Skip to main content
OpenGeni is the platform layer that makes long-running AI agents safe to trust with real work: durable, replayable sessions; human approvals; governed credentials and memory; and a choice of where every session runs, either a managed sandbox or your own hardware. The included web app lets you use OpenGeni directly as an agent harness. Your own products can use the same session-based API to create, steer, observe, interrupt, and replay agent runs, with OpenGeni owning durable session state, event history, approvals, and outputs.

​
Two ways to use it

Agent harness

Use the OpenGeni app to give agents work, follow their progress, steer sessions, approve tool use, and answer questions. Start with your own tasks or work with your team.

Integrate into your product

Add agent capabilities to your product with the API, SDK, and React components. Your product owns its users and interface; OpenGeni handles agent execution and durable sessions.

​
Choose a deployment

For either usage mode, choose where OpenGeni runs. Use the managed service at app.opengeni.ai, or self-host OpenGeni in your own infrastructure. For product integrations, check which features your chosen deployment supports. The new chat setup currently requires a compatible API deployment and source workspace packages; the hosted API does not yet support that setup.

​
What you get

  • Durable, replayable sessions. Every event lands in a Postgres event log. Live streams backfill from it, so a browser reload, a new client, or an audit replays the same history.
  • Sessions that finish the job. A session can carry a goal with success criteria. Stopping is an explicit act: the agent completes the goal with evidence, pauses it with a rationale, or a human interrupts.
  • Governance built in. Human approvals gate tool use, agents pause durably for structured answers, and credentials are brokered per session.
  • Your hardware as a first-class target. A Connected Machine runs sessions on a computer you enroll. It only dials out, receives no platform credentials, and can be revoked with one click.
  • Workspace memory and knowledge. Agents save and correct durable facts, and workspace documents are indexed for retrieval.

​
Where to go next

Concepts

Sessions, turns, goals, approvals, compute targets, and memory.

SDK reference

Explore the chat integration, full session API, and React components for your product.

Connect a machine

Enroll a laptop, build server, or GPU box and run sessions on it directly.

Source on GitHub

The runtime, web app, SDK, deployment artifacts, and deeper engineering docs.