I need a vector store with metadata filtering and permission-aware retrieval

For metadata filtering + permission-aware retrieval, you want a vector store that can:

1. store embeddings,
2. attach structured metadata to each chunk/document,
3. filter during search,
4. enforce access control before results are returned.

Good options

1. PostgreSQL + pgvector

Best if you want a flexible, production-friendly setup.

Why it fits:

• Store vectors and metadata in the same DB
• Strong SQL filtering on metadata
• Easy to implement permission checks with joins / row-level security
• Good for “permission-aware retrieval” because access rules can be enforced in the database

Typical pattern:

• documents table with embedding, metadata, tenant_id, acl, etc.
• Query with vector similarity + WHERE clauses for allowed users/roles/projects

Pros

• Very strong filtering
• Easy ACL integration
• Familiar tooling
• Great for multi-tenant apps

Cons

• May need tuning for very large scale

---

2. Pinecone

Good managed vector database with metadata filtering.

Why it fits:

• Native metadata filters
• Scales well
• Simple API

Permission-aware retrieval

• Usually handled by encoding permissions into metadata and filtering by user/tenant/group
• Works well, but true complex ACL logic may need an app-side authorization layer

Pros

• Easy to use
• Fast
• Managed

Cons

• Less flexible than SQL for complex access rules
• Costs can rise with scale

---

3. Weaviate

Strong choice if you want richer filtering and schema support.

Why it fits:

• Metadata filtering is built in
• Supports structured schema
• Can model tenant/user fields for access filtering

Pros

• Powerful filtering
• Good developer experience
• Open source + managed options

Cons

• More operational complexity than Postgres
• Permission logic still usually app-defined

---

4. Qdrant

Very good for payload-based filtering and access conditions.

Why it fits:

• Fast vector search
• Rich payload filtering
• Good for metadata-driven retrieval

Permission-aware retrieval

• Store ACL-related fields in payload
• Filter by user groups/tenant/document visibility

Pros

• Fast and lightweight
• Strong filtering
• Good self-hosted option

Cons

• ACL logic is still mostly implemented via filtering design

---

Best approach for permission-aware retrieval

If permissions matter a lot, I’d recommend:

Postgres + pgvector

because you can:

• join against a permissions table,
• enforce access in SQL,
• use row-level security,
• avoid accidentally returning restricted items.

Example idea:

• docs(id, embedding, metadata, tenant_id, visibility)
• doc_acl(doc_id, principal_type, principal_id, permission)
• Query only docs where the current user has access

---

If you want the simplest managed solution

Choose:

• Pinecone if you want managed and scalable
• Qdrant if you want strong filtering and self-hosting

---

Recommendation by use case

• Strict enterprise permissions / multi-tenant app → Postgres + pgvector
• Managed SaaS with moderate ACL needs → Pinecone
• Open source + rich filtering → Qdrant or Weaviate
• Complex authorization logic → Postgres + pgvector

If you want, I can also give you:

1. a comparison table,
2. a recommended schema for ACL-based retrieval, or
3. code examples for pgvector / Pinecone / Qdrant.