ModernRelay/omnigraph

Lakehouse native graph engine with git-style workflows

What it solves

Omnigraph is a lakehouse graph database designed specifically for AI agents. It provides a durable, versioned state and coordination layer that allows fleets of agents to operate on parallel isolated branches, ensuring that changes to the knowledge graph are reviewed and merged safely, similar to Git. It solves the problem of managing shared, multimodal retrieval and agentic memory across large-scale agentic workflows.

How it works

Omnigraph uses the Lance columnar format for open, versioned storage on S3-compatible object stores. It implements a "cluster" model where graphs, schemas, stored queries, and security policies are declared as code (via cluster.yaml). The system supports multimodal retrieval by combining graph traversal, vector ANN, full-text search, and Reciprocal Rank Fusion in a single query runtime. Security is enforced server-side using Cedar policies.

Who it’s for

It is built for developers building fleets of AI agents that require a shared, versioned memory or a "company brain" for unified organizational knowledge. It is also suitable for R&D and ML data layers where experiments and trials need to be versioned for training and evaluation.

Highlights

  • Git-style Branching: Agents can enrich the graph on isolated branches and merge changes after review.
  • Multimodal Retrieval: Unified query runtime for graph, vector, and full-text search.
  • Infrastructure Native: Runs on any S3-compatible object store (AWS S3, R2, GCS, or on-prem via MinIO/RustFS).
  • Declared as Code: Cluster configurations, schemas, and policies are managed via a Terraform-style plan and apply workflow.
  • Cedar Policy Enforcement: Server-side security and access control enforced on every mutation.

Related

  • Project
    lancedb/lancedb

    A multimodal AI lakehouse and vector database built on the Lance columnar format for fast, scalable search over petabytes of vectors and multimodal data.

  • Project
    lance-format/lance

    An open lakehouse format for multimodal AI that provides high-performance vector search, random access, and native storage for images, video, and audio.

  • Project
    laminlabs/lamindb

    An open-source data management tool for multimodal AI that provides end-to-end traceability, versioning, and governance for datasets and the code that produces them.

  • Project
    memgraph/memgraph

    Memgraph is an in‑memory graph database (C/C++) that combines sub‑millisecond graph traversals with built‑in vector, text, and geospatial indexes. It ships a 40+ algorithm library (MAGE), an “Atomic GraphRAG” query model, and an AI Toolkit for LLM‑aware workflows, while offering enterprise features like Raft HA, multi‑tenancy, and fine‑grained security.

  • Project
    xoai/sage-wiki

    A graph-based knowledge base that uses an LLM compiler to turn raw documents into an interlinked wiki and knowledge graph for AI agents and humans.