A Graph Database That Branches Like Git and Bills Like S3

What it does

Omnigraph is a Rust-based graph database and server that stores its data in the Lance columnar format on any S3-compatible object store. You declare graphs, schemas, stored queries, and Cedar access policies as code inside a cluster directory, then converge the state with plan-and-apply semantics. Clients query and mutate the graph via CLI, HTTP, or TypeScript SDK, while the server enforces policies on every write path.

The interesting bit

Instead of versioning as an afterthought, Omnigraph treats the graph like a codebase: agents work on isolated branches, changes are reviewed, and merges happen across the entire graph structure. Because the underlying Lance format is branchable and time-travelable by design, you get git-style workflows without bolting a separate versioning layer on top of a traditional database.

Key highlights

• Multimodal retrieval in a single query runtime: graph traversal, vector ANN, full-text search, and Reciprocal Rank Fusion.
• Security as code: Cedar policies bound per-graph or server-wide, enforced identically across HTTP, CLI, and embedded SDK paths.
• Infrastructure-agnostic deployment: runs on any S3-compatible store—on-prem RustFS or MinIO, AWS S3, R2, or GCS—so data stays inside your storage boundary.
• Declarative management: a cluster.yaml defines the multigraph configuration; changes are previewed and applied idempotently.
• Embedded local mode for quick experiments: file-backed graphs that need no server or object store.

Caveats

• The Python SDK is marked “coming soon,” so Python-first stacks must use the HTTP/OpenAPI interface for now.
• Local development requires more than just Rust: the README notes that full CI and some test flows need protobuf-compiler, and S3 integration tests expect an S3-compatible endpoint like RustFS.
• With 506 stars, the project is early; the README pitches ambitious use cases but does not make explicit maturity or performance claims.

Verdict

A strong candidate if you are building agent fleets that need durable, branchable shared memory and want to keep data on existing object storage. Look elsewhere if you need a proven, general-purpose graph database with a mature ecosystem today.