Petrixia. In development

Platform

One intelligence, multiple robots.

Petrixia separates the robot's intelligence from its particular hardware. That allows new robot bodies, sensors and capabilities to be introduced without rebuilding the whole system.

Components

What the platform is made of.

Agent and conversation system

Turns what a person said into what should happen next, and says something useful back.

Memory

Durable, attributable recall: what was learned, from whom, when, and how sure the system is.

Procedure engine

Long-running tasks that survive interruption — pause, resume, re-plan, or hand back.

World model

Rooms, objects, people and their relationships, kept current from what the body actually perceives.

Permissions and safety

The boundary the rest of the platform cannot argue with: leases, permission checks, audit.

Robot and tool interface

One hardware-independent contract that a simulated body and a physical body both implement.

Simulation

Petrixia Town: a full world for developing and scoring behaviour before it reaches hardware.

Fleet and organisation services

Multiple sites, robots, staff and residents under separate administration.

The authority boundary

The world keeps the truth. The agent keeps its mind.

Most agent demonstrations let the model both act and report on its own success. Petrixia is built the other way round: an authoritative world holds ground truth, and the agent reaches it only through the same narrow gateway a human operator uses.

The authority boundary A human operator and a Petrixia instance both send commands through a single actor gateway into the authoritative world, which owns the map, bodies, objects and permissions. CONTROLLERS Human operator direct control & override Petrixia conversation · memory · plans SINGLE BOUNDARY PACT actor gateway identity & body leases permission checks deterministic parsing every action audited AUTHORITATIVE WORLD Ground truth map, world time, bodies objects, inventories, relations movement & collision timed world processes results the agent cannot edit Perception is radius-limited: Petrixia receives actor-relative observations only, and may hold a private map that is incomplete, stale or wrong. The world map remains ground truth and cannot be overwritten by the agent.
A keypress from a human operator and a decision from Petrixia produce the same command, pass the same permission checks, and get the same honest answer back — including blocked and partial.

One protocol for both

Every action — a person pressing a key, or Petrixia deciding to fill a cup — becomes the same structured command. Behaviour observed in testing is behaviour produced by the real interface, not by a special path built for the demonstration.

Actions become true by happening

A request does not succeed because the agent believes it did. Turning on a tap starts a timed world process; the cup fills; events report the level and, eventually, that it is full. Belief and outcome stay separable.

A task, as the world receives it

# the same commands a human operator's actions produce
cup: take
cup: place under tap
tap: turn on
world: wait until cup is full, timeout 20 seconds
tap: turn off
cup: take

Permissions and safety

Safety by boundary, not by promise.

A body is leased, never owned

An agent never owns a robot body. It holds a temporary, revocable lease on a body the world owns, and the lease can be withdrawn at any moment.

Permission sits outside the agent

Permission checks, collision and world rules live in the gateway and the world — where the agent cannot argue with them, reinterpret them, or route around them.

Everything is on the record

Commands are idempotent and audited; every action writes an ordered event. What was intended and what actually happened can always be compared afterwards.