Omnigraph
Omnigraph is a typed, versioned graph database for agent memory, organizational knowledge, and hybrid retrieval.
Omnigraph is a typed, versioned graph database for agent memory, organizational knowledge, and hybrid retrieval.
What the product does and how it is positioned
Omnigraph combines a typed ontology with object-storage-backed graph data and Git-style branching, review, and merge workflows.
Its documented retrieval model combines graph traversal with vector, full-text, or hybrid ranking, and it can run embedded, in a user's cloud account, or on-premises.
Source-supported ways to use the product
Organize company knowledge, claims, decisions, and action items in one typed graph.
Maintain versioned memory for agents or tasks with branch-based writes and review before merging.
Connect people, projects, decisions, documents, and related organizational knowledge.
Represent hypotheses and results in a graph with changes available for review.
The documented workflow, where available
Describe typed nodes, edges, fields, constraints, and optional embeddings in the graph schema.
Create an agent or task branch so proposed writes do not directly modify the main graph.
Apply mutations on the branch, where each write has an author and is checked against the schema.
Inspect additions, changes, removals, and conflicts before accepting the branch.
Merge the reviewed branch into main, while retaining readable graph history and snapshots.
Omnigraph documents a branch-based model for agent memory: agents write to separate branches, schema and policy checks apply during writes, and changes can be diffed before merging into main.
Checks to run with your own material and workflow
What was checked and when
Answers based on the source-checked product record
It is described as a versioned graph database for typed organizational knowledge, agent memory, context graphs, and research graphs.
It supports graph traversal, vector search, full-text search, and hybrid ranking in queries over the same typed graph data.
The documented storage options include S3-compatible storage, Azure Blob in preview, and local disk, with embedded, cloud-account, and on-premises deployments.
Yes. The documented MCP server exposes an MCP endpoint, and the MCP integration can let clients read schemas, run queries, and write where policy permits.
Changes can be made on branches, inspected through diffs, reviewed for conflicts, and merged into main after approval.