Environment state management

ACP computers persist files, packages, secrets, runtime choices, and optional GUI state across work. Environment state management is about deciding when to reuse, checkpoint, fork, or isolate that state.

Treat computers as durable environments with forks, snapshots, and reusable runtime state.

Reuse state intentionally

A persistent computer is useful because it carries state forward. Reusing that state lets threads build on installed dependencies, cloned repositories, and prior outputs instead of redoing setup every time.

Reuse a computer when the same codebase or workspace should stay live.

Keep setup work out of the hot path for repeated execution.

Match the lifetime of a computer to the lifetime of the work it supports.

Fork and checkpoint safely

Snapshots and forks make it possible to branch environment state just like code. ACP makes this a product primitive so you can experiment or recover without corrupting the original working environment.

Checkpoint before risky changes, migrations, or refactors.

Fork a computer when a thread should branch in a different direction.

Use snapshots as recovery points and operational history.

Govern what the environment can do

Environment state management also means deciding what a computer is allowed to contain or access. ACP exposes runtime versions, Dockerfile extensions, secrets, internet access, packages, and MCP servers as explicit configuration surfaces.

Use explicit runtime config rather than implicit setup inside prompts.

Keep secrets and internet permissions visible and reviewable.

Model computers as governed execution assets, not hidden sandboxes.