basalt
Star the repo Sign in Start a project
Beta

Basalt Vector

The open source vector toolkit for Postgres

Use the Basalt client libraries to store, index, and query your vector embeddings at scale. Build AI applications on the Postgres database you already run — no second system to operate.

Quickstart Explore documentation

Postgres + vectors

Use the vector column type to store, query, and index your embeddings at scale inside a Postgres instance.

Nimbus AI and more

Connect any large-language-model or embeddings API, including Emberface, Aurora Forge and Modelhub.

Secure and scalable

Basalt is SOC 2 Type II compliant, and ships with row-level security and a fine-grained permissions system.

Deploy globally

Choose from 14 distributed regions, or self-host the whole stack on infrastructure you already own.

Leverage the tools you already use

Nimbus AI Emberface Aurora Forge Linkweave
Skiff Notebooks Loomlab
Basalt Vector

Simple yetpowerful APIs

Small, predictable client libraries for managing and querying vector stores in Postgres.

Explore documentation
Open in Skiff 
docs = vx.create_collection(name="docs", dimension=3)

# add records to the collection
docs.upsert(
    vectors=[
        (
            "vec0",
            [0.1, 0.2, 0.3],   # the vector's identifier
            {"year": 1973}      # associated metadata
        ),
        (
            "vec1",
            [0.7, 0.8, 0.9],
            {"year": 2012}
        )
    ]
)
Store

Efficiently upsert millions of vectors alongside the metadata that matters.

Index
Query

What you can build
with Basalt Vector

Scale from a weekend experiment to a production AI workload without changing databases.

Semantic search

Search your own knowledge base by meaning instead of keywords.

View example

Chat assistants

Give an assistant durable, content-grounded long-term memory.

Template coming soon

Hybrid search

Combine semantic and full-text ranking with plain SQL filters.

Template coming soon

Image similarity

Turn images into embeddings to detect near-duplicates and lookalikes.

Open in Skiff

Data management

Automatically tag, deduplicate and cluster records in your store.

Open in Skiff

Recommendations

Surface related articles, videos, products and places automatically.

Template coming soon

Powerful featuresScale to millions

Develop, integrate and deploy secure, enterprise-grade AI applications at unusual speed.

Explore documentation

Fully managed or self-hosted

Start on our managed cloud platform, or run the identical stack inside your own infrastructure. You choose.

Global & multi-region

Provision and configure a fleet of read replicas across regions to cut query latency for distant users.

Integrated

Store embeddings in the same database as your transactional data, which simplifies the app and cuts a network hop.

No vendor lock-in

Basalt is built on open source tools, so migrating in — or out — is a schema dump away.

Automatic backups

Point-in-time recovery keeps every write for 28 days, so a bad migration is a rollback, not an incident.

Highly scalable

Designed for high recall and low tail latency at global scale, from 10k vectors to 240 million.

Teams building on
Basalt Vector

We picked Basalt over the specialised vector databases because it let us stay GDPR compliant from day one with very little extra work.

Mikael Fenn, Co-founder at Draftpilot

Read customer story

We tried three other vector databases — Vectorly, Weavepoint, Pinehold. If you only need nearest-neighbour search they're great, but the moment you have to store real metadata alongside it, things get painful.

Cal Prentice, CEO at Answerly

Read customer story

Basalt Vector for enterprise

Talk to one of our engineers about scaling Basalt Vector and managing embeddings at volume.

Fill out the enterprise form

Pick your building blocks

Every Basalt product works on its own and gets better next to the others, so you can adopt exactly as much as you need.