fork() for machines

Branch live VMsinstantly.

Virtual machines with git-like semantics. Fork a live VM: memory, processes, sockets, and all in milliseconds. Run a thousand agent trajectories in parallel. Roll back a bad deploy in one call.

Used by agent swarms, CI preview environments, and teams who ship on Friday afternoons.
/ how it works

Three primitives.
Branchable, committable, resumable VMs.

01branch

Branch

Clone a running VM to take it in a different direction. Do it a hundred times. The parent keeps running.

$ vers branch vm-abc123
→ vm-9f2e77
02commit

Commit

Freeze a running VM to a permanent commit. Tag stable states, ship known-good builds, and continue moving fast in new branches.

$ vers commit vm-abc123 --tag stable
→ c1a2b3c4
03restore

Restore

Boot a VM from any commit. Reproduce a bug from the exact state it shipped in, spin up a clean demo, onboard a new engineer without setup scripts, or roll back a bad deploy in one call. Any commit, any time, the full running state.

$ vers restore c1a2b3c4
→ vm-d41c09 booted

Simple Syntax
Familiar Commands

~/vers
$vers branch

# Cloning current moment...

 

> Files [copied]

> Processes [carried]

> Memory [mapped ]

 

Branched vm-abc123 → vm-9f2e77 in 0.586ms

/ branch

Fork a running VM in less time than it takes to sip your coffee. Memory, processes, and sockets are carried into the child.

memory · processes · sockets · all carried
/ mental model

It's git, for running machines.

A root VM is a main branch. Every branch is a living fork. Every commit is a restorable snapshot. Walk the tree, prune dead experiments, roll back production in one call.

  • root: base image
  • branch: live fork
  • commit: content-addressable snapshot
  • restored: booted from a commit
A tree of live virtual machines: one root VM, three branches (two agent trajectories and a CI preview), each with its own commits and sub-branches, and one restored rollback.ROOTvm-abc123base · ubuntu-24.04BRANCHvm-9f2e77agent_1 · path ABRANCHvm-55dd01agent_2 · path BBRANCHvm-ff22c8ci · preview-#482COMMITc1a2b3tag: stableBRANCHvm-1e04rolled backCOMMITd7f09apre-deploy snapshotRESTOREDfrom c1a2b3one-call rollback
/ use cases

Built for teams that branch compute,
not just code.

/ parallel

Agent swarms

Fork one VM into a thousand. Let agents explore in parallel, keep the winners, discard the rest.

/ ci+cd

CI preview environments

Every PR gets a perfect copy of production. Branch at HEAD, boot an ephemeral deployment, tear it down on merge.

/ reliability

Bad-deploy rollback

Commit before risky migrations. If the deploy blows up, restore the commit: live memory and all.

/ durability

Long-running LLM tasks

Checkpoint long chains at every tool call. Resume from the last known-good state on crash.

/ faq

Questions we actually get.

Most sandbox platforms give you ephemeral VMs: boot, run, throw away. Vers gives you a tree of live machines you can branch, commit, and restore: memory, processes, sockets and all. Branches fork in milliseconds with no cold boot, so you're not paying for warm pools sitting idle. Different primitive, different jobs.