From the source
Amp now lets teams run orbs on their own compute, with the company's orb provider dispatching spawn, pause and delete requests to a program the customer supplies.
- The setup requires infrastructure that can start machines quickly, sleep and resume them just as fast, keep them alive for long periods, and scale to zero so idle orbs cost nothing.
- Users run `amp orb provider` inside their own environment; it connects out to Amp and waits for orb operations, then hands each one to a local script.
- Amp publishes example dispatchers for Docker, E2B,
Apple's container tool, Modal and
Vercel Sandbox, each mapping the three operations onto that platform's own primitives.
- Modal, for instance, has no pause, so the sample snapshots the filesystem and resumes from the last snapshot, while the Vercel example copies Amp's image and keeps one snapshot.
