pgvector vs sqlite-vec
A side-by-side comparison of pgvector and sqlite-vec, two Vector DB tools, drawn from Ignaite's continuously-verified listings.
Compared from listings verified as of
At a glance
| Attribute | pgvector | sqlite-vec |
|---|---|---|
| Category | Vector DB | Vector DB |
| Pricing | FREE | FREE |
| License | Open source | Open source |
| Deployment (differs) | Self-host | Local |
| Platforms | API | API |
| Model support | Model-agnostic | Model-agnostic |
| Vendor (differs) | pgvector community | Alex Garcia |
| Capabilities |
|
|
The honest brief
pgvector
Keeps vectors in your existing Postgres, so you JOIN against relational data and back it all up together.
- No new database to operate
- JOIN embeddings with relational data
- Free and open source
- Works on Supabase, Neon, any managed Postgres
- Scales worse than dedicated vector DBs
- Tuning HNSW/IVFFlat is on you
- No built-in hybrid search out of the box
sqlite-vec
Embeds vector search inside the SQLite file itself, so RAG can run fully local — in the browser via WASM or on a Raspberry Pi — with no server.
- Zero dependencies, pure C
- Runs anywhere SQLite runs
- Bindings for Python, JS, Ruby, Go, Rust
- Local-first, no server needed
- MIT / Apache 2.0 licensed
- Exhaustive (brute-force) search, not ANN
- Not built for very large datasets
- Single-node, embedded only
When to pick which
pgvector and sqlite-vec cover the same capabilities. They differ on deployment:
- Deployment
- Self-host · Local