Command many coding agents.
Kalauz turns a repo into a control room. Every agent runs in its own git worktree, every change is reviewable in place, and profiles keep work, client, and personal credentials from bleeding into each other.
A control room, not a chat window.
The sweet spot is three to five workspaces: enough leverage to feel like more hands, few enough to stay in command. Every block earns its place.
Parallel by worktree
Each workspace gets its own branch and git worktree. True isolation, no worktree gymnastics, no agents stepping on each other.
Review in place
Read every change on the diff canvas, run it, and act on it without switching tools or windows.
Native on every OS
A single binary on C# and Avalonia. macOS, Windows, and Linux as first-class citizens, with no heavy runtime to drag around and light enough to keep five workspaces alive at once.
Your agents, your CLIs
Bring the agents you already trust, like Claude Code or Codex, along with the tools already on your PATH. Kalauz routes them; it doesn’t replace them.
Profiles with separate credentials
Switch between work, client, and personal profiles. Each one carries its own agent API keys, git identity, model provider, and privacy defaults.
Sealed ground
Work stays contained to its worktree. An agent can’t reach across into another workspace’s files, branch, or history.
New work is one move.
Workspace, branch, worktree. The same shape, repeated for every agent you set running.
Open a workspace
Point Kalauz at a repo. A workspace is one move: a fresh branch and its own git worktree, sealed off from everything else on disk.
Run the yard
Hand each workspace to an agent like Claude Code or Codex, and let them work in parallel. Nothing one agent does can touch another’s ground.
Review and ship
Read the diff in place, run it, open the PR, merge. You stay at the junction routing the work, never leaving the window.
Credentials live in the right profile.
Each profile carries its own agent API keys, git identity, model provider, and privacy defaults. Switch the whole set in one move, so a client repo never runs with your side-project credentials.
- Org agent keys
- [email protected] git identity
- Bedrock provider · data privacy mode on
- Your own keys
- you@home git identity
- Direct provider · telemetry your call
Run Kalauz on your own infrastructure.
Hand it to a developer and keep every token on your own rails. Turn on data privacy mode and point Kalauz at your own model provider, so the work never leaves your boundary.
Data privacy mode
Telemetry off, prompts and code stay local to the session. Nothing is sent anywhere you didn’t configure.
AWS Bedrock
Route agents through your Bedrock endpoint and models. Spend and logs run through your AWS account, in your region.
Google Vertex AI
Same for Vertex: your GCP project, your quotas, your data-residency rules. The org owns the pipe, and the developer just works.
Leverage, not chaos.
Five agents feels like extra hands; fifty is chaos. Kalauz is built for the range where one person can still keep command of many hands.
Workspaces in flight: the range that feels like more hands without losing the thread.
macOS, Windows, and Linux from one native codebase. No second-class platform.
A single native build on C# and Avalonia. No Electron, no runtime to babysit.
Questions & answers
The short, exact version. Each answer names the tradeoff.
Take command of the yard.
Download Kalauz for your OS, point it at a repo, and run your first parallel workspace. It’s one binary, with nothing to babysit.