LOCUSTLocust with
Claude Code, as one of your team
Locust runs Claude Code as a teammate on your own computer, with Claude Code's own sign-in, on your Claude plan, beside teammates on other agents. Free and MIT, with no Locust account.
How Locust runs it
claude -p --output-format stream-json --permission-prompt-tool
Locust starts Claude Code in its print mode and reads its stream of events. When Claude Code would ask permission, the request comes to Locust, which shows it on a card and answers with what you chose. It never reads the agent's screen. If it is not installed yet, Locust's first launch shows how to get it.
Before it acts
Asks you first
In Approve each action: every file change, command and connector call, before it runs. In Ask and Plan, every connector call. In Edit, a connector call the teammate was not given, a command Claude Code does not run on its own say, and a change to a file outside the folder.
Without asking
Reading files, in every mode. In Edit, changes to files in the folder and the commands Claude Code runs on its own say. In Auto, everything.
Five modes, from Ask, which changes nothing, to Auto. Every agent, side by side: What Locust can stop.
What Locust adds to Claude Code
- More than one account: add one named Work, sign in with the agent’s own sign-in, and pick it on a teammate. Its usage is counted on its own.
- When your plan runs out, the stopped turn offers to carry the work on with another agent in one press, told what was asked and what was done. Settings can make it happen by itself, only to an agent you picked.
- Turns on your own cloud computer, a GitHub Codespace, streamed here like any turn; the change waits there until you apply it or open a pull request.
- Locust's browser beside the conversation: your teammate opens pages, clicks, types and reads the console and the network, and asks you on a card before a login, a payment, a send or a download.
- Every teammate on its own agent and model, side by side, with one Board for what needs you, what is working and what is done.
- Blind compare: the same task on several agents, names hidden until you keep one.
- Hand-offs by name between teammates on different agents, and a checker you pick, on another AI if you like, who must approve before a run counts.
- A local record of every run, kept if the window closes, and Undo for a teammate’s turn.
- Your phone from anywhere, sealed end to end, and Discord, Slack or Telegram if you turn them on.
Locust with the others
Try Locust
Free and open source (MIT). Windows, Mac and Linux. No Locust account: it runs the agents you already have, signed in as you.