Vector Store Implementations

[RyanNg1403]

Add support for 6 additional vector storage backends following the existing architecture pattern at src/core/vector_storage/backend/.

Vector Store Details

1. ChromaDB: HTTP/persistent modes, metadata filtering (npm install chromadb)
2. FAISS: Local high-performance search, multiple index types (npm install faiss-node)
3. Pinecone: Cloud database, namespaces, rate limiting (npm install @pinecone-database/pinecone)
4. Weaviate: GraphQL-based, schema management (npm install weaviate-ts-client)
5. PgVector: PostgreSQL extension, ACID transactions (npm install pg @types/pg)
6. Redis: In-memory, RedisSearch module (npm install redis @redis/search)

Success Criteria

• All implementations follow existing architectural patterns
• Comprehensive test coverage (100%)
• Proper error handling and logging
• Configuration documentation updated

[onestardao]

one thing to watch out for when adding multiple backends like this is that FAISS and some others have silent failure modes if the index is created but never properly loaded or activated before queries.
this can pass tests in a clean environment but break in deployment when ingestion or config ordering changes.
worth having a minimal verification step per backend to confirm data is actually queryable before marking the test as passed.

[RyanNg1403]

@onestardao really helpful insight! I'll be looking at that.

[onestardao]

This falls under what we classify as No.14 – vector store backend mismatch / hidden operational modes in our AI failure Problem Map.
We’ve documented and fixed these FAISS + multi-backend inconsistencies so the retrieval layer behaves identically across modes.
Full details here: WFGY Problem Map — MIT licensed, free to integrate.

[romanbsd]

I implemented the pgvector store: #198