Choose your lanes
Set a design model, a code model, or use one provider for both.
Many agents.
One codebase.
Choose one model for design and frontend work and another for backend and logic. Muxtra gives each CLI an isolated copy of your project, then brings the verified work together.
Claude Code
Design lane
Codex
Code lane
More CLIs
Coming next
muxtra — zsh
$ muxtra team "Build account settings"
…
✓ Design + code team ready
Open two terminals and run:
muxtra launch build-account-settings-design
muxtra launch build-account-settings-code
Primary checkout unchanged
2 isolated workspaces
Built to work across agents
Claude Code
Codex
More coming
Claude Code
Codex
More coming
Isolation by default
Same project.
Separate worlds.
Every task gets its own branch and worktree automatically. Agents can edit, run, and test simultaneously without overwriting the files another agent is using.
- 01 One isolated workspace per task
- 02 Overlapping claims surfaced early
- 03 Your primary project stays untouched
Checkout redesign
agent/claude/checkout-redesign
Flaky tests
agent/codex/fix-flaky-tests
Search filters
agent/codex/search-filters
One verified result
All checks passed
The workflow
Describe the work.
Muxtra handles the machinery.
Pick the right provider and model for each responsibility. Muxtra handles the worktrees, shared context, verification, and safe composition from the terminal.
Work in native CLIs
Keep every provider’s full TUI, reasoning, tool calls, approvals, and conversation.
Combine safely
Muxtra checks the combined result before your primary project advances.
muxtra claim
$ muxtra claim "src/settings/**"
Claimed write: src/settings/**
Agent: claude (account-settings-design)
OVERLAP — another agent has claimed paths that intersect yours:
codex (account-settings-code) write src/settings/schema.ts
Claims do not block. Coordinate, narrow your paths, or ask the human.When paths cross
Nothing gets quietly overwritten.
Isolation prevents accidental live edits. Claims make overlapping intent visible, and guarded composition catches conflicts that only appear when finished work meets.
Another task has claimed an overlapping path
Muxtra shows which declared paths intersect.
One command, two focused lanes
Native agent CLIs.
Muxtra coordination.
Prompt, approve tools, and follow reasoning in each agent's complete native interface. Use Muxtra to coordinate their workspaces, checks, conflicts, and final result.
muxtra — zsh
$ muxtra setup
Committed .muxtra/project.yaml as "Configure Muxtra".
✓ Muxtra is ready.
$ muxtra team "Build the checkout"
…
✓ Design + code team ready
Open two terminals and run:
muxtra launch build-the-checkout-design
muxtra launch build-the-checkout-code
Ready for both native CLIs
Task branches retained
Built around the work
More progress.
Less orchestration.
Provider independent
Use the coding agents you prefer instead of committing your workflow to one vendor.
Task-first isolation
Every workspace has a recorded base, provider, task, and path to verification.
Verified before combined
Repository-owned checks decide when finished work is ready to come together.
Common questions
Good to know.
What happens to my primary checkout?
Agent work stays in isolated task workspaces. Muxtra only advances the primary checkout after the selected tasks combine cleanly and every configured check passes.
Can agents from different providers work together?
Yes. Muxtra is designed around the repository, not a single model provider. Claude Code and Codex CLI are supported today, with more adapters planned.
Can two agents edit the same file?
They can work independently without overwriting each other. Muxtra surfaces overlapping claims and stops unsafe combinations when Git or your checks find a conflict.
Is Muxtra open source?
Yes. The Muxtra CLI is available on GitHub under the MIT License.