ModernRelay/omnigraph

Lakehouse native graph engine with git-style workflows

Summary

Omnigraph is a Rust, object-storage-native graph database built around typed node and edge schemas, versioned Lance data, and Git-style branches. It lets agents and applications store durable context and combine graph traversal, vector approximate-nearest-neighbor search, full-text search, and Reciprocal Rank Fusion with schema enforcement and server-side policies.

What it is useful for

Omnigraph fits systems where several agents or services need a shared, changing model. A coding agent could store issues and dependencies in a “dev graph”; a research team could record experiments and evaluation datasets in branches; an internal knowledge assistant could query a company graph. Typed .pg schemas define node and edge types, keys, and constraints, which the engine enforces on writes.

Its retrieval runtime combines graph traversal, vector ANN, full-text search, Reciprocal Rank Fusion, and filtering in one query. Branching supports isolated proposals: create a branch, write and review there, then use a three-way merge. The README also describes time travel across the whole graph.

A cluster declares graphs, schemas, stored .gq queries, embedding providers, and policies in cluster.yaml. Cedar policies are enforced server-side on every mutation through the HTTP server, CLI, or embedded SDK.

Data lives in open, versioned Lance on local disk or an S3-compatible store. Documented options include RustFS, MinIO, AWS S3, R2, and GCS. That makes Omnigraph relevant to lakehouse, knowledge-graph, and context-graph workloads built around object storage.

How engineers can use it

The README provides this release-binary installer for omnigraph and omnigraph-server:

curl -fsSL https://raw.githubusercontent.com/ModernRelay/omnigraph/main/scripts/install.sh | bash

It installs both binaries into ~/.local/bin. A documented Homebrew alternative is:

brew tap ModernRelay/tap
brew install ModernRelay/tap/omnigraph

Create a cluster directory containing cluster.yaml, a .pg schema, stored queries, and a policy bundle. After configuring local or S3-compatible storage, validate and converge the cluster:

omnigraph cluster validate
omnigraph cluster plan --target ./company-brain
omnigraph cluster apply --target ./company-brain
omnigraph-server --cluster company-brain --bind 0.0.0.0:8080

With a default target configured in ~/.omnigraph/config.yaml, everyday operations can run by stored query or mutation name:

omnigraph query search_docs --params '{"q":"AI safety"}'
omnigraph mutate add_person --params '{"name":"Mina"}'
omnigraph branch create --from main agent/ingest-42
omnigraph branch merge agent/ingest-42 --into main

The README lists TypeScript SDK, MCP server, and HTTP/OpenAPI clients; the Python SDK is still marked “coming soon.” Azure blob roots remain a qualification preview and require the checked-in admission wrapper. The hosted cloud service is in private preview. For contributors, GQT and DST tests run separately, some flows require protobuf-compiler, and S3 integration tests expect an S3-compatible endpoint.