ModernRelay / omnigraph

项目摘要

Omnigraph is a lakehouse-native graph engine designed for multi-agent coordination and context assembly. It combines graph traversal, vector search, and full-text retrieval with git-style branching workflows, allowing hundreds of agents to operate on parallel isolated branches and merge changes safely. Built in Rust, it runs on S3-compatible object stores and uses the Lance columnar format for versioned, time-travelable storage.

为什么值得关注

Omnigraph introduces git-style branching and versioning to graph databases, enabling safe parallel operations by fleets of AI agents. Its multimodal retrieval engine combines graph traversal, vector ANN, and full-text search with Reciprocal Rank Fusion in a single query runtime, while its lakehouse-native architecture ensures data never leaves the user's object store.

优势

• Novel git-style branching for safe parallel agent operations on graph data
• Unified multimodal retrieval combining graph, vector, and full-text search with Reciprocal Rank Fusion
• Lakehouse-native architecture with versioned, time-travelable storage via Lance format
• Strong security model with Cedar policy enforcement server-side on every mutation
• Infrastructure-agnostic design running on any S3-compatible object store

局限性

• No automated tests despite having CI, raising concerns about reliability and production readiness
• Very new project (3 months old) with limited contributor base and unproven track record
• No examples provided, making it harder for new users to evaluate and adopt

使用场景

• Multi-agent coordination with isolated branching and safe merge workflows
• Knowledge graph construction and querying for organizational memory
• Context assembly for RAG systems combining multiple retrieval methods