Docs
Filesystem tunnels
Two ownership rules: a project-mirror the AI owns, and a requirements-feed you own.
Two tunnels, two owners
Cockpit keeps the Project VM and the local desktop in sync with two one-way tunnels. Direction and ownership are the product rules — not a shared scratch disk.
Project-mirror (VM → local, AI-owned)
The Project VM is the source of truth for application code. A project-mirror copies that tree to the local machine so the desktop AI can read and edit it. Treat the mirrored copy as AI-owned: Agent mode is allowed to change it. Do not treat it as the place you drop regulated source-of-record files.
Requirements-feed (local → VM, user-owned)
The requirements-feed copies local briefs, specs, and tickets onto the VM so Ask / Plan / Agent can see them. You own that feed. The agent should not overwrite it. Use it for human-authored requirements, not generated Mule code.
- project-mirror: VM → local. AI-owned working tree for the Mule project.
- requirements-feed: local → VM. User-owned input the model may read.
- Do not invert the directions. A local AI-owned mirror is not a backup of your laptop documents.
Why this matters on regulated benches
Banking, government, and enterprise programs need a clear line between generated code and user-authored requirements. Keeping those trees separate avoids mixing an AI working copy with evidence, policy, or client-owned specifications.