Local runner · planned capability
Local runner
The intended execution architecture separates the control center — where you cast and review — from the runner, where code actually executes. This page describes that architecture and states plainly what exists today: nothing but the simulation.
The intended execution path
Planned
- 01DeveloperCasts the task and holds every approval gate.
- 02Forge control centerTracks the task, its plan, its state, and its artifacts.
- 03Policy / permission layerEnforces the gates — dependency changes, network, git operations — before anything reaches the runner.
- 04Runner (local, self-hosted, or managed)The process that actually executes: on your machine, your infrastructure, or ours.
- 05Isolated task workspaceA disposable copy of the relevant code, created per task, destroyed after review.
- 06Configured model providerReceives only the scoped context the policy layer allows, using the credentials you configure.
- 07Tests, verification, review artifactThe output: a verification report, self-review, and diff that flow back to the control center for your review.
Each stage hands to the next: developer → control center → policy layer → runner → isolated workspace → model provider → verification and review artifacts back to you.
What a local runner means
With a local runner, your code never leaves your machine except for the scoped model calls you explicitly configure — your keys, your chosen provider, your network. The Forge control center coordinates the task, but execution, the workspace, and the repository all stay on hardware you control. The policy layer runs alongside the runner, so permission gates hold even when the control center is remote.
Local vs. managed
| Aspect | Local / self-hosted | Managed |
|---|---|---|
| Where code lives | Your machine or infrastructure | Isolated workspaces we operate |
| Model credentials | Yours — you pick the provider | Configured per organization |
| What leaves your boundary | Only scoped model calls you configure | Task context, within the policy layer's limits |
| Operations burden | You run the runner | None |
Today
Available in prototype
No runner exists yet — local, self-hosted, or managed. The prototype simulates execution entirely in your browser with a deterministic trace; no code is executed and nothing leaves your machine at all. The runner is planned work, and the security properties it must satisfy are laid out in the security architecture.