What we have built
These are the systems we use in our own work and run for the people who visit this site. Each one is listed with its status.
Each one is evidence of what we can build for you through our services. How a client project runs and the videos we have delivered follow the list.
Innovations by readiness
Savanax / Dialectica
Multi-perspective reasoning that keeps the disagreement
- Hard questions in policy, care and compliance rarely have one right answer. Most AI tools average the viewpoints away and hand back a single confident paragraph.
- Our own reasoning engine, deployed in the cloud. It runs a question through more than 20 reasoning perspectives, each grounded in its own source material, then maps where they agree and where they pull apart. Behind it sits a corpus of Australian disability legislation, guidance and decisions.
Read more about Savanax / DialecticaDecision-support and knowledge tools built on the same approach, scoped to your own documents and questions.
Through AI and software systems.
Context Scanner
Plain summaries of a codebase, written for AI assistants
- AI coding assistants start every session knowing nothing about your project: its structure, its conventions, or the rules it has to meet. Teams re-explain all of it, every time.
- A command-line tool that reads a codebase and writes a set of plain summaries an assistant can load at the start of a session, including checks for gaps against Australian accessibility, privacy, security and AI-governance expectations. Tested, and used daily across our own repositories.
Read more about Context ScannerA scan of your codebase with a written summary of what it found, or the tool set up inside your own team.
Through AI and software systems and AI capability and Vybrid sessions.
Decision Bus
One inbox for every decision an AI agent wants a person to make
- AI coding agents can do a great deal of work unattended. The risk is not that they do too little; it is that nobody can say afterwards who approved what, or when.
- The system we use to govern our own AI-built work. Every action an agent wants to take that needs a human call lands in one inbox, where a person approves or refuses it from their phone. Every decision is recorded. A large test suite keeps it honest.
Read more about Decision BusThe same governed way of working, applied to the AI and software projects we deliver for you.
Through AI and software systems.
Project Layer
Where operators plan and approve the work the Bus carries out
- A decision inbox on its own answers "may this happen?" but not "what are we trying to do, and how far along is it?" Operators need the plan and the approvals in one place.
- The console that sits alongside the Decision Bus. Operators set out the work, break it into pieces AI agents can carry, approve each step and see what has been done. In production for our own delivery.
Read more about Project LayerPlanned, visible and approved AI work on the projects we build with you.
Through AI and software systems.
AxelV1
Autonomy-preserving decision support for people with large support teams
- A person with a large support team generates a constant stream of messages, reminders, observations and roster changes. When the team carries all of it by hand, the person at the centre is the last to be asked and the first to be overruled.
- An event-driven support system that routes everyday events through policy-based automation while the person stays in charge: the system extends the judgement of the person and their team, every automated decision can be overridden, and anything new or complex escalates to a human. Built on Savanax for the hard calls. In production.
Read more about AxelV1Decision-support systems for care, support and case-management teams, designed around the preferences of the person at the centre.
Through AI for NDIS providers and Savanax.
Axel avatar and aXread
A talking avatar and an Easy Read assistant, running on this site
- Websites are dense. Many people would rather ask a question and hear the answer than read a page, and many need the answer in plain, short sentences.
- A self-hosted talking avatar, Axel, who answers questions on this site, and aXread, an assistant with an Easy Read mode and text-to-speech on every reply. Both are live here as a pilot.
Read more about Axel avatar and aXreadAn avatar or Easy Read assistant for your own site, with your own knowledge base and boundaries.
Through Accessible content and video and aXread.
Document Layer
A governance document store where every change is on the record
- Policies, plans and registers change hands and change shape. When a question comes later, the organisation needs to show what the document said at the time and who changed it.
- A working prototype, with tests, of a document store where every change is logged, tamper-evident and replayable: you can wind a document back to any point and see how it got there. Demonstrated with synthetic material only.
Read more about Document LayerA demonstration on synthetic material, and a conversation about what a governed document store would mean for your records.
Through AI and software systems.
PSight
Biometric wellness that follows the state of the nervous system
- Most wellness apps ask people to notice they are struggling and then pick the right exercise. By the time a person notices, the window for an easy intervention has often passed.
- A closed-loop prototype grounded in Polyvagal Theory. It reads heart rate variability and EEG signals, works out whether the nervous system is over- or under-aroused, and suggests the kind of exercise that fits that state. A working prototype.
Read more about PSightA demonstration of the prototype, and a conversation about state-aware wellness and regulation tools for your own program.
Through AI and software systems.
Ersher
Give it a website address; get a containerised, runnable version of that site
- Websites are tied to wherever they were first built and hosted. Moving one, rebuilding it, testing changes safely or bringing it inside a governed delivery pipeline takes weeks of hand work, and the knowledge of how it fits together lives in one person's head.
- Ersher, in development, takes a URL and produces a self-contained, containerised version of that site that can be run, inspected and worked on anywhere. It runs on the Decision Bus and the Project Layer as core infrastructure, so every step is planned, signed off by a person and recorded. It is planned as part of the AI-supported work pipeline: the way aXai takes on a site and delivers governed work on it.
Read more about ErsherToday: website, AI and accessibility work delivered through the Decision Bus and Project Layer. Ersher is in development; bringing us a URL is the planned way a project starts.
Through AI and software systems and Decision Bus and Project Layer and AI-supported work.
MCP connectivity
Letting other AI tools connect to the Bus and the Document Layer
- Organisations already use several AI tools. Governance only works if those tools can hand their decisions and documents to the same record, rather than each keeping its own.
- Planned: connections that let other AI tools reach the Decision Bus and the Document Layer through the Model Context Protocol, an open standard for connecting AI tools to systems.
Read more about MCP connectivityToday, the systems we build keep their decisions in the Bus. If your organisation wants its other AI tools to share that one record, we would like to hear about it.
Through AI and software systems.
Built with partners
Disability Services Exchange (DSX)
An NDIS provider discovery platform, built with DisabilityX. Participants search providers by location, service type and specialisation and see each provider's verification level. Designed for participants first, with Easy Read built in.
About DSXDisabilityX
A disability-led website fronted by Axel, with news and a podcast. We built it and keep it running; it shares its Easy Read assistant with this site.
Visit DisabilityX
What these products have in common
- Accessibility from the start. Easy Read, keyboard use and screen reader support are part of the design.
- Built for the people who use them. DSX began with what participants need to know about a provider.
- Made to run. These are working systems that we look after ourselves, so we live with our own decisions.
- Plain about what they found. Savanax shows disagreement; Context Scanner reports exactly what it checked.
How a client project runs
Your project runs on the same governed system as our own work. The plan sits in the Project Layer, a person signs off each step an AI agent carries through the Decision Bus, and every decision is recorded, so you can see what was done, by whom, and when.
You get a fixed written quote before work starts and a delivery you can audit afterwards. Client work falls into seven service families, set out on What we do. Talk to us about work similar to yours and we will walk you through how a comparable build runs.
Delivered work: videos
Two published videos, Never Funded and This Is AI-Supported Work, are on the AI avatar videos page, both presented by Axel's AI avatar. The second also opens our AI-supported work page.
Your own videos are made the same way, at a fixed price, from induction and training videos to service explainers.
What the statuses mean
- Live
- Deployed and running for people outside aXai.
- In production
- Used every day in our own delivery work.
- Pilot
- Running on this site while we learn from it.
- Prototype
- Working code with tests, shown on synthetic material only.
- In development
- Being built.
- Planned
- Designed, next in line.
Why aXai
aXai is a disability-led AI, software and accessibility company. We build AI systems, software, and concepts and prototypes ready to scale, and we deliver them through AI-supported work: skilled people with disability in real, properly paid roles, working to clear quality standards, with AI removing the barriers and people keeping the accountability.
Every project runs on our own infrastructure, the Decision Bus and Project Layer, so each step is planned, signed off by a person and on the record. That is our operating advantage. Read how it works.
Have a problem that looks like one of these?
Tell us what you are trying to do. We will say which of these applies, and what we would build for you.
Get in touch