> ## Documentation Index
> Fetch the complete documentation index at: https://docs.opengeni.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Quickstart

> Use OpenGeni as an agent harness and run your first session.

Use OpenGeni as an agent harness to give agents work, follow their progress, and steer their sessions from the app. This quickstart uses the managed service at [app.opengeni.ai](https://app.opengeni.ai), with nothing to deploy. You can also [self-host the same platform](/guides/self-host).

To add agent capabilities to your own app, start with [Integrate your product](/guides/integrate-your-product).

<Steps>
  <Step title="Create an account and organization">
    Sign up with email and password. After sign-in, name your organization. OpenGeni creates it with you as owner, along with a Personal workspace for you. Team members join by invitation.
  </Step>

  <Step title="Connect a model">
    Sessions need inference. Organization owners choose one of:

    * a ChatGPT/Codex or SuperGrok subscription, connected through a device-code login;
    * a provider key such as OpenRouter or Vercel AI Gateway; or
    * prepaid OpenGeni credits.

    In organization **Models** settings, choose which workspaces can use each subscription or gateway and which models it enables.

    You can skip this during setup and do it later from settings.
  </Step>

  <Step title="Start a session">
    Describe the task. Optionally attach a repository, upload files, or pick the tools the agent may use. The session starts on a managed sandbox by default.
  </Step>

  <Step title="Follow and steer">
    The timeline streams the agent's messages, tool calls, and commands live. Messages you send while a turn is running queue behind it. Steer interrupts the running turn and delivers your message now. Pause holds the session, and Cancel ends it.
  </Step>

  <Step title="Answer the agent">
    When a tool needs approval, the session waits until you approve or reject it. When the agent asks a structured question, answer it in place and the same turn continues.
  </Step>
</Steps>

## Make it keep going

For work that should not stop at the first plausible answer, give the session a goal with success criteria. While the goal is active the agent keeps working until it explicitly completes the goal with evidence or pauses it with a rationale. See [Goals](/concepts/goals).

## Use your own computer

Enroll a laptop or server as a Connected Machine and target it when creating a session. The agent runs there directly, under your files and your git credentials, with no cloud box in the loop. See [Connect a machine](/guides/connect-a-machine).

## Call it from code

Everything the web app does goes through the public API. See [Integrate your product](/guides/integrate-your-product) and the [SDK reference](/reference/sdk).