🧩 Reliability & honesty fixes
- Streaming answers now carry an honest truncation notice instead of silently cutting off mid-word (#336)
- Silent embedding/memory-search failures now surface as real warnings instead of disappearing at debug level (#341)
⚙️ Scaling
- Optional Redis/RQ task queue for autonomy actions - degrades to inline execution if unset (#367)
- Rejection-pattern reporting for autonomy decisions, honestly scoped - auto-suggestion deliberately deferred until there's enough real data (#368)
🐳 Quickstart
docker compose up -dnow starts only db+app by default; voice/neo4j/redis/worker moved behind opt-in profiles (voice/graph/queue/full) (#381)
🤝 Community
- First external contribution merged: indirect graph-boosted retrieval via
search_related_entities(), courtesy of @hardik47278 (#373, closes #25)
📄 Docs
- Documented that custom AI Employee role creation already works via
PATCH /employees/{role}(#371) - Fixed a stale ROADMAP.md claim about SENSITIVE_DOMAIN_ROLES enforcement (#370)
- Documented real Ollama-in-Docker setup steps (#356)
- README: removed hosted-platform overclaims, replaced the comparison table with prose, noted the new ragleap-observability package (#334, #342, #343)
✅ Testing
Full suite: 253 passed, 0 failures.