---
title: Agents
description: pi, Claude Code, Codex, Cursor, OpenCode and Grok Build, side by side in one space. Each is its maker's official CLI, signed in through its own login.
updated: 2026-10-29
canonical: https://turnpanel.com/docs/agents
markdown: https://turnpanel.com/docs/agents.md
llms: https://turnpanel.com/llms.txt
---

# Agents

pi, Claude Code, Codex, Cursor, OpenCode and Grok Build, side by side in one space. Each is its maker's official CLI, signed in through its own login.

## The agents TurnPanel runs

TurnPanel does not contain an agent. It runs the ones you already use, in your spaces, and gives them the same memory, skills, tools and browser.

- **pi.** A version of the open-source pi agent, built in and ready at once. It works with any model: a local one, one from your own provider, or a TurnPanel-hosted one.
- **Claude Code.** Anthropic's `claude` CLI. You sign in through its own login: open its link and paste back the code it gives you.
- **Codex.** OpenAI's `codex` CLI, signed in through its own device-code login, or with an API key.
- **Cursor.** Cursor's `cursor-agent` CLI, signed in through its own device-code login.
- **OpenCode.** The `opencode` CLI, with an API key from a provider.
- **Grok Build.** xAI's `grok` CLI, signed in through its own device-code login.


## Installing an agent in a space

Pick your agents during setup and TurnPanel installs them into your first space. Later, a space admin can add, update or remove agents in the space's settings under Harnesses. Versions are pinned, so nothing updates on its own.


## Signing in to each agent

You sign in to each agent once, through that agent's own login, from any space where it is installed or from your profile. Your jobs then use your account in every space you open. Someone else's sign-in is never visible to you, and yours is never visible to them. An admin can remove a person's sign-in but cannot see or use it.

How it works depends on the agent: a link and a code to enter, a link and a box to paste a code into, or an API key that you save as your own secret. The dialog walks you through it. TurnPanel keeps the sign-in encrypted and hands it only to your own processes of that agent.

> **Automations use your sign-in**
> An automation runs as the person who set it up, so its jobs use that person's agent sign-in, whether they made it themselves or asked their agent to. One set to run as the space has no person behind it and gets no agent sign-in; it can use pi with local models or API-key providers.


## Models

pi uses TurnPanel's model picker (see [Models](/docs/models)). Every other agent uses its own models, listed after you sign in, with options such as effort. TurnPanel does not route their requests.


## Agents you run yourself

A Claude Code in your own terminal or an editor can connect to a space too. Paste the space's MCP address and a key into any MCP client. Everything such an agent does is limited by the key. A command that sets this up for you, `turnpanel connect`, is not available yet.


## Agents can start agents

Every agent knows it can start another job on any agent and model the space allows. A space admin can limit which ones, in the space's settings under Models.