Omnigraph

Graph Database

Omnigraph is an open source graph database built for workloads where many AI agents share, read and update the same body of knowledge. Instead of running on a dedicated server with its own disks, it stores every graph in the open Lance format on local disk or any S3-compatible object store, which keeps storage cheap and the data fully in your hands. Graphs are versioned like code: agents and people work on branches, review changes and merge them back. Written in Rust and released under the MIT license, it ships as a single CLI and server, with an MCP server for connecting agents directly.

• Typed ontology: .pg schema files declare node and edge types, keys and constraints, and the engine enforces them on every write, so every agent works against the same model of the domain.
• Multimodal retrieval: One .gq query combines graph traversal, vector nearest-neighbour search, BM25 full-text search and filters, merged with Reciprocal Rank Fusion, to assemble precise context in a single round trip.
• Git-style branching: Create a branch per agent or task, diff and review changes, run three-way merges and time-travel across the full history of the graph.
• Object-storage native: Runs on local disk, RustFS, MinIO, AWS S3, Cloudflare R2 or Google Cloud Storage, with data kept in open, versioned Lance files.
• Declared as code: A cluster.yaml describes graphs, schemas, stored queries, embedding providers and policies, applied with a Terraform-style plan and apply workflow.
• Security as code: Cedar policies are enforced server-side on every mutation, with bearer authentication, server-resolved actors and audit tracking.
• Agent-ready tooling: Includes an MCP server, SDKs and an agent skill that teaches coding agents how to operate a cluster.

Omnigraph is a strong option for teams building agent memory, company knowledge graphs, context graphs or versioned research data, who want an alternative to Neo4j, TigerGraph or Amazon Neptune without running a separate database cluster. Its branching model, open storage format and combined graph and vector search make it especially well suited to AI applications where context has to be both reliable and reviewable.