Isolation should be automatic
Each task receives a dedicated branch and worktree, recorded from an exact base commit and kept separate until verification.
About Muxtra
More agents should mean
more finished work.
Keep the agent interfaces you already like. Lose the workspace juggling, crossed changes, and uncertainty about what is safe to combine.
The idea
Parallelism needs coordination.
Coding agents are individually capable, but real projects rarely divide themselves into perfectly separate files. Two agents can make correct changes in isolation and still produce a broken result together.
Muxtra exists to make that coordination understandable. It gives every task room to work, surfaces declared path overlap, verifies the committed result, and proves the final composition before it reaches your project.
What we believe
Coordination with clear boundaries.
Every task should have a clear owner, boundary, and path to verification.
Providers should coexist
Use Codex for one task, Claude Code for another, or hand a workspace to a remote client. The project—not the provider—owns the workflow.
Verification beats optimism
“The agent finished” is not the same as “the work is ready.” Muxtra records exact commits and reruns checks on the combined result.
Failure should be recoverable
Failed composition happens in temporary space. Task branches remain available. Uncommitted work is never silently discarded.
Open source
Built in the open.
Muxtra is an MIT-licensed CLI for coordinating coding agents across providers. The current release keeps each provider’s native interface intact while Muxtra owns isolation, role context, verification, and safe composition.