pgvector vs Chroma
Both are free/open-source alternatives to Pinecone. Here's how they stack up — verified facts, no spin.
pgvector
Vector search inside the Postgres you already run — zero new infrastructure.
pgvector is an open-source extension (PostgreSQL license) that adds vector similarity search to PostgreSQL itself. If your app already has Postgres, this is the no-new-moving-parts answer: embeddings live next to your relational data, joins and filters are just SQL, and every major managed Postgres (RDS, Cloud SQL, Azure, Supabase, Neon) supports it. Recent releases added parallel HNSW index builds, iterative scans for filtered queries, and halfvec quantization — it now handles serious workloads, not just prototypes.
Chroma
The fastest way to prototype RAG — embedded, local-first, Apache-2.0.
Chroma is an open-source, AI-native vector store built for developer speed: pip install, three lines of Python, and you have persistent local vector search — no server required. It persists to disk by default in its embedded mode and also runs as a client-server deployment when an app graduates from notebook to production. For prototypes, local agents, and small-to-mid RAG apps it is the lowest-friction option in the ecosystem.
Side by side
| pgvector | Chroma | |
|---|---|---|
| Sovereignty Score | 94 | 90 |
| Open source | Yes | Yes |
| Self-hostable | Yes | Yes |
| Local-first | Yes | Yes |
| License | PostgreSQL | Apache-2.0 |
| Pricing | Free — an extension of PostgreSQL; runs wherever your Postgres runs | Free open source; optional managed Chroma Cloud |
pgvector edges it on the Sovereignty Score, but the right pick depends on the trade-offs below.
pgvector
Strengths
- +No new database to operate — it's your existing Postgres
- +Vectors join directly with relational data in SQL
- +Supported by every major managed Postgres provider
- +HNSW indexes, filtered iterative scans, quantization
Trade-offs
- −Very large collections (hundreds of millions of vectors) favor a dedicated engine
- −Tuning happens in Postgres terms — indexes, memory, vacuum
Chroma
Strengths
- +Embedded mode: vector search with zero infrastructure
- +First-class LangChain / LlamaIndex integration
- +Local persistence by default — data survives restarts
Trade-offs
- −Not built for large distributed production clusters
- −Fewer enterprise features than Milvus/Weaviate
Facts verified 2026-07-15. Licenses and pricing change — spotted something out of date? That's a correction we want.