Parallel agents

Run a yard of coding agents at once, each in its own workspace.

Kalauz is built around running agents in parallel. Each one lives in its own isolated workspace, so a yard of agents can grind through unrelated tasks at the same time without conflicts.

One agent per workspace

Every workspace runs against one branch and one worktree. That isolation is what makes parallelism safe: there's no shared working directory to race on, and a failed run only ever affects its own worktree. A workspace can hold more than one chat session, but they all operate on the same branch.

Screenshot — placeholderThe sidebar with several workspaces running at once, each showing its own status.

How many at once?

There's no hard cap — run as many workspaces as your machine will carry. In practice, Conductor's guidance is also ours: 3–5 at a time. More than that and you can't keep up with the reviews; fewer and you don't really need the tool.

Watching the yard

The sidebar is the control room. Each workspace card shows its name, branch, a changed-files badge, and a spinner while a session is running. You route your attention to whichever one needs it, and jump between them with the command palette (⌘K).

Run scripts and concurrency

Chats always run in parallel. The one thing you may want to serialize is the Run script — if several workspaces each try to bind the same dev-server port, they'll collide. Set a project's run-script mode to nonconcurrent and Kalauz gates it: only one workspace in that project runs its Run script at a time, and the others wait with a notice. See Testing for how run scripts work.

What's next?

Agent modes — picking a model and how hard it thinks.
Testing — setup and run scripts inside a workspace.