Summary

Omnigraph is a lakehouse-native graph engine designed as an operational state and coordination layer for AI agents. It provides Git-style versioning and branching for graph data, enabling robust context assembly and memory management. The engine targets developers building agentic systems, offering multimodal retrieval and a "data as code" approach for schema, context, and security.

How It Works

This engine leverages the Lance format for its open storage layer, natively supporting object storage like S3 and RustFS. Its core innovation lies in applying Git-like workflows (branching, merging) to graph data, facilitating versioned R&D, agent memory, and company knowledge bases. It offers multimodal retrieval combining graph structures with vector embeddings, full-text search, and filters, optimized for assembling complex contexts.

Quick Start & Requirements

Installation is straightforward via a bash script (curl -fsSL https://raw.githubusercontent.com/ModernRelay/omnigraph/main/scripts/install.sh | bash) or Homebrew (brew tap ModernRelay/tap && brew install ModernRelay/tap/omnigraph). A one-command local bootstrap (curl -fsSL https://raw.githubusercontent.com/ModernRelay/omnigraph/main/scripts/local-rustfs-bootstrap.sh | bash) sets up a local RustFS instance and Omnigraph server, requiring Docker. Building from source requires a Rust stable toolchain. Links to starter graphs and agent skills are available in the omnigraph-cookbooks repository.

Highlighted Details

• Git-style versioning and branching for graph data.
• Multimodal retrieval (graph, vector, FTS, filters).
• Object storage native support (S3, RustFS) with native blob-as-data.
• "As Code" philosophy: Schema, Context, Security, and planned Dashboards.
• Server-side enforcement of Cedar security policies on mutations.
• Lance format as an open, efficient storage layer.

Maintenance & Community

The project encourages collaboration via its Slack community for questions and feedback. Contributions are welcomed, with a preference for discussing design and specs via issues before submitting large code changes.

Licensing & Compatibility

The project's license is not explicitly stated in the provided README, which may require further investigation for commercial use or integration into closed-source projects.

Limitations & Caveats

The "Dashboards AS CODE" feature is noted as "coming." The local bootstrap process relies on Docker, and full CI/testing may require a protobuf compiler. No other specific limitations, alpha status, or known bugs are detailed in the README.