Capabilities
Your goal becomes a graph.
Type one goal. The Conductor plans it into a DAG of tasks, dispatches them across six Claude Code sessions, and shows you what every terminal is doing, live.
How a mission runs
One brief in, a mission out.
Plan
The Conductor, a reasoning model you bring your own key for, breaks the goal into discrete tasks and maps the dependencies between them. Independent tasks can run at the same time. Dependent ones wait their turn.
Dispatch
Tasks fan out across the fleet, six real Claude Code CLI sessions running on your existing Claude login. The DAG decides what each session picks up next, so six terminals stay busy without stepping on each other.
Verify
Nothing counts as done on a session's word alone. Work is re-checked before it counts, finished slots are detected and marked, and a premature done gets caught. See verification.
Live activity strips
Read all six at a glance.
A mission you cannot see is a mission you cannot trust. The mission view pairs the DAG with a live activity strip for every terminal, so the state of the whole fleet fits in one look.
A strip per terminal
Each session gets a live strip showing what it is doing right now, so you scan the fleet in seconds instead of cycling through six panes.
Drop into any session
The strips are summaries, not walls. Every terminal is the real Claude Code TUI, and you can watch it scroll or type into it whenever you want a closer look.
Recovery is part of the run
The Conductor watches every terminal, answers permission prompts, recovers stalls and rate limits, and re-sends tasks after a session comes back. See smart steering.
Questions
Missions, answered.
What happens when one task depends on another?
The DAG holds the dependent task until the work it needs has finished and been verified, then dispatches it to a free session. You never have to sequence the work by hand.
Can I step in mid-mission?
Yes. Each of the six terminals is a real Claude Code session, so you can type into any of them at any point. Shift+tab cycles that terminal's permission mode when you want tighter control. See permission modes.
What marks a task as done?
Verification, not self-report. Loom re-checks the work before it counts, and the mission only advances on results that hold up. When it is over, review everything in the source control panel before you push.
Hand it the work.
Walk away.
macOS, Linux, and Windows. Around 13 MB. Free and open source.