AI coding agents your bot can hand work to

A coding agent is good at code and bad at knowing when to run. A Casama bot can hand a task to Claude Code, Codex or Cursor, on a sandboxed computer, and come back with a summary and the changed files.

Updated

What the bot does

Where it stops to ask

You set these rules per bot and per action. See how approvals work.

What it looks like

  1. You, from your phone: "Use Claude Code to fix the failing tests in the billing package."
  2. The bot starts the agent on its computer with the repository checked out.
  3. Ten minutes later: "Fixed 3 tests. Changed 2 files. Here's the diff."
  4. You ask it to open a pull request. It asks first, then does.

Why its own computer matters

Each Casama bot works on a sandboxed Linux machine of its own, with a browser, files and a shell. It keeps going when you close your laptop, it can't reach your own files, and you can watch its screen or take over at any time. Before a command that deletes data, its computer is saved, so the work can be undone.

Set it up

  1. For the bot's computer: add the coding agent's API key as a named secret.
  2. For your own computer: install the agent's command-line tool, sign in, and share the project folder in the desktop app.
  3. Ask the bot in plain words which agent to use and where. The Coder persona brings a code-change skill.

Ready-made skills, starter bots and routines are in the gallery.

Questions

Which coding agents work?

Claude Code, Codex, Cursor, Pi and T3 Code on your own computer; Claude Code, Codex and Cursor on the bot's computer; Cursor in the cloud.

Does Casama replace my coding agent?

No. It runs the one you choose, on a schedule or from your phone, with approvals around it.

Can I use it with MCP?

Yes. Casama has an MCP server, so a coding assistant can also hand work to your bots.

See what else Casama bots do.

Join the waitlist Run it on your own server