features
Everything your agents do, on one screen.
Six areas, one dashboard on 127.0.0.1:7777. Each area has its own page with an episode, short loops and screenshots.
From an idea to a plan your agents can run
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.
Explore planningEvery decision your agents wait on, in one place
A permission prompt in one terminal, a question in another. The Inbox puts every decision that waits on you in one list you work from the keyboard, and an agent clears the part that never needed you.
Explore inboxHow your agents are doing, and what to change
A retry here, a lint error there. Health reduces a date range to one sentence about the fleet and the agents that moved it, names the friction you can remove today, and turns patterns into recommendations with the evidence attached.
Explore healthEvery Claude Code session, across every project
The daemon reads the transcripts Claude Code already writes and indexes them as they land. A list that tells working from stuck, a detail view that replays the run as a conversation, and a rail with models, agents, files and cost. Nothing to install in your projects.
Explore sessionsWhat your agents learned, and whether to trust it
Forecasts sound confident, and miss. A run that lands far from its forecast becomes a lesson candidate with the evidence attached. Nothing becomes active without your accept, every active lesson is re-measured, and the local classifier only acts once it has earned it.
Explore learningWhat one project knows about itself
A session re-reads the repo and guesses the architecture. Knowledge keeps what one project knows about itself in one place: its memory files in one editor, an architecture map that says how stale it is, and code graphs agents can query instead of grepping.
Explore knowledgethe whole dashboard
Nine pages in the sidebar.
What each one is for, in one line.
What waits on you, what runs now, and this week's loop: plan, run, measure, learn, change.
Every decision in one queue, and it reaches your phone, so a waiting run is unblocked without going back to the desk.
Every transcript indexed as it is written, with the tool calls, files and errors of each run.
Every session blocked on a human, oldest first, with the reply ready for your clipboard.
The planner interview, the phase timeline with its dependencies, forecasts, runs and completion reports.
Agent scorecards, friction, estimates, an evidenced advisor, and cost against the live quota window.
Lessons from runs that missed their forecast, a local classifier, and how honest the forecasts are.
Memory files in one editor with a conflict guard, a stale-fact lint, and the architecture map.
Every agent, skill, hook and command on the machine, with origin and a content-addressed version history.
why it holds together
Six rules, each enforced by something you can run.
A core that names no one
Brand tokens are forbidden anywhere under plugins/, and a CI ratchet keeps the count at zero. Each project's flavor arrives at runtime from its own project.json.
Flow goes up only
A component is born project-local, moves to a domain pack when a second project needs it, and reaches the core only when every project does. Nothing is copied downward.
One worktree per run
Every phase run and every dispatched task gets its own branch in an isolated git worktree. Your working tree is never the agent's workspace.
Verdicts that can say no
A read-only verifier diffs the work against the persisted start point and returns PASS, FAIL or INCONCLUSIVE. Done means the criteria are ticked and the report is written, not that a process exited 0.
A phase graph, not a wish list
Every phase carries measurable acceptance criteria, a forecast and a completion report. Runs execute the graph in dependency order, and the markdown on disk stays the source of truth.
Everything on the record
Every session transcript is indexed into local SQLite: tokens, cost, approvals, forecasts against actuals, and the lessons drawn from the gap. Nothing leaves the machine.
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.