How It Works
No black boxes. Every task, every model call, every decision is routed, governed, and recorded.
The execution path
- Task enters — via the Control Centre, an agent mission, or a website request.
- AI Fabric — the single entry point applies governance and policy.
- AI Policy — classifies the task (coding, vision, offline, auto) and selects the route.
- OmniRoute — picks the model from 874 across 19 namespaces, with health checks and fallbacks.
- Execution — the model runs; the result is captured with provider, model, tokens, and audit ID.
- Audit — the execution is appended to the audit trail, visible in the Control Centre.
01 Task →
02 AI Fabric →
03 AI Policy →
04 OmniRoute →
05 Model →
06 Audit record
Governance model
Approval lifecycle
Outbound actions follow a documented lifecycle: PENDING → APPROVED → EXECUTING → EXECUTED, with REVIEW_REQUIRED and CANCELLED paths. No other transitions exist.
Outbound mode
The system runs in FROZEN, PILOT, or PRODUCTION mode. Today it is in PILOT: outbound is limited, governed, and human-approved.
Do-Not-Contact
A first-class DNC register is checked on every outbound request. Consent is tracked per contact. We never auto-contact without permission.
See it in action
Book a discovery call for a live demonstration of the execution path and audit trail.
Book a Discovery