Platform Guide

Governance & security

Permissions, budgets, approvals, and audit are properties of the platform, not conventions the application layer is trusted to follow.

Permissions

Role-based permission scopes apply on every route, with resource ACLs beneath them. Access is evaluated at the point of retrieval or dispatch, so an identity without access never reaches the resource — as opposed to reaching it and having the result filtered afterward.

Budgets, limits, and approval gates

Budgets, usage limits, and approval gates wrap every AI operation. Ceilings are enforced at execution: an agent that would exceed its budget stops and escalates rather than continuing and billing. Sensitive or irreversible actions pause for one or more human approvals before they run.

Audit and attribution

Structured logs, OpenTelemetry traces, and per-call model attribution — tenant, initiator, purpose, tokens, cost — make every answer explainable after the fact. The record is generated by the running system, which is what makes it usable as evidence rather than as a reconstruction.