Omnigraph

Omnigraph is a graph database built on object storage, designed for many long-running agents that share one typed world model, retrieve context from it, and write back to it. Agents and UIs become stateless consumers of durable state, and agents work with graph data the way engineers work with code: branched, typed, reviewed, merged, and audited.

Schemas declare node and edge types, keys, and constraints, and the engine enforces them on every write. One query combines graph traversal, vector nearest-neighbor search, full-text (BM25) search, and Reciprocal Rank Fusion with filtering. Agents write to their own branch per agent or task, and changes land through a three-way merge, with time travel across the whole graph.

A deployment is a cluster declared as code: a YAML file lists graphs, schemas, stored queries, embedding providers, and policies, and a plan step previews changes before an apply step converges them, in the style of Terraform. Cedar policies are enforced server-side on every mutation, per graph, per branch, or server-wide, with hashed bearer tokens and actors resolved on the server. Data is stored in the open, versioned Lance format on local disk or any S3-compatible store (RustFS, MinIO, AWS S3, R2, GCS), with Azure in preview.

Access comes through a CLI, a TypeScript SDK, an MCP server for LLM hosts, and an HTTP API with an OpenAPI spec. An agent skill teaches coding agents to operate clusters, evolve schemas, write data, and manage branches and policies. Example uses include a company knowledge graph, versioned agent memory, a decision-trace context graph, an issue and dependency graph for coding agents, and an R&D data layer.