External Agent Access and Integration Control
Status: Reserved / In development. Desktop currently provides a status projection under Advanced & Diagnostics → Permissions & Access. Token issuance and external execution remain disabled until the Runtime action registry and external gateway infrastructure are enabled.
External Agent Access is designed to let third-party tools—such as IDE extensions, external AI assistants, and automation workflows—interact with Nimi through scoped, time-bounded access tokens.
For the platform-level specification and security model, see Platform → External Agents.
Current Desktop Behavior
| Control Dimension | Current Behavior | Purpose |
|---|---|---|
| Service Status | Displays Runtime's external gateway readiness | Shows whether the local gateway service is reachable |
| Action Registry | Keeps token issuance disabled while the action registry is empty (EXTERNAL_AGENT_ACTION_REGISTRY_EMPTY) | Fails closed so tokens cannot be issued before permitted actions are registered |
| Token Ledger | Shows current token projection state and why issuance is unavailable | Explains why external access cannot yet be granted |
| Revocation Contract | Defines the revocation interface in the specification as a fail-closed control | Reserves immediate revocation support before token issuance opens |
Planned Security Boundaries
When external token issuance and gateway execution are enabled, the protocol requires three core safeguards:
- Least Privilege: Each token is restricted to explicit actions and data scopes.
- Bounded Local Audit: External calls, parameters, and authorization decisions are recorded in sanitized, size-bounded local audit events.
- Fail-Closed Gateway: Operations not present in the admitted action registry are rejected at the gateway.