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

# SET UP YOUR AGENT

> Point a coding agent at this page so it installs the Oumi CLI and uses it for you

> [!NOTE]
>
> **Copy these instructions for your agent**
>
> ```text theme={null}
> Set up Oumi from https://docs.oumi.ai/cli/agents.md
> ```

Paste the prompt above into Claude Code, Codex, Cursor, or any other coding agent. The agent reads this page, installs `oumi-cli`, and asks you to sign in through your browser. After that, it runs Oumi for you.

## WHAT YOUR AGENT CAN DO

Instructions for the agent: after setup, you can do the following with `oumi-cli`.

* Train a model from a dataset: [https://docs.oumi.ai/cli/workflows/train-a-model.md](https://docs.oumi.ai/cli/workflows/train-a-model.md)
* Evaluate a model and find where it fails: [https://docs.oumi.ai/cli/workflows/evaluate-a-model.md](https://docs.oumi.ai/cli/workflows/evaluate-a-model.md)
* Deploy a model behind an inference endpoint: [https://docs.oumi.ai/cli/workflows/deploy-a-model.md](https://docs.oumi.ai/cli/workflows/deploy-a-model.md)
* Upload, version, and analyze datasets: [https://docs.oumi.ai/cli/commands/datasets.md](https://docs.oumi.ai/cli/commands/datasets.md)
* Upload and manage project files: [https://docs.oumi.ai/cli/commands/files.md](https://docs.oumi.ai/cli/commands/files.md)
* Create and manage evaluators: [https://docs.oumi.ai/cli/commands/evaluators.md](https://docs.oumi.ai/cli/commands/evaluators.md)
* Score live deployment traffic with monitors: [https://docs.oumi.ai/cli/commands/monitors.md](https://docs.oumi.ai/cli/commands/monitors.md)
* Track and cancel long-running operations: [https://docs.oumi.ai/cli/commands/operations.md](https://docs.oumi.ai/cli/commands/operations.md)

## SET UP

Instructions for the agent: complete these steps in order. If your environment blocks the install,
explain why and give the human the commands to run themselves. Do not attempt another install method.

**1. Install the CLI**

Follow [https://docs.oumi.ai/cli/installation.md](https://docs.oumi.ai/cli/installation.md) to install `oumi-cli`, then confirm it runs.

**2. Ask the human to sign in**

Stop here and ask the human to run `oumi-cli auth login` in their own terminal. The command opens the Oumi platform in a browser. The human signs in and clicks **Authorize**, and the CLI saves the key. No key is copied.

You cannot run this command yourself, because it needs an interactive terminal.

If the machine has no browser (for example, over SSH or in a container), tell the human to run `oumi-cli auth login --no-browser`, open the printed URL on any device, click **Authorize**, and paste the key that the page shows at the prompt. Never ask the human to paste the key in chat. Never put the key in a command you run.

After the human confirms, verify the login:

```bash theme={null}
oumi-cli auth status --output json
```

**3. Choose the organization and project**

Ask the human which organization and project to use. Set both once:

```bash theme={null}
oumi-cli organizations set <ORG_ID>
oumi-cli projects set <PROJECT_ID>
```

The human can list valid IDs with the `organizations list` and `projects list` commands. See [https://docs.oumi.ai/cli/commands/organizations.md](https://docs.oumi.ai/cli/commands/organizations.md) and [https://docs.oumi.ai/cli/commands/projects.md](https://docs.oumi.ai/cli/commands/projects.md).

## LEARN THE CLI

Instructions for the agent: run this command, and follow its output for all later use of `oumi-cli`.

```bash theme={null}
oumi-cli agent-instructions
```

That output is the source of truth for flags, output formats, and exit codes. It matches the version of the CLI you installed.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.