Vector

Supabase Vector is an open-source toolkit for building AI applications that combines PostgreSQL with the pgvector extension, enabling developers to store, index, and search vector embeddings alongside relational data in a single database. Founded in 2019 and headquartered in San Francisco, CA, it targets teams who want to avoid managing a separate vector database by leveraging Postgres's existing ecosystem. The Free plan includes 500 MB database storage, 50,000 monthly active users, and unlimited API requests, making it suitable for prototyping and small projects. The Pro plan starts at $25/month per project, offering 8 GB database size, 250 GB egress, and daily backups stored for 7 days, appealing to production applications that need predictable scaling without leaving the Postgres environment.

Supabase Vector uses pgvector to perform efficient similarity search via IVFFlat and HNSW indexes, supporting L2 distance, inner product, and cosine distance. It integrates directly with Supabase's managed Postgres service, allowing developers to run vector queries alongside standard SQL operations. Real-time updates are supported through Supabase's Realtime engine, enabling live synchronization of vector data changes to clients. The platform provides advanced querying capabilities, including hybrid search combining vector similarity with SQL filters, and supports indexing strategies that trade recall for speed. For example, an IVFFlat index with 100 lists can reduce search time by 90% compared to brute-force scanning on a dataset of 1 million 768-dimensional vectors. The Pro plan includes 250 GB cached egress and 100 GB file storage, with compute add-ons scaling from a Micro instance (2-core ARM, 1 GB RAM, $10/month) to a 16XL instance (64-core ARM, 256 GB RAM, $3,730/month).

Supabase Vector competes directly with specialized vector databases like Pinecone, Weaviate, Milvus, Qdrant, and Chroma, as well as vector search features in MongoDB Atlas and Redis. A 2023 Supabase benchmark showed pgvector on a Postgres instance with 8 GB RAM achieving 95% recall at 1ms latency for 1 million vectors, comparable to Pinecone's performance at a fraction of the cost. The platform's unified data storage eliminates the need for data synchronization between a relational database and a vector store, reducing operational complexity. However, dedicated vector databases like Pinecone offer managed scaling for massive workloads (e.g., 10 billion vectors) and built-in filtering optimizations that pgvector may not match. Supabase's open-source model and Postgres compatibility give it an edge for teams already invested in the Postgres ecosystem, while alternatives like Weaviate provide native multi-tenancy and hybrid search out of the box.

Honest trade-offs: Supabase Vector's Free plan pauses projects after 1 week of inactivity, limiting its use for hobby apps that don't receive regular traffic. The Pro plan's 8 GB database size per project can be restrictive for large-scale vector workloads; a single dataset of 10 million 768-dimensional vectors consumes roughly 30 GB, requiring compute add-ons that increase monthly costs. Community support on the Free plan is limited to community forums, with no guaranteed response times. Beginners may find the combination of Postgres, pgvector, and Supabase's tooling complex to set up, especially when tuning index parameters like `lists` and `probes` for optimal recall-latency trade-offs. A Reddit user noted that using an existing Postgres database for vector search is often better than adopting a separate vector database, but this approach requires expertise in Postgres performance tuning and index management.

Free ($0/month): 500 MB database, 50,000 MAUs, 5 GB egress, 2 active projects. Pro ($25/month): 8 GB database, 100,000 MAUs, 250 GB egress, daily backups. Team ($599/month): SOC2, SSO, 14-day backups. Enterprise: custom pricing with 24×7 support.