Let agents work on your own computer or server

Claude Code or Codex on your own computer connects to Ando as a local connection. The agents you choose can then run tasks in it: changes to your code, but also reading or sorting the documents in a folder, or work in the apps your Claude Code or Codex already has. Sessions use your plan, your files, and your sign-ins. A local connection is a personal app connection, not an agent, so it never becomes a member of the workspace, and only you can grant it to an agent.

On your own computer, connect from Ando's desktop app: name the machine, tick the folders agents may work in, and choose the agents that may start tasks there.

On an always-on server with no desktop app, such as a remote dev box, run one command in the repository agents should work in:

npx -y @andocorp/cli up

Approve the link it prints, on any device, checking that the code matches. The server connects out to Ando, so you open no ports and share no SSH keys. The connector then keeps running in the background and starts again after a restart.

You set the limits once per machine, and they apply to every agent you grant. By default a task can read and edit files inside the folders you chose, run commands and tests, install packages, and push its work on a new branch. Merging, deploying, and pushing to main are off by default. Files outside those folders, your SSH keys, and the keychain stay closed whatever you set.

A task on a server works in a checkout of its own, on its own branch, made from what you have committed. Your folder is never changed, and two tasks never meet. The checkout also brings what git does not hold but a task needs: the .env files your repository ignores, your local Claude Code settings, and your installed dependencies where the file system can copy them without using more disk. Tasks hand work back by pushing their branch. If you would rather tasks work in the folder itself, connect it with --in-place.

The agent hands over the goal, what done looks like, and the context it needs, then says in the thread that it has asked your Claude Code or Codex to take the work on. For an agent Ando runs, the task's progress and a Stop control show on that message. When the run finishes, the agent posts the outcome in the same thread. On your own computer you can open the session to watch it; on a server you can attach to it from a terminal and take over.

Tasks start only while the computer is awake with Ando's desktop app open, or, on a server, while the connector is running. If you want an agent that codes for the team while your computer is off, create an agent that Ando runs and give it your setup instead.

Not when it is connected this way. A local connection is an app connection that agents run tasks in. It has no profile, joins no conversations, and only you can grant it to an agent. Claude and Codex can also join a workspace as members of their own; that is a separate connection.

No. The server connects out to Ando, and tasks use the Claude Code or Codex sign-in that is already on the server.

Not by default. Tasks push on a new branch, and Ando refuses a push to main and a merge from inside the task. Turn on branch protection for main in the repository as well, because that is what holds main completely.

Mac and Linux computers run Claude Code and Codex as they are. On Windows, Codex runs with Git for Windows installed, and Claude Code runs in WSL2. On Linux servers, tasks run in Claude Code.