Task success
Did the agent finish the task it was given, to the bar you set?
example 41 / 50 tasks
Pre-launch concept
One control layer for every coding agent you already run.
Muxboard is a desktop app that switches harnesses, carries task memory between models, and routes planning to Claude while cheaper models handle routine edits.
- Harness switching
- Cross-model task memory
- Planner / editor routing
- Tool-call review
- Weekly cost & quality report
How it works
Four steps, one desktop. Muxboard sits above the agents you already use instead of replacing them.
-
01
Route
Pick a task and Muxboard sends it to the right harness. Switch harnesses mid-task without losing context, because task memory moves with the work, not with the model.
-
02
Plan
The Claude API breaks the task into concrete steps. Planning goes to Claude, and routine edits go to cheaper models.
-
03
Review
Claude reviews tool calls as they happen, so risky or off-plan actions get flagged before they land and a human can step in when needed.
-
04
Report
Every run leaves a trace. Each week, Claude turns those traces into a cost and quality report you can actually act on.
A look at the desktop
Illustrative mockup, not a screenshot of a shipping product. All names and numbers below are placeholders.
Plan (by Claude)
- ✓Map current middleware call sitesplanner
- ✓Extract token check into helpereditor
- Update 6 route handlerseditor
- Add tests for expired-token patheditor
- Final diff reviewplanner
$ mux route --task refactor-auth --harness B
→ memory synced · planner: claude · editor: routine-tier
$ mux review --live
✔ write src/auth/token.ts approved
⚠ shell `rm -rf build/` held for human
Three numbers that matter
Muxboard measures outcomes, not tokens. These are the metrics it tracks. Values shown are labeled examples only.
Human escalation
How often a run needs you: held tool calls, stuck plans, handoffs.
example 7 of 50 runs
Dollars per merged change
Total model spend divided by changes that actually merged.
example $X.XX / merge
Status: pre-launch
Muxboard is in internal testing. Credits currently cover the planner and the eval loop, roughly 20 to 50 internal runs a day. No public build, pricing, or signup flow yet.
This page has no backend and collects no data. The waitlist form below is a placeholder: nothing you type is sent or stored.
Want early access?
Tell us which coding agents you run today and what you'd want Muxboard to route.