Skip to main content
Opengeni gives your product an agent that does real work: it calls your APIs, asks for approval before it acts, pauses for answers, and keeps going until the job is done. You add one backend route and one React component. Opengeni runs the sessions, streaming, approvals, history, and compute.

​
How it works

  1. Sign up at app.opengeni.ai and get an API key.
  2. Add one backend route and <OpenGeniChat /> to your app, or let Opengeni or your coding agent do it.
  3. Your users get agents in your product that call your APIs, ask before acting, and finish the work.
The Opengeni web app: an agent found the cause of checkout errors, ran tests, opened a pull request, and is asking for approval to release the fix to production

​
Choose your path

Embed in your product

Add the full agent conversation to your app. Hand the integration to your coding agent, or follow four manual steps.

Use Opengeni for your team

Sign in at app.opengeni.ai, connect a model, and give agents work from the web app.

Self-host

Run the same Apache-2.0 platform on your own infrastructure with Helm and Terraform.

​
The fastest way to embed

Most teams integrate Opengeni with a coding agent such as Claude Code, Codex, or Cursor. Install the Opengeni plugin, give your agent one prompt, and it wires up your backend, your users, and the UI.
See Embed with your coding agent and the plugin guide.

​
What you get

  • A complete conversation UI. OpenGeniChat gives each user their chats and a conversation with streaming replies, tool calls, approvals, questions, attachments, queueing, and steering. Brand it with CSS tokens.
  • Your tools, your users. The agent calls your product through an MCP server or an OpenAPI description, as the signed-in user, with short-lived tokens you mint.
  • Durable sessions. Every event lands in a Postgres log. Reloads, new clients, and audits replay the same history, and work survives worker restarts.
  • Agents that finish. Give a session a goal with success criteria and it keeps working until it completes the goal with evidence or pauses with a reason.
  • Humans in control. Tool approvals and structured questions pause the exact turn until someone answers.
  • Background work. Scheduled tasks and inbound webhooks start agents without a user in the loop.
  • Your choice of compute. Sessions run with no sandbox, in a managed sandbox, or on a machine you enroll.

​
Where to go next

How it fits together

The browser, your backend, and Opengeni, and how tenants map to workspaces.

Concepts

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

For AI agents

The integration skill, llms.txt, and the docs MCP server.

Source on GitHub

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