Lab
Everything we have built, with its status
The first two are the pipeline itself; they run underneath every job. Ideas are tested by building them, and the status says how far each one has got.
Decision Bus
Sits between the AI agents that build our work and anything they try to do. Checks every action against the rules, holds the ones that need a person, and records every decision in a log that cannot be edited.
In productionProject Layer
Work is planned before any AI starts on it: the steps, and what finished looks like. A person reads that plan and approves it. Once approved it is locked — change it and it goes back for approval, and nothing outside it can be touched. Every approval, change and refusal is added to the record, so what was meant to happen sits beside what did.
In productionaXstudio
One accessible interface for every aXai product: order, review, approve and pay, one question at a time, for people whose disability includes cognitive impairment.
AlphaAvatar assistant
Our own talking avatar, built and run on our own hardware, so the face, the voice and the words stay ours to set. The model that writes the words runs in Sydney. Everything it says is on screen as well as spoken, Easy Read is one tap away, and it can send Annette a message for you. Axel AI Assisted runs on it.
LiveSavanax
Multi-perspective reasoning engine that maps disagreement rather than forcing consensus. Twenty-three reasoning perspectives, each its own service with its own source material.
LiveAxelV1
Autonomy-preserving decision support.
PilotContext Scanner
Codebase documentation generator. Deterministic parsing, not generation: its whole dependency list is a TypeScript parser.
In productionDocument Layer
A central store for the documents a piece of work depends on. Every change is sealed onto the one before it, and the state is a replay of that log rather than a copy that can drift.
PrototypeTrading
An automated trading system on racing exchange markets: it watches live prices, decides, places small orders, and records every one — 114,005 of them over 29 days. It includes machine learning, and it is the one place where a model’s claim is settled by the market within minutes. The page shows what the record says, in the numbers it was measured in.
In productionPSight
Nervous-system regulation from heart-rate variability and EEG.
ParkedErsher
Specified: give it a website address and get back a containerised, runnable version of that site. It starts once the Decision Bus and the Project Layer are hardened, because it runs on both.
In developmentMCP connectivity
Planned: letting an organisation’s other AI tools hand their decisions and documents to the same record, through the Model Context Protocol.
PlannedMoontrade
A currency-trading prototype that asked whether lunar phase, geomagnetic activity and weather carry any signal a market has not already priced. Built properly, on 319,336 rows, so the answer would be worth having. The answer is no — and the page keeps the numbers that say so.
Parked
Live: in use on real work. Pilot: in use with one partner. In production: used inside aXai. Alpha: built, not yet used with a customer. Prototype: working, and shown on synthetic material. In development: specified, and being built. Planned: specified, and next in line. Parked: paused on purpose.