A tour of the system: who talks to what, where your data lives, and why one tenant can never reach another. Click any box for a plain-language explanation.
You drive everything from one place: the arc command line, a chat message, a Telegram topic, or the web overview.
All traffic enters through a hardened edge over encrypted transport, with mutual TLS to the core. The backend is not exposed to the public internet.
The core checks that you are allowed to act on a project and coordinates the AI team. It is the single place where access is enforced, so one tenant can never reach another tenant's data.
Your tasks are carried out by AI workers, a dedicated set per project, acting within that project's boundary.
Workers run on Claude using your own subscription or API key. Your model access stays yours.
Each project's data lives in a single source of truth, isolated per tenant. Semantic search runs inside the same secure boundary.
Keys and secrets are kept encrypted at rest. Workers get only the minimum they need.
On the cloud plan, each user runs in their own isolated workspace on a separate host, with no access to the database or secrets.
Backend isolated from the public internet. Access checked on every request, per tenant. Secrets encrypted at rest. Your own model keys, never ours.