How top AI teams scale to 1,000+ agents

Omnigraph is a revolutionary AI tool designed to scale to over 1,000 agents by utilizing a native graph database with branching and a typed ontology. This innovative solution offers a unified retrieval engine, combining vector, BM25, and graph capabilities, to streamline the retrieval and updating of context for multiple agents sharing a world model.

Omnigraph's graph database allows agents and UIs to function as thin, stateless consumers of durable state, making it an ideal choice for long-horizon AI agents. The tool is engineered to be run by coding agents, with two primary ways to utilize it: teaching an agent a playbook or allowing the agent to set everything up from scratch.

The Omnigraph CLI carries an operational playbook covering various aspects such as cluster mode, config surfaces, schema evolution, query linting, data writes, branches, Cedar policy, and common gotchas. This playbook ensures seamless integration and operation for AI agents. Omnigraph also provides ready-to-run graphs with real seed data, enabling users to quickly see how the tool can be tailored to a specific domain.

For first-time users, the cluster guide offers an introduction to deploying and managing clusters, including writer ownership transfer and authentication. The deployment guide covers containers, object stores, and Azure writers. Applying a cluster policy grants the authenticated operator config_manage permissions, as well as read and schema_apply permissions for schema changes.

Omnigraph's architecture supports server targets, allowing users to deploy local edits using specific configurations and targets. Stored queries and mutations can be run by name, enabling agents to write in isolation. With its seamless control-plane login and organization operations, Omnigraph offers a comprehensive solution for AI teams seeking to scale to large agent networks while maintaining efficiency and coordination.