Skip to content
agentvfs ★ GitHub

Agent orchestrators running many tasks against one workspace

Per-task isolation with rollback

Fork a vault per task and checkpoint before risky steps, so a bad execution rolls back instead of corrupting the workspace an agent depends on.

The problem

When an agent runs many tasks in sequence, one bad mutation can poison the workspace for everything after it, and you cannot always tell which command did the damage.

How agentvfs helps

Fork a fresh vault per task for isolation, and use allow_with_checkpoint so risky commands save a rollback point automatically. If a step goes wrong, checkpoint restore snaps the workspace back.

What that looks like in practice

  • avfs vault fork gives each task its own millisecond-cloned workspace.
  • Policy can auto-checkpoint before risky commands, so a rollback point always exists.
  • checkpoint restore recovers cleanly without re-running earlier work.
  • ChangeSummary tells you exactly which files each task touched.

Try it on this workflow

See the architecture, more use cases, or the guides.