Add RAG to the database you already run.
Pick the database you have. See whether to extend it or add a vector store, then the embedding, rerank, gateway, and trace tools around it.
Start from your database
Common first stops
These are navigation, not a ranking. They cover the usual fork: extend Postgres, use a dedicated store, then add an embedding API and a reranker.
-
pgvector
Open-source vector similarity search for Postgres. The README documents exact and approximate nearest neighbor search, and says to combine it with Postgres full-text search for hybrid search.
-
Pinecone
Managed vector database. The docs list full-text, sparse-vector, hybrid search, and filter-by-metadata guides. The pricing page lists Starter, Builder, Standard, and Enterprise.
-
OpenAI embeddings
Hosted embedding models. The pricing page lists text-embedding-3-small, text-embedding-3-large, and text-embedding-ada-002.
-
Cohere Rerank
Cohere's rerank models. The Rerank 5 announcement lists API prices for the Pro and Fast tiers.
Catalog
- Databases22
Vector search inside a database you already run, and dedicated vector stores.
- Embeddings7
APIs and open weights that turn text into vectors.
- Rerankers4
Cross-encoders that reorder a short list of retrieved chunks.
- Gateways4
Proxies in front of model APIs: keys, fallbacks, and logs.
- Observability5
Traces for prompts, retrieval, and model calls.
- Frameworks5
Libraries that wire retrieval, tools, and model calls.