The future is Australian-made
Matilda is out of open beta
Amanda Baughan · 1 min read
News from the lab
Our team in Melbourne writes up its launches, research and engineering here.
All posts
Show
Australian-made AI is independent AI
Matilda is live. MC-3 and our own cloud are next. Australia’s technology independence is no longer a pipe dream, and organisations in New Zealand, Singapore and beyond want it too.
Dave Lemphers · 6 min read
Why we’re getting right behind open weights
The strategy papers your people paste into public AI services can teach someone else’s model how you think. They should be teaching yours.
Dave Lemphers · 7 min read
Introducing Matilda’s Open Beta: Australian AI, Built for Trust
Today, we’re opening Matilda to a broader group of beta users. Matilda is available at matilda.maincode.com and on the Apple App Store. Matilda is Maincode’s Australian AI system: built from metal to model, designed for thoughtful use, and held to a higher standard for how AI should behave. We’re building Matilda because Australia should have more than access to powerful AI. We should have AI systems that understand our context, reflect our standards, run on infrastructure we can trust, and gi
Luke Borgnolo · 5 min read
Building the machine a model actually wants
A model, once you stop thinking of it as math and start thinking of it as a machine, is a dataflow. The weights stream in once and then sit still, the key and value for every token pile up and get read again by every token that follows, and the activations move from a projection into attention into a mixture of experts and back into the residual while a small sum runs between the GPUs so each one ends up holding the whole answer. Serving a model is really just moving those tensors along a path,
Dave Lemphers · 7 min read
Writing our own inference engine in Rust on the AMD MI355X
On the small all-reduce that tensor-parallel inference runs for every token, our from-scratch Rust inference engine comes out about 1.3x faster than AMD's RCCL at its best, sitting on the physical floor of the MI355X.
Dave Lemphers · 12 min read
Same Dictionary, Different Geometry
Recent work shows Muon representations are more robust and transferable. We ask what that means anatomically: at 124M scale, firing-pattern matching says the sparse features are the same — Muon recovers the AdamW feature set up to seed noise, packaged into a hotter, sparser, more crowded, and more seed-reproducible residual-stream geometry.
Max · 8 min read
MC-2: Australian Based Inference
MC-2 is Maincode's bare-metal AMD MI355X cluster in Melbourne — benchmarked as a production inference platform across model families, prompt shapes, and concurrency levels.
Max & Sara · 5 min read
What it's like being an AI Researcher on the Matilda team
We interview a lot of candidates who are making the move from academia into industry. It's a big transition and it makes sense that people carry assumptions from one world into the other. There's genuine overlap between the two, but the overlap is incomplete. The part that tends to surprise people most is how much engineering is involved. This post is specifically about what it means to be an AI researcher on the Matilda team, where engineering isn't a side activity but the bridge between havin
Dave Lemphers · 4 min read
Model Release: Maincoder-1B
Our newest 1B parameter coding model that scores 76% on HumanEval
Kees Bakker · 2 min read
The ROI of an AI Token Factory
Why owning an on-premise AI Token Factory enables 7.7x more token generation over APIs, cutting AI costs by 87%
Kees Bakker · 4 min read
Designing Matilda: The Next Generation of AI UX
SXSW UX Recap
Amanda Baughan · 5 min read
Inference Time Licensing
If you ask me where I learned the word “tree,” I cannot give you a neat citation.
Kees Bakker · 2 min read
It's down...ish
The Hardest Outage is the One You're Not Having
Dave Lemphers · 4 min read
The Language of Computer Control Agents
The current focus of frontier AI research is on Computer Control Agents (CCA). The goal of this research is the automation of workflows in web browsers and operating systems on computers and mobile devices. In our previous post, we discussed that most CCAs operate by employing large language models (LLM). Which language do CCAs speak? Evolved Natural Language The origins of natural language date back more than 100,000 years [Miyagawa et al., 2025]. Humans evolved language for coordination,
Kees Bakker · 2 min read
Moving Beyond Static AI: The Need for Dynamic Personalisation and Where We Go Next
Why the one-size-fits-all approach to interaction modelling falls short and why we need to re-think the role of the human expert.
Dave Lemphers · 7 min read
Co-discovery: Humans in Automated Scientific Discovery
Why end-to-end automation of scientific discovery is probably not where the biggest opportunity lies right now.
Kees Bakker · 3 min read
The Future of Computer Control Agents
The current frontier in the fast-paced world of AI research is Computer Control Agents or CCA. OpenAI released Operator and Anthropic its computer use capability. Large language models (LLM) that interact with websites and the apps on our phones like humans do. Vision is enabled via screen captures and actions are mouse clicks, screen touches and keystrokes. How did we get here? The AI Revolution The revolution of language-based AI models started with OpenAI's belief in the scaling hypothes
Kees Bakker · 3 min read
Why Principles Matter
You can’t carry lightning on a shoestring
Dave Lemphers · 4 min read
There are no posts in that category yet.