Zuse

Zuse is an open source desktop application that provides developers with a unified workspace to run multiple AI coding agents simultaneously across their Git repositories. Rather than bouncing between separate terminal windows, browser tabs, and IDE extensions for each agent, Zuse consolidates everything into a single interface where developers can orchestrate Claude Code, Codex, Cursor, Gemini, Grok, OpenCode, and Kiro working in parallel on different tasks. The application runs locally on macOS, connecting to whatever AI provider subscriptions a developer already has rather than requiring new accounts or selling its own model credits.

The coordination problem Zuse solves becomes apparent once you try running more than one AI coding assistant on the same codebase. Hand two agents different tasks and they will both want to make changes, often to the same files, often creating conflicts that take longer to untangle than the original work would have taken manually. Zuse sidesteps this entirely by giving each task its own Git worktree, an isolated working directory that shares the repository history but keeps each agent's changes separate until you explicitly merge them. Two agents can edit the same codebase without ever touching each other's work in progress, and when they finish you review their diffs independently before deciding what makes it into your main branch.

The workflow starts with opening a repository in Zuse and selecting which agent you want to handle a particular task. You describe what needs to happen, the agent runs using your existing subscription, and Zuse captures everything in a persistent session. The workspace displays the conversation history, the files being edited, terminal output, diffs, check results, browser runs if applicable, and pull request context all in one place. Nothing hides in a background process or requires switching to another tool to inspect. If the agent stalls on something or you decide a different model might handle the problem better, you can swap agents mid task without losing the session context. The worktree, the changes made so far, and the conversation all stay intact while the new agent picks up where the previous one left off.

Developers who maintain multiple AI subscriptions will find Zuse particularly useful because it lets them use those subscriptions together rather than in isolation. Someone paying for Claude and Codex and Cursor can spin up all three on different parts of a project without needing to manually coordinate which tool is editing which files. The orchestration layer that Zuse provides is agnostic about which provider powers each task, it simply routes your existing credentials and lets you leverage each model's strengths on the tasks where those strengths matter most. One agent might be better at refactoring while another handles documentation, and Zuse makes it trivial to split work that way.

The parallel execution model scales to realistic workloads. If you have twenty issues in Linear or a backlog of tickets in whatever tracker you use, you can assign each one to its own worktree with its own agent and let them all run at once. The interface gives you visibility into all of them from a single dashboard, showing progress, terminal output, and diffs as they emerge. You review changes when each task completes, approve or request revisions, and merge the results when you are satisfied. This turns what would otherwise be serial work, waiting for one agent to finish before starting another, into a supervisory role where you watch many tasks progress simultaneously and intervene only when something needs human judgment.

Zuse does not abstract away the underlying Git mechanics or hide what agents are doing behind a polished consumer interface. The workspace exposes branches, changes, commands, permissions, and results explicitly, which makes it powerful for developers already comfortable with Git workflows but potentially overwhelming for those looking for a more guided experience. Session persistence means you can close Zuse, come back later, and pick up exactly where you left off with full context retained, but understanding what happened in a session requires reading terminal output and diffs rather than summaries.

The project is open source under an MIT license with the source code available on GitHub. There is no hosted service, no subscription fee for Zuse itself, and no lock in to a particular provider. You download the macOS desktop application, point it at your repositories, configure your AI provider credentials, and run it locally. Cloud and mobile experiences are under development but not yet generally available, so for now the workflow is desktop and local only. The open source nature means you can inspect exactly what Zuse does with your credentials and code, and contribute improvements if you want. For solo developers it is a way to parallelize work that would otherwise be sequential, and for teams it provides a shared surface where everyone can see which agents are working on which tasks and step in when needed.