From an idea to a plan your agents can run
Agents write code fast. Planning makes them build what you meant.
Describe what you want in plain words. A planner reads your repo, interviews you one question at a time, and writes a phased plan with acceptance criteria and forecasts. The same page runs it, one phase at a time or the whole graph, each phase in its own git worktree.
An interview, not a form
The planner asks one question at a time on the left, while the plan as it stands sits on the right. Every option comes with pros, cons and a recommended pick. Pick several, or answer in your own words: a reply the wizard cannot parse turns into a conversation instead of a dead end.
- The running plan rebuilds after every answer
- Every answer and its reasoning is kept in the plan
- Refine, or tell the planner to proceed
Markdown stays the source of truth
A plan lands as plain docs in your private workspace: the goal, the decisions from the interview, and phases with their sizes and what each one depends on. Progress is the acceptance-criteria checkboxes inside them, and ticking one in the dashboard rewrites that exact line.
- Each phase is a markdown doc with acceptance criteria
- Pick a model per phase, run one phase or the whole plan
- A finished phase writes a report of what shipped
Forecast first, then measure
Before anything runs, the planner forecasts each phase's size, duration, outcome and the areas it will touch. Afterwards the actuals are scored against the forecast. When a phase fails, the planner proposes a fix, and you review the revision as a diff before anything changes.
- See where a phase went off plan
- Revisions are staged proposals until you apply them
- A run that lands far from the plan becomes a lesson candidate
screenshots
The planning pages, up close.
Click any screenshot to open it full size. The projects are fictional.
Run your agents like a fleet.
Free on GitHub. One binary on your machine, no cloud, no account. Your sessions show up the moment it starts.