TRUST & OPERATIONS Live

AI, policy and physical control

The governing boundary for Oddessy demos, APIs, agents and industrial integrations.

Last updated 27 July 2026 · Documentation version 2026.07

Authoritative statement

Safety-rated controls remain authoritative

Oddessy perception and agent outputs are probabilistic operational inputs. They do not replace safety PLCs, relays, emergency stops, interlocks, qualified operators or site commissioning.

Oddessy safety flow from evidence through a verified model, policy and draft recommendation to accountable human approval and authoritative industrial controls — Oddessy produces bounded operational inputs; the commissioned site-control path remains authoritative.

What Oddessy may do

Observe approved image, telemetry and system context.
Return structured classifications, explanations and provenance.
Draft or request a bounded action under a configured policy.
Manage approved edge software and report acknowledgements.

What Oddessy does not claim

That an AI response proves a robot, motor, valve or PLC changed state.
That a public demo is connected to production equipment.
That the QCS6490 reference controller is a safety PLC.
That MCP or Argo can override an interlock, approval or emergency stop.

Production responsibility

A production deployment needs site-specific hazard analysis, network segmentation, least-privilege identities, retention and privacy controls, tested fail-safe behaviour, model validation and functional-safety engineering.