The parts that make multi-agent work supervisable.

Eight capabilities, each solving a specific way that autonomous agents go wrong when nobody is watching. Open any one for the detail.

01Coordination

Agent teams

A team is a set of agents with named roles, a shared task list, and a shared repo. Give the team an objective and the lead breaks it into tasks, assigns them, and keeps the board moving. Every agent works in its own process, so five agents mean five things happening at once, not five things queued.

Named roles: lead, builder, reviewer, tester, docs
Agents run in parallel processes, not one queue
Delegation and hand-off between teams

Kanban board

The board is the interface. Tasks flow from backlog to done as agents claim, work, and finish them — no refresh, no polling. Drag a card to reprioritise, click one to read the full agent transcript, or pull something back to review if it looks wrong.

Five columns: backlog, assigned, in progress, review, done
Real-time status as agents claim and finish work
Drag to reprioritise or reassign mid-flight

Code review

Nothing lands because an agent said it was done. Every task produces a diff you can read in the app. Accept it, reject it, or leave a comment and send it back — the agent picks the comment up and revises. Reviewer agents can do the first pass so you only see what survived.

Per-task diff view, file by file
Accept, reject, or comment and send back
Reviewer agents do the first pass

Token analytics and budgets

Every run is metered. Break spend down by team, agent, task, project, model or individual run, watch the trend line, and set a monthly ceiling in tokens or currency. Alerts fire at 80% and again at 100%, and scheduled runs can hard-stop instead of quietly draining the budget overnight.

Breakdown by team, agent, task, project, model, run
Trend lines and month-end forecast
Monthly budget in tokens or currency

Organizations and global overview

Past two or three teams you need a floor plan. Organizations let you nest teams into any structure you like, and the global overview draws all of them at once — who is working, what is blocked, which task depends on which, and which teams are talking to each other right now.

Nest teams into departments, squads, any depth
One live map of every organization
Task progress and dependency lines

More on organizations and global overview

overview · all orgs

06Control

Terminal workspace

Agents run commands and so do you — in the same window. Open a shell scoped to a team's workspace or a plain local shell, keep persistent history across restarts, and use autocomplete that knows your project. No alt-tabbing to check whether the build actually passed.

Team-scoped and local shells side by side
Persistent history across restarts
Autocomplete aware of the project

Cross-team communication

Coordination happens in messages, not in your head. Agents send direct messages, comment on tasks, and link work across teams. Every exchange is logged and readable, so when something goes sideways you can see the sentence where it went wrong.

Direct agent-to-agent messages
Threaded comments on any task
Shared task links across teams

Solo mode

Not everything needs a committee. Solo mode gives you a single agent that writes its own task list, works through it, and reports back on the same board. When the job gets bigger, add roles and the solo agent becomes the lead of a real team — same project, same history.

One agent that plans its own task list
Same board, same review flow
Promote to a full team without losing history

Stop being the message bus.

Install it, open a repo, and give three agents an objective. Free, no account, no API key.

Download free Find your use case

macOS · Windows · Linux