---
title: "lizard agents"
description: "lizard agents connects your ChatGPT, Claude, or API-key credentials for coding agents that run in sandboxes, and sets each agent's default model."
url: https://lizard.build/docs/cli/agents/
updated: 2026-10-12
---

# lizard agents

Connect the credentials that coding agents use in sandboxes, and set each agent's defaults. Run the agents with [`lizard sandbox prompt`](https://lizard.build/docs/cli/sandbox.md#coding-agents).

## Usage

```bash
lizard agents status
lizard agents login [chatgpt] [flags]
lizard agents add <credential> [flags]
lizard agents remove <credential>
lizard agents use <agent> [flags]
```

Credentials belong to your account. Connect a credential once to use it for agent runs in any sandbox. For a ChatGPT login, the platform keeps the long-lived token and hands a sandbox a short-lived access token for each run. Lizard passes API keys and the stored Claude subscription token unchanged in each run's environment. Use credentials you can revoke. These commands accept `-p, --project` but ignore it.

## Agents and credentials

| Agent | Name | Accepts |
|-------|------|---------|
| Codex | `codex` | one of: `chatgpt`, `openai`, `deepseek`, `openrouter` |
| Claude Code | `claude` | one of: `claude`, `anthropic`, `openrouter` |
| pi | `pi` | any of: `claude`, `chatgpt`, `anthropic`, `openai`, `openrouter`, `llmgateway` |
| OpenCode | `opencode` | any of: `chatgpt`, `anthropic`, `openai`, `openrouter`, `llmgateway` |
| Prime Agent | `prime` | any of: `claude`, `chatgpt`, `anthropic`, `openai`, `openrouter`, `llmgateway` |

Kimi Code (`kimi`, with `kimi` or `moonshot` credentials) is listed but can't run yet.

## Subcommands

### `lizard agents status`

Show which credentials are connected and which agents can run. Plain `lizard agents` does the same.

### `lizard agents login [chatgpt]`

Connect your ChatGPT subscription with a device code. The CLI prints a URL and a code; open the URL, sign in, and enter the code within 15 minutes. The command waits until you approve.

| Flag | Description |
|------|-------------|
| `--open` | Open the sign-in page in your browser |

### `lizard agents add <credential>`

Add an API key: `anthropic`, `openai`, `openrouter`, `llmgateway`, `deepseek`, or `moonshot`. For a Claude subscription, run `claude setup-token` on your machine and add its token as `claude`.

| Flag | Description |
|------|-------------|
| `--key <value>` | The key or token. Without it, the CLI reads stdin, or asks |

### `lizard agents remove <credential>`

Remove a connected credential (alias `logout`).

### `lizard agents use <agent>`

Set an agent's credential and defaults.

| Flag | Description |
|------|-------------|
| `--credential <name>` | For Codex and Claude Code: which credential to use |
| `-m, --model <model>` | Default model (`none` clears it) |
| `--effort <level>` | Default reasoning effort (`none` clears it) |

## Examples

### Connect ChatGPT and run Codex

```bash
lizard agents login chatgpt --open
lizard sandbox create -t codex
lizard sandbox prompt sb_abc123 "add a health check endpoint and test it"
```

### Add a Claude subscription for Claude Code

```bash
claude setup-token
lizard agents add claude
lizard sandbox create -t claude
lizard sandbox prompt sb_abc123 --agent claude "fix the failing tests"
```

### Run Codex on an OpenAI API key

```bash
lizard agents add openai < openai.key
lizard agents use codex --credential openai --effort high
```

## See also

- [lizard sandbox](https://lizard.build/docs/cli/sandbox.md#coding-agents): `sandbox prompt`, `conversations`, and `interrupt`
- [Sandboxes](https://lizard.build/docs/sandboxes.md): templates and lifecycle