omnigraph

Why we included this project

Omnigraph treats organizational knowledge as a versioned graph that agents can read from and write back to, rather than a pile of static documents. What makes it worth a look is the Git-style workflow applied to data: branches let many agents enrich the graph in parallel, and changes get reviewed and merged instead of overwriting one another. The query runtime combines graph traversal, vector search, and full-text lookup in one place, with Reciprocal Rank Fusion to blend the results, which helps when you are assembling context for coding or research agents. Everything is declared as code in a cluster config and enforced server-side, and storage sits on Lance over any S3-compatible object store, so the data stays on infrastructure you control. Teams already running several agents that need shared, safe, queryable state will get the most out of it.