Rohit Arora 9e8a9fd6e9 feat: add pgvector semantic memory, PgBouncer pooling, and Alembic migrations ... pgvector — semantic memory retrieval: - Postgres Dockerfile builds pgvector v0.7.4 from source (Alpine) - memories table gains embedding vector(384) column + IVFFlat cosine index - EmbeddingService: lazy-loads all-MiniLM-L6-v2 (384 dims) via sentence-transformers; embed_text() / embed_batch() with graceful degradation if model unavailable - DBManager: asyncpg vector codec registration (text-cast fallback), upsert_memory_embedding(), get_semantically_relevant_memories() retrieves by cosine similarity rather than importance score PgBouncer — connection pooling: - Transaction-mode pooling: 200 max clients → 20 Postgres connections - Sits between all app services (FastAPI + 4 Celery workers) and Postgres - DATABASE_URL in .env now routes through westworld-pgbouncer:5432 - Added to docker-compose.yml and CI/CD pipeline (build-deploy.yaml) Alembic — versioned schema migrations: - alembic.ini configured for async SQLAlchemy + asyncpg - env.py uses async_engine_from_config; reads DATABASE_URL from env - Initial migration 0001_initial_schema.py: creates vector extension, all 13 tables (including embedding column), all 14 indexes - Downgrade drops tables in correct FK dependency order Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>		2026-03-23 23:20:03 -04:00
..
build-deploy.yaml	feat: add pgvector semantic memory, PgBouncer pooling, and Alembic migrations	2026-03-23 23:20:03 -04:00