Use cases
Build it six ways by lunch.
Run six parallel attempts at the same idea, or split one prototype into six parts. Keep what works, delete the rest, and lose nothing but tokens.
Two ways to fan out
Parallel attempts or parallel parts.
Six takes on one idea
Give the Conductor one idea and six directions, and each Claude Code session builds its own variant. Instead of arguing about the approach, you compare working versions.
Six parts of one build
Or split a single prototype: auth, schema, UI shell, API, tests, and docs land as separate tasks in the mission DAG, built simultaneously from one brief.
Pick the winner
Loom auto-detects running dev servers and opens them in an inline web preview, so you click through each attempt without leaving the app. The git graph shows what each one cost in code.
The brief
Name the variants, not the steps.
For a variants run, the brief is the list of directions. The Conductor plans one task per direction, dispatches them across the fleet, watches every terminal, and verifies each attempt actually runs before it counts as done.
# one idea, six variants
Prototype an onboarding flow for the app.
One approach per session: wizard, single page,
checklist, chat-style, video-first, blank canvas.
Each variant runs on its own port with seed data.Built for throwaways
Cheap experiments, fast loops.
Design first
For website prototypes, the Design tab works theme-first: it settles the look and the page plan before the fleet builds anything, so variants differ on substance, not accidents.
Preview inline
Every dev server the fleet starts is auto-detected and opened in a preview tab. Six attempts, six tabs, one window.
Nothing wasted but tokens
Loom is free and open source, and the fleet runs on your existing Claude login. An abandoned prototype costs only what your providers charged for it, which makes trying six things rational instead of indulgent.
Hand it the work.
Walk away.
macOS, Linux, and Windows. Around 13 MB. Free and open source.