Open-source graph database on object storage, with typed schemas and Git-style branches.
What it does
On this page
Omnigraph keeps a typed property graph as Lance tables inside storage you already own, then serves it to agents and applications through a CLI, an HTTP server or MCP.
Who it is for
Teams running several agents that need one shared model of a domain, and who would rather hold that state in their own bucket than in a hosted memory service.
How it works
Schemas in .pg files declare node and edge types, keys and constraints, and the engine checks them on every write. Reads combine graph traversal, vector search and full-text ranking inside a single query. Writes land on branches that merge like code, so an agent can propose a change for review instead of editing the main graph. Access control is written as Cedar policy and enforced in the engine, with an actor recorded for the audit trail. Graphs live on local disk, S3-compatible storage or Azure Blob, and the format stays open enough for other tools to read the files.
Pricing
The engine, CLI and server are MIT-licensed and free to run; you pay for your own storage and compute. A managed cloud is in private preview.
Notable limits
Azure Blob support is listed as a preview.
Source: https://www.omnigraph.dev/