Code Factory
One chat, one branch, one machine, one thing at a time. That’s not a workflow — it’s a bottleneck with a cursor in it. Write the spec once, fork your environment into a dozen identical stations, and let a fleet of agents work the floor in parallel. Review from the desktop app. Or your phone.
The model stopped being the constraint
Agents can build almost anything you can describe. The hard part isn’t the model anymore — it’s how many you can run, watch, and merge before you lose the thread. The default workflow is artisanal: one attempt at a time, hand-tended.
A factory needs identical stations and a foreman who can see the whole floor. That’s an infrastructure problem, not a better prompt.
Spec in, parallel work out
Write the spec
Describe the outcome and the constraints once. The spec is the contract every agent on the floor builds against.
Fork the floor
Snapshot a known-good Agent Computer and fork it into identical stations — one per task, attempt, or variant.
Run in parallel
Agents work at the same time on isolated machines. No shared state, no two of them editing the same line.
Review & merge
Watch results land in the desktop app, compare attempts, keep the winner, drop the rest in one move.
# Snapshot a known-good machine as the factory template heyvm fork main-env --name spec-142-a heyvm fork main-env --name spec-142-b heyvm fork main-env --name spec-142-c # Point an agent at each station against the same spec heyvm connect spec-142-a --agent claude-code heyvm connect spec-142-b --agent claude-code heyvm connect spec-142-c --agent opencode # Review every attempt side by side in the desktop app
Hosted playground vs. your factory floor
Replit, Lovable, Vercel own distribution: your project lives in their cloud, shaped by their runtime, shipped through their pipe. Great for one app. A factory needs the opposite — your machine, your stack, and the freedom to fork it as wide as the work gets.
A hosted playground
- One project, one hosted environment
- Their cloud owns distribution and runtime
- Parallelism capped by the product
- Ships to their cloud, on their terms
A heyo code factory
- Fork one machine into a fleet of stations
- Your hardware, your stack, your tools
- Parallelism bound only by your hardware
- Ships to infrastructure you already run
Turn one machine into the factory floor
Spec-driven, parallel, on infrastructure you control end to end.