +66,7 @@ export function GardenEditorPage() {
   const journalObjectId = useEditorStore((s) => s.journalObjectId)
   const setJournalObjectId = useEditorStore((s) => s.setJournalObjectId)
   const journalCounts = useJournalCounts(gid)
+  const capabilities = useCapabilities()
   const journalTotal = useMemo(
     () => [...(journalCounts.data?.values() ?? [])].reduce((a, b) => a + b, 0),
     [journalCounts.data],
@@ -385,8 +388,21 @@ export function GardenEditorPage() {
     },
   ]
 
+  // Only when the instance actually has the assistant configured. A tab that
+  // opens onto an apology is worse than no tab.
+  if (capabilities.data?.agent) {
+    railTabs.push({
+      id: 'chat',
+      label: 'Assistant',
+      render: () => <ChatPanel gardenId={gid} canEdit={canEdit} />,
+    })
+  }
+
+  // 100dvh, not 100vh: on mobile Safari/Chrome 100vh is the *largest* viewport
+  // (URL bar hidden), so with the bar showing the editor overflowed and pushed
+  // the canvas bottom + Fit button under the browser chrome (#85).
   return (
-    <div className="flex h-[calc(100vh-8rem)] flex-col gap-3 md:flex-row">
+    <div className="flex h-[calc(100dvh-8rem)] flex-col gap-3 md:flex-row">
       <div className="shrink-0 md:w-40">
         <h1 className="mb-2 truncate text-lg font-semibold tracking-tight" title={garden.name}>
           {garden.name}
@@ -422,6 +438,15 @@ export function GardenEditorPage() {
         >
           History
         </Button>
+        {capabilities.data?.agent && (
+          <Button
+            variant="ghost"
+            className="mt-1 w-full text-sm"
+            onClick={() => setRailTab(railTab === 'chat' ? null : 'chat')}
+          >
+            💬 Assistant
+          </Button>
+        )}
       </div>
 
       <div className="relative flex min-h-0 flex-1 flex-col gap-2">
@@ -504,8 +529,9 @@ export function GardenEditorPage() {
 
       {clearing && focusedObject && (
         <ClearBedModal
+          objectId={focusedObject.id}
           objectName={objectDisplayName(focusedObject)}
-          plops={focusedPlops.map((p) => ({ id: p.id, version: p.version }))}
+          plopCount={focusedPlops.length}
           gardenId={gid}
           onClose={() => setClearing(false)}
         />

web/src/pages/PublicGardenPage.tsx

@@ -53,7 +53,7 @@ export function PublicGardenPage() {
   }
 
   return (
-    <div className="flex h-[calc(100vh-8rem)] flex-col gap-3">
+    <div className="flex h-[calc(100dvh-8rem)] flex-col gap-3">
       <div className="flex flex-wrap items-center gap-2">
         <h1 className="truncate text-lg font-semibold tracking-tight" title={garden.name}>
           {garden.name}

web/src/pages/SettingsPage.tsx

@@ -0,0 +1,164 @@
+import { useEffect, useState } from 'react'
+import { Alert } from '@/components/ui/Alert'
+import { Button } from '@/components/ui/Button'
+import { Select } from '@/components/ui/Select'
+import { TextField } from '@/components/ui/TextField'
+import { toast } from '@/components/ui/toast'
+import { errorMessage } from '@/lib/api'
+import {
+  conflictSettings,
+  useSettings,
+  useUpdateSettings,
+  type EffectiveAgent,
+  type InstanceSettings,
+} from '@/lib/settings'
+import { usePageTitle } from '@/lib/usePageTitle'
+
+// agentEnabled is a tri-state on the wire (null = inherit env, true, false); the
+// form models it as three named choices so "inherit" is a deliberate pick, not
+// an empty control.
+type EnabledChoice = 'inherit' | 'on' | 'off'
+const toChoice = (v: boolean | null): EnabledChoice => (v === null ? 'inherit' : v ? 'on' : 'off')
+const fromChoice = (c: EnabledChoice): boolean | null => (c === 'inherit' ? null : c === 'on')
+
+/** Admin-only instance settings (#79). The one that matters today is the agent
+ *  model; the API key stays in the environment and is only ever reported as
+ *  present/absent, never shown or edited. */
+export function SettingsPage() {
+  usePageTitle('Settings')
+  const settings = useSettings()
+  const update = useUpdateSettings()
+
+  if (settings.isPending) {
+    return <p className="text-sm text-muted">Loading settings…</p>
+  }
+  if (settings.isError) {
+    return <Alert>{errorMessage(settings.error, "Couldn't load settings.")}</Alert>
+  }
+  return <SettingsForm data={settings.data} update={update} />
+}
+
+function SettingsForm({
+  data,
+  update,
+}: {
+  data: { settings: InstanceSettings; effective: EffectiveAgent }
+  update: ReturnType<typeof useUpdateSettings>
+}) {
+  const [model, setModel] = useState(data.settings.agentModel)
+  const [enabled, setEnabled] = useState<EnabledChoice>(toChoice(data.settings.agentEnabled))
+  const [version, setVersion] = useState(data.settings.version)
+  const [error, setError] = useState<string | null>(null)
+
+  // Rebase the form when the query cache updates (e.g. a successful save writes
+  // the new row back), so the version we submit is never stale.
+  useEffect(() => {
+    setVersion(data.settings.version)
+  }, [data.settings.version])
+
+  const eff = data.effective
+
+  const save = () => {
+    setError(null)
+    update.mutate(
+      { agentModel: model.trim(), agentEnabled: fromChoice(enabled), version },
+      {
+        onSuccess: () => toast.info('Settings saved.'),
+        onError: (err) => {
+          const current = conflictSettings(err)
+          if (current) {
+            // Someone else saved first. Adopt their row so the next attempt is
+            // clean, and say so rather than silently discarding this edit.
+            setModel(current.agentModel)
+            setEnabled(toChoice(current.agentEnabled))
+            setVersion(current.version)
+            setError('Someone else changed these settings just now — reloaded their version. Re-apply your change if you still want it.')
+            return
+          }
+          setError(errorMessage(err, "Couldn't save settings."))
+        },
+      },
+    )
+  }
+
+  return (
+    <div className="flex max-w-2xl flex-col gap-6">
+      <div>
+        <h1 className="text-xl font-semibold text-fg">Settings</h1>
+        <p className="mt-1 text-sm text-muted">
+          Instance-wide, admin-only. Changes take effect immediately — no restart.
+        </p>
+      </div>
+
+      <section className="flex flex-col gap-4 rounded-lg border border-border p-4">
+        <div>
+          <h2 className="text-sm font-semibold text-fg">Garden assistant</h2>
+          <p className="mt-1 text-xs text-muted">
+            The model runs against your Ollama Cloud key, which is set in the environment and never
+            shown here.
+          </p>
+        </div>
+
+        <AgentStatus eff={eff} />
+
+        <TextField
+          label="Model"
+          name="agentModel"
+          placeholder={eff.model || 'ollama-cloud/glm-5.2:cloud'}
+          value={model}
+          onChange={(e) => setModel(e.target.value)}
+          hint={
+            model.trim() === ''
+              ? `Empty — inheriting ${eff.model || 'the built-in default'} from the environment.`
+              : 'A majordomo model spec. A comma-separated list is a failover chain.'
+          }
+        />
+
+        <Select
+          label="Enabled"
+          name="agentEnabled"
+          value={enabled}
+          onChange={(e) => setEnabled(e.target.value as EnabledChoice)}
+          options={[
+            { value: 'inherit', label: 'Inherit from environment' },
+            { value: 'on', label: 'On' },
+            { value: 'off', label: 'Off' },
+          ]}
+        />
+      </section>
+
+      {error && <Alert>{error}</Alert>}
+
+      <div className="flex justify-end">
+        <Button onClick={save} disabled={update.isPending}>
+          {update.isPending ? 'Saving…' : 'Save changes'}
+        </Button>
+      </div>
+    </div>
+  )
+}
+
+// AgentStatus surfaces the gap the capabilities endpoint exists for: enabled +
+// a key does not guarantee the assistant is actually running (an unresolvable
+// model leaves it down), and that's precisely what an admin needs to see.
+function AgentStatus({ eff }: { eff: EffectiveAgent }) {
+  const [tone, text] = eff.agentLive
+    ? (['ok', `Live on ${eff.model}`] as const)
+    : !eff.hasApiKey
+      ? (['warn', 'No API key set in the environment — the assistant is off.'] as const)
+      : !eff.enabled
+        ? (['warn', 'Turned off.'] as const)
+        : (['warn', `Configured but not running — check the model (${eff.model}).`] as const)
+
+  return (
+    <div className="flex items-center gap-2 text-sm">
+      <span
+        className={
+          'inline-block h-2 w-2 rounded-full ' + (tone === 'ok' ? 'bg-emerald-500' : 'bg-amber-500')
+        }
+        aria-hidden
+      />
+      <span className={tone === 'ok' ? 'text-fg' : 'text-muted'}>{text}</span>
+    </div>
+  )
+}

web/src/router.tsx

@@ -14,6 +14,7 @@ import { GardensPage } from '@/pages/GardensPage'
 import { GardenEditorPage } from '@/pages/GardenEditorPage'
 import { PublicGardenPage } from '@/pages/PublicGardenPage'
 import { PlantsPage } from '@/pages/PlantsPage'
+import { SettingsPage } from '@/pages/SettingsPage'
 import { meQueryOptions } from '@/lib/auth'
 import { queryClient } from '@/lib/queryClient'
 import { safeRedirectPath } from '@/lib/redirect'
@@ -48,6 +49,20 @@ async function requireGuest(context: RouterContext, redirectTo: string) {
   }
 }
 
+// requireAdmin: authenticated AND admin. A non-admin who navigates to /settings
+// is sent to /gardens rather than shown a page whose API calls would 403. This
+// is convenience routing, not the security boundary — the server's requireAdmin
+// is authoritative.
+async function requireAdmin(context: RouterContext, path: string) {
+  const me = await context.queryClient.ensureQueryData(meQueryOptions)
+  if (!me) {
+    throw redirect({ to: '/login', search: { redirect: path } })
+  }
+  if (!me.isAdmin) {
+    throw redirect({ to: '/gardens' })
+  }
+}
+
 const indexRoute = createRoute({
   getParentRoute: () => rootRoute,
   path: '/',
@@ -109,6 +124,13 @@ const plantsRoute = createRoute({
   component: PlantsPage,
 })
 
+const settingsRoute = createRoute({
+  getParentRoute: () => rootRoute,
+  path: 'settings',
+  beforeLoad: ({ context, location }) => requireAdmin(context, location.href),
+  component: SettingsPage,
+})
+
 // Public read-only garden by share token. Deliberately has NO beforeLoad auth
 // guard, so a logged-out visitor viewing a shared link is never redirected to
 // /login or OIDC.
@@ -125,6 +147,7 @@ const routeTree = rootRoute.addChildren([
   gardensRoute,
   gardenEditorRoute,
   plantsRoute,
+  settingsRoute,
   publicGardenRoute,
 ])