Supported agents

Coming soon

Which AI agents you can use in the Taller, how they connect with your own account and what permissions you give them.

The Taller works with any command-line AI agent: the agent picker simply launches that program in a terminal, inside the workspace. Many come already set up to launch with one click.

Most used agents

Claude Code (Anthropic): deep integration, with visible usage and limits and quick account switching.
Codex (OpenAI): deep integration, with visible usage and account switching.
Cursor CLI, Gemini, GitHub Copilot CLI, OpenCode, Aider, Goose and many more, with automatic setup.

If your agent isn't on the list, you can still open an empty terminal and launch it by hand.

Your account, on your computer

Each agent uses your own session: the one you signed into once in the terminal (for example, Claude Code keeps its own in ~/.claude). The Taller detects it on its own, with no extra setup.

Setting up Claude Code

Install Claude Code following Anthropic's guide.
Sign in once from any terminal.
Open a workspace in the Taller and choose Claude Code in the picker.

Usage and limits

For agents that allow it, the status bar shows how much of your plan you've used and how long until the limits reset. The Taller reads that data from the agent itself on your disk: it makes no extra calls and doesn't ask for another authorization. That's why the numbers update when the agent writes them, not in real time.

If you have several accounts, you can switch from one to another with one click, even with sessions open.

Agent permissions

By default, many agents launch in a mode that doesn't ask for confirmation before each command. It's convenient, but keep in mind:

A workspace is a separate copy of the code, not a safe box: the agent can read the files and use whatever network it can reach.
If you don't fully trust the task, change permissions to Manual in the agent settings. That way the agent will ask you for confirmation.

Each agent's status

Each agent tab shows its status: working, waiting for your answer, done, or done and not reviewed. That way you can tell at a glance which one needs your attention.

An agent won't start? See troubleshooting.