Governance software for autonomous robots

Robot Expert System

AI that Answers to You.

The Robot Expert System (RES) is an embodied Trust Layer for Autonomous Systems — a programmable conscience that runs inside the agent, with a safety floor that cannot be overridden and a tamper-evident record of every decision it makes.

The Laws, with the Owner's rules above them, govern what a robot may do — in an order no one can change.

Start here What are The Laws? Honest status

Unplug everything.Nothing changes.

A conscience that depends on a network is not a conscience. The judgment, the record, and the refusals live inside the robot. The cloud is a courier, never a crutch.

See it proven →

network · cloud · certifier

inside the robot

#41 · unlock the door — delivery

pass — Owner rule satisfied

#42 · record the visitor's face

refused — bystander — stays out of the record

#43 · fetch the 8 am medication

pass — within the Laws

Illustrative demo with simulated decisions; every scenario line is owner-approved copy. The loop never reads the network state: cadence and content are identical either way. That is the claim.

What it is

An autonomous agent makes consequential decisions in the world. Its capability can be verified on a bench; its conduct cannot. The tools that try to govern conduct today sit outside the agent, in the cloud, where they can be switched off, bypassed, or simply lost with the connection.

The RES is not ordinary robot software. It is the robot's conscience: the trust-and-record layer that runs on the robot itself.

Offline and durable. It runs with no network and operates the same on day one and day one thousand.
Unbypassable below the floor. The owner defines the agent's limits and can make them stricter at any time, but no one — including the owner — can lower the immutable safety floor. There is no in-the-moment override of a refusal.
Self-recording. Every decision is written to a tamper-evident, append-only record that travels with the agent and can be independently audited later.
Reproducible. A decision can be re-derived and checked, not merely asserted.

What it is Not

The RES is not a

Capability or performance layer
Replacement for a maker's autonomy stack
Guarantee that any particular decision was wise

It governs and records; it does not make the agent smarter.

Isaac Asimov spent a career showing that unexamined laws fail. This catalog is what examining them looks like.

Current maturity and current claims live on the Status page.

Who checks this?

The RES submits to an independent certifier it does not control: GRRC. The separation is enforced cryptographically, so any attempt to self-clear is detectable. Any certificate can be checked from outside, by anyone.

Verify it yourself at grrc.ai →

Karel

Hard questions. Straight answers.

The ones a skeptic would ask, answered without spin.

“Anything can be hacked. Why believe this?”

We never say unhackable. The claim is narrower and checkable: the record cannot be altered silently. Tampering shows as a broken chain to anyone who verifies it.

“Is this real today?”

Yes, at early maturity: one implementation, in owner use, no shipped robot. The Status page says exactly what is true now and what is planned, and it stays honest as that changes.

“Who does it really answer to: you, or me?”

You. The Owner writes the rules and can tighten them at any time. Beneath them sits a safety floor no one can lower: not the maker, not us, not even you.

Written answers, approved by the owner of this site. Nothing here is a live agent.

New here? Start with the plain-language walkthrough.

Start here