Maincode is an Australian AI research company. We're building AI that earns trust in the real world, to move humans forward. We create intelligent systems that understand context, anticipate needs, and turn ideas into action, unlocking entirely new ways for people to work and create. The future isn’t just about software that stores information. It’s about technology that thinks, adapts, and acts. We are pioneering the next generation of AI-powered, action-driven systems that amplify human capability, accelerate workflows, and make work feel effortless. We believe AI should do more than assist, it should empower. If you're passionate about building the next era of intelligent software, join us.
External link for Maincode
Melbourne, VIC 3000, AU
Now available on Windows and Linux, in addition to macOS. Give it a go!
Want more of Matilda? We're excited to launch Matilda Desktop. It brings chat, coding, browsing, and agent orchestration together into one single surface, so you can stop tab-hopping between your terminal, browser, and AI dashboard. Whether you're writing code, running your agents, or just asking questions, it all happens in one place. Try it here: https://lnkd.in/gPyxQPpr
The Matilda API is here. Served onshore from Melbourne, on our own GPUs. Two SDKs, live now: 1. Client SDK for chat, streaming, structured output, file uploads. Point your existing code at Matilda and go. 2. Agent SDK for named agents, tool loops, multi-turn sessions, durable stream resume. Agents that pick up where they left off, no interruptions. Setup takes minutes: get your API key, install the SDK, start building. No complex auth flows, no configuration overhead. Australian-made. Ready when you are. Let's build. https://lnkd.in/gEH6-EnT
Maincode reposted this
Very proud to see three papers from the Maincode core team accepted to the #EMNLP2026 Industry Track 🎉📚 What I really value about Maincode is that we are not only trying to build great AI products. We also do serious research, publish at leading peer-reviewed conferences, and put our work in front of the broader research community to be challenged, tested, and improved. 🔬 And what makes these three papers particularly meaningful is that every author is part of the Maincode core team. This is research coming directly from the people building the technology. 💻🧠 These papers were not projects we started simply because we wanted publications. They emerged naturally from building and exploring Matilda. As we pushed Matilda further, we kept encountering problems that went beyond a single product feature: long-context and RAG serving, million-token inference on real hardware, and how an AI assistant should reflect an Australian voice without reducing it to caricature. They started as real product and engineering needs. We then realised that the underlying questions were much broader than Matilda itself. So we generalised the problems across models, systems and domains, studied them rigorously, and shared what we learned through peer review with the wider research community. 🌏 Product development gives us important research questions. Research gives us better ways to build. And when the results are useful beyond our own product, we want to share them. 🤝 Very proud of the Maincode team ❤️ (Dave Lemphers, Maxwell Twelftree, Jared Collette, An Chi He, Amanda Baughan, Luke Borgnolo)— and excited to keep turning the challenges we encounter while building Matilda into research that contributes back to both the Australian and global AI communities. 🌏🇦🇺
Want more of Matilda? We're excited to launch Matilda Desktop. It brings chat, coding, browsing, and agent orchestration together into one single surface, so you can stop tab-hopping between your terminal, browser, and AI dashboard. Whether you're writing code, running your agents, or just asking questions, it all happens in one place. Try it here: https://lnkd.in/gPyxQPpr
Maincode reposted this
We're open sourcing mainarch, our reference inference engine. It's written in Rust and it talks straight to the kernel driver. No ROCm, no HIP, no HSA runtime. It opens /dev/kfd, builds the GPU dispatch packet by hand, rings the doorbell, and runs a real model on an MI355X end to end. We're putting it out because this part of AI engineering is mostly undocumented and sits inside a handful of companies, and we think more people should get to see how it actually works. It's the companion to From Tin to Tokens. The book explains it, the repo is the thing itself, running. Apache 2.0, free, one command to try it. Built at Maincode in Australia with love! Repo in the comments. #OpenSource #Rust #AI #GPU #AMD
Maincode reposted this
Almost everything genuinely known about serving large models fast lives inside a handful of companies, and almost none of it gets written down. That's why we decided to put all our learnings into a book that you can have for free. From Tin to Tokens is a free 192-page book on building an LLM inference stack from scratch in Rust. It goes from the driver interface, where you are writing ioctl structures by hand, to an HTTP endpoint serving thousands of concurrent users: attention kernels, a paged KV cache, a scheduler, multi-GPU collectives, a quantisation pipeline, and the correctness apparatus that proves any of it works. It came out of the work we do at Maincode, where we build frontier AI in Australia on our own machines. The inference layer is where the physics of the hardware meets the economics of the product, and it is where most of the difference between a demo and a business gets decided. And we made it free because this kind of knowledge needs to be shared, and it's super cool to boot! And if you find something wrong, which I'm sure you will, let us know! Link to the repo is in the comments. #Rust #LLM #GPU #Inference
Maincode reposted this
Super excited to announce Matilda is now available on Google Play!! https://lnkd.in/g2kCUWHj Thanks to all our Android fans for your patience! Love shipping more and more Australian-made AI Matilda gold!
Maincode reposted this
Our latest paper was accepted into MOSS @ COLM 2026 :) We looked at whether Muon actually changes the features a model learns, or mostly just how it stores them. On a 124M GPT-2-class model the SAE features match AdamW up to seed noise. The useful part of Muon’s advantage lives in the value circuit, applying it only to WV/WO recovers most of the loss and reproducibility gains, while doing the same on the QK scoring circuit doesn’t. We are hiring researchers and engineers across Melbourne and Seattle please reach out if this kind of work interests you. PDF attached and OpenReview in the comments.
Matilda Code is here. Built from the ground up in Australia. Matilda Code is an AI coding agent that lives in your terminal, reads your codebase, writes features, fixes bugs, and ships alongside you. Powered by Matilda, our Australian-built AI. Most other AI coding tools send your code and data overseas. Matilda Code keeps it on Australian soil. If you're a dev building here, your IP and your customer data shouldn't have to leave the country just to get a decent coding assistant. It's available now, and it's free while we keep building. https://lnkd.in/gY9cPjuv
Maincode reposted this
More than a year and a half ago, Maincode set off on a mission to create AI in Australia. We had no idea where we were going, who was coming, or where we would end up. Today, I'm so proud and excited to release Matilda into open beta. You can sign-up at https://lnkd.in/g8EStXDf or download Matilda from the App Store, https://lnkd.in/gaj-28wG None of this would have been possible without the incredible Maincoders who have worked tirelessly to bring Matilda to life. Congratulations team, Australia is proud of you! Sara E. Rodriguez Gomez An Chi He Yue Yang Kees Bakker Jared Collette Maxwell Twelftree Amanda Baughan, PhD Rea Kenan Luke Borgnolo