Changeset 0.15.0 patches (#72)

docs/ROADMAP.md

@@ -749,15 +749,30 @@ See [DESIGN-0.14.0.md](DESIGN-0.14.0.md) for full spec.
 
 ---
 
-## v0.15.0 — Compaction
+## v0.15.0 — Compaction ✅
 
 Depends on: utility model role (v0.10.0).
 
-- [ ] Auto-compaction service: background job that calls utility model to summarize
-- [ ] Channel-scoped: triggers when context exceeds threshold
-- [ ] Summaries stored as system messages
-- [ ] Configurable: per-channel opt-in/out, summary model override
-- [ ] Admin controls for resource limits
+- [x] Compaction service: extracts messages, calls utility model to summarize, replaces with summary
+- [x] `Summarize & Continue` button: user-triggered compaction via existing summarize endpoint
+- [x] Channel-scoped context threshold triggers (configurable min messages, min chars, activity gap)
+- [x] Summaries stored as system messages with `is_summary = true` metadata
+- [x] Background scanner: periodic candidate detection with cooldown and in-flight dedup
+- [x] Configurable: per-channel opt-in/out (`compaction_opt_out`), global threshold overrides
+- [x] Admin controls: global enable/disable, threshold/cooldown settings via `global_settings`
+- [x] Context budget guard rail: estimated tokens must exceed 80% of ceiling before compaction
+- [x] Token estimation: `(len + 3) / 4` heuristic with path-from-summary support
+- [x] Rate limiting on summarize endpoint
+- [x] Comprehensive test suite: scanner candidates, guard rails, live pipeline integration tests
+
+Bug fixes during CI stabilization:
+- [x] Fixed `channels.deleted_at` → `channels.is_archived` (channels use soft-archive, not soft-delete)
+- [x] Fixed NULL model column scan in scanner (`COALESCE(c.model, '')`)
+- [x] Fixed test channel backdating (BEFORE UPDATE trigger override via INSERT)
+- [x] Fixed vault nil-check fallback in role resolver for plaintext API keys
+- [x] Fixed parallel test DB races (`-p 1` serialization)
+- [x] Fixed proactive token refresh cascade: failed refresh no longer nukes active session
+- [x] Fixed model visibility reset: preferences fetch failure preserves existing hidden state
 
 ---
 
@@ -1160,6 +1175,7 @@ based on need.
 
 **UX / Multi-Seat**
 - ~~Per-chat model/preset persistence (server-side)~~ ✅ _(v0.12.0 — `channels.settings.last_selector_id` JSONB merge, localStorage write-through cache)_
+- Per-provider model preferences: `user_model_settings` unique key is `(user_id, model_id)` — same model from different providers shares one visibility toggle. Needs `provider_config_id` dimension in DB constraint, store, API, and frontend `hiddenModels` keying. Frontend composite key (`configId:modelId`) already exists in `App.models[].id`.
 
 **Knowledge Bases — Future**
 - Hybrid search: combine vector similarity with full-text `tsvector`, re-rank
@@ -1184,4 +1200,4 @@ based on need.
 - Multi-tenant SaaS mode
 - Plugin/extension marketplace
 - Virtual scroll for long conversations
-- SQLite backend option (single-user / dev)
+- SQLite backend option (single-user / dev)