FreeVisit Website
TL;DR - pgvector
- pgvector is an open-source PostgreSQL extension for vector similarity search, allowing users to store and query vectors directly within their database.
- It supports various distance metrics (L2, inner product, cosine, L1, Hamming, Jaccard) and vector types (single-precision, half-precision, binary, sparse).
- pgvector offers both exact and approximate nearest neighbor search, with approximate search enabled by HNSW and IVFFlat indexes for improved performance.
Pricing: Free forever
Best for: Individuals & startups
Pros & Cons
Pros
- Native PostgreSQL
- No separate service
- Open source
Cons
- PostgreSQL only
- Limited compared to specialized DBs
Key Features
Vector storageSimilarity searchPostgreSQL nativeMultiple indexesExact and approximate searchOpen source
Pricing Plans
Open Source
Free
- PostgreSQL extension
- Vector similarity search
- No per-vector pricing
- 60-80% cheaper than Pinecone
About pgvector
By Toolradar Team·
pgvector is an open-source PostgreSQL extension for vector similarity search. Enables storing embeddings and performing similarity queries directly in your PostgreSQL database.
Reviews
No reviews yet. Be the first to review pgvector!
pgvector FAQ
What types of vector similarity search does pgvector support?
pgvector supports both exact and approximate nearest neighbor search. It can work with single-precision, half-precision, binary, and sparse vectors.
Which distance metrics are available in pgvector?
pgvector supports several distance metrics including L2 distance, inner product, cosine distance, L1 distance, Hamming distance, and Jaccard distance. The Hamming and Jaccard distances are specifically for binary vectors.
How can I install pgvector on a Linux or Mac system?
To install pgvector on Linux or Mac, clone the repository, navigate to the directory, and then run `make` followed by `make install`. This process supports Postgres 13 and newer versions.
What is the purpose of adding an index in pgvector?
Adding an index in pgvector enables approximate nearest neighbor search, which improves query speed by trading off some recall. This means queries with an approximate index may yield slightly different results compared to an exact search.
Which index types does pgvector support for approximate nearest neighbor search?
pgvector supports HNSW and IVFFlat index types for approximate nearest neighbor search. An HNSW index, for example, creates a multilayer graph to optimize query performance.
How do I calculate cosine similarity using pgvector?
To calculate cosine similarity, you can use the formula `1 - (embedding <=> '[3,1,2]')`. The `<=>` operator returns the cosine distance, so subtracting it from 1 yields the similarity.
Source: github.com