Omnigraph: Open source graph database for your AI agents
On November 12, 2026, a developer tool that keeps graphs as files in the user's own cloud storage went live on Product Hunt, where the listing showed 282 upvotes and 35 comments. Product Hunt describes the project as an object-storage native graph database that adds branching and a typed ontology, so that AI agents can share one knowledge base. The engine is released under the MIT license, and its code sits in a public GitHub repository under the ModernRelay organization.
Graph data is held in Lance, an open columnar format, on S3-compatible storage, local disk, or Azure Blob, where Azure support is still a preview. Writes are checked against a typed schema on every change, and access rules written in Cedar are enforced inside the engine. Each agent can work on its own branch, and a change reaches the main graph only after review and merge, with time travel available across the whole graph.
Retrieval runs through one query path that mixes graph traversal with vector search, full-text search, and hybrid ranking through Reciprocal Rank Fusion. Deployment options include an embedded mode, bring-your-own-cloud installs on AWS, Azure or GCP, and on-premises setups. The project site lists version 0.15.0 and about 2.1k GitHub stars.
Why it matters
For founders building agent products, the appeal is that shared memory and context can live in storage the customer already controls. That reduces lock-in concerns for buyers in privacy-sensitive sectors and removes a recurring per-call memory fee from the cost base. Teams can also review and validate changes before they enter shared context, a control that enterprise buyers tend to ask about during procurement.
For investors, the launch shows the agent data layer splitting between open formats that anyone can self-host and managed offerings that charge for hosting. The MIT license invites forks and outside contributions, and the project's own FAQ says the engine is free, so customers pay only for their storage and compute. That sets a price floor for any commercial competitor and makes hosting, support, and enterprise features the likely revenue paths.
Key points
- 1The launch on Product Hunt showed 282 upvotes and 35 comments in the listing.
- 2Graphs are stored as open Lance files on S3-compatible storage that the user controls.
- 3Each agent writes on a separate branch, and merges into main require review first.
- 4One query can combine graph traversal, vector search, full-text search, and hybrid ranking.
- 5The engine is MIT-licensed, and the project site lists version 0.15.0 with about 2.1k GitHub stars.
Analysis
Graph databases are an established category, and Neo4j is the best-known vendor. Wikipedia notes that Neo4j added vector search in 2023, launched agentic AI offerings in 2025, and introduced its Infinigraph architecture in September 2025. Omnigraph's own FAQ calls Neo4j mature and positions Omnigraph as the better fit for agent workloads specifically. The real difference is less the query surface than the combination of Git-style branching, open storage, and in-engine policy, which few incumbents package together.
The model carries a learning curve. An architecture teardown by Bicarait, published in September 2026, says engineers must learn graph schema files, Cedar policy syntax, and data branching concepts, and that the tool is not a drop-in replacement for Postgres. The Product Hunt maker comments also say that Omnigraph stores and connects data but does not handle ingestion pipelines, so teams must build that layer themselves. Release notes for version 0.11.0 list fixes for false merge conflicts and stale fast-forward results, which suggests the branching layer is still maturing.
Several things are worth tracking next. The project is pre-1.0, and the v0.11.0 notes already flagged changes to identity fields and result formats, so breaking changes are plausible. Buyers should watch whether the managed Cloud preview and bring-your-own-cloud options turn open-source users into paying customers. The Azure path is still labeled preview. Fork activity and contributor growth on GitHub will likely show adoption before revenue does.
