Skip to content

What Forkstage is

Forkstage keeps a warm, frozen copy of your whole stack: the application, its database with data in it, the queue, the workers, every service your docker-compose.yml or Dockerfile describes. Whenever a person or a coding agent needs somewhere real to run code, Forkstage forks that copy into a private microVM in about 100 ms and gives it a public address. The fork is used, then destroyed. What remains is a proof: a record of what ran inside it.

Coding agents change code faster than anyone can check it. To know whether a change works, the code has to run against the real shape of the system: migrations applied, seed rows present, the other services answering. The two usual places to do that both fall short:

  • Shared staging is one environment for many people. Someone’s migration breaks someone else’s test; resetting it takes a ticket; nobody fully trusts what it shows.
  • Local Docker starts in minutes, is not production-shaped, lives on one laptop, and an agent running in the cloud cannot reach it.

So the agent guesses, and the reviewer guesses after it.

  1. Bake once. forkstage bake sends your repository to a Forkstage host, builds and starts every service, runs migrations and seeds, waits until the app answers, and snapshots the running microVM: memory and disk. About a minute, repeated only when the setup changes.
  2. Fork many. forkstage up restores that snapshot into a fresh microVM with its own copy-on-write disk and its own network. Ready in about 100 ms. Each fork is complete and private: its own database, its own queue, its own address https://<id>.forkstage.run.
  3. Strike when done. forkstage down destroys the fork. So does the idle stop, the hard stop, and the pull request closing. The proof stays for 30 days.

Engineering teams that ship with coding agents and want every change, human or agent, to be exercised against a real stack before anyone reads the diff. Typical uses:

  • An agent reproduces a bug against real seed data, fixes it, runs the tests, and the proof lands with the pull request.
  • Every pull request gets a clickable copy for the reviewer, with tests already run.
  • A developer wants a throwaway database with the right schema and rows, right now, without touching anyone else’s.
  • Not a hosting platform. Forks are temporary by design and stop by themselves.
  • Not a replacement for CI. It runs your tests inside a real copy and records them, but it does not gate merges (yet).
  • Not a place for real customer data. Bake with seed data; anyone with a fork’s link can open it while it runs.

The control plane (API and dashboard) runs in Amsterdam. Hosts that run forks are bare-metal machines in Frankfurt with Firecracker microVMs. The CLI runs on your machine and talks to the control plane over HTTPS. See Security and data.