███╗ ███╗ █████╗ ██╗ ███╗ ██╗ ██████╗ ██████╗ ██████╗ ███████╗
████╗ ████║ ██╔══██╗ ██║ ████╗ ██║ ██╔════╝ ██╔═══██╗ ██╔══██╗ ██╔════╝
██╔████╔██║ ███████║ ██║ ██╔██╗ ██║ ██║ ██║ ██║ ██║ ██║ █████╗
██║╚██╔╝██║ ██╔══██║ ██║ ██║╚██╗██║ ██║ ██║ ██║ ██║ ██║ ██╔══╝
██║ ╚═╝ ██║ ██║ ██║ ██║ ██║ ╚████║ ╚██████╗ ╚██████╔╝ ██████╔╝ ███████╗
╚═╝ ╚═╝ ╚═╝ ╚═╝ ╚═╝ ╚═╝ ╚═══╝ ╚═════╝ ╚═════╝ ╚═════╝ ╚══════╝
$ whoami
Maincode is an Australian AI research company building MatildaMATILDA[PRIVATE BETA]
Now in private beta — request early access: an assistant that understands your work, learns how you operate, and helps you move faster, safely. We're building AI that earns trust in the real world, to move humans forward.
Now in private beta — request early access: an assistant that understands your work, learns how you operate, and helps you move faster, safely. We're building AI that earns trust in the real world, to move humans forward.
$ cat status.txt
LOCATIONMelbourne, Australia
FOCUSAI research - Product development - Applied intelligence
CONTACThello@maincode.com
TIP: ↑↓ arrows to navigate - type help for commands
$ ls -la /research/
// PRODUCT AND INFRASTRUCTURE
matilda/
2026-04
Intelligent AI assistant - flagship product
Matilda is our flagship AI assistant, built to understand complex workflows,
reason about context, and take meaningful action. Conversational AI that
actually gets work done.
LLM RAG Multi-agent Real-time API
LLM RAG Multi-agent Real-time API
inference-cluster/
2025-06
GPU inference infrastructure
High-availability inference infrastructure serving Matilda at scale.
Custom batching, model sharding, and dynamic routing for
sub-200ms response latency globally.
CUDA Kubernetes Terraform GCP vLLM
CUDA Kubernetes Terraform GCP vLLM
// RESEARCH
context-engine/
2026-02
Long-context reasoning framework
Proprietary framework for maintaining coherent reasoning across
extended context windows. Enables Matilda to track complex multi-step
workflows without losing the thread.
Transformers Attention mechanisms Memory systems Python
Transformers Attention mechanisms Memory systems Python
action-layer/
2025-11
Tool-use and agentic execution
The layer that turns Matilda from a chatbot into an agent.
Structured tool calling, safe execution sandboxing, and
multi-step planning with real-world side effects.
Agent framework Sandboxing Tool use Planning
Agent framework Sandboxing Tool use Planning
eval-suite/
2025-09
Evaluation and safety benchmarks
Internal evaluation framework for measuring model capability,
safety alignment, and task completion accuracy. Automated
regression testing across 10K+ scenarios.
Benchmarking Safety RLHF CI/CD
Benchmarking Safety RLHF CI/CD
// RESIDENCY
residency/
3–6 mo
Australian-made AI research residency
// BLOG
$ cat about.md
Maincode exists because we believe AI should do more than generate text. It should
understand context, deliver useful results, and earn trust.
Every interaction, every time.
We're building Matilda to be the AI assistant we always wanted: one that grasps the
full picture of what you're working on, remembers what matters, and helps you
move through complexity on your terms.
Founded in Melbourne, our team spans AI research, product engineering, and design.
We're Australian-made by design: independent, close to the people we build for, and
here to set a higher standard for AI safety. We ship fast, measure everything, and
build in the open where we can.
$ cat principles.txt
01Move Humans Forward. Create technology that improves humankind
02Do The Hard Thing. Make timeless choices
03Take Point. Are excited to do what others won't
04It's 2AM in the Lab. Thrive on challenges
05Stay Foolish. Know just enough to not know better
06Are High Voltage. Optimise for full power
07Ship. Deliver stuff that matters
███╗ ███╗ █████╗ ████████╗ ██╗ ██╗ ██████╗ █████╗
████╗ ████║ ██╔══██╗ ╚══██╔══╝ ██║ ██║ ██╔══██╗ ██╔══██╗
██╔████╔██║ ███████║ ██║ ██║ ██║ ██║ ██║ ███████║
██║╚██╔╝██║ ██╔══██║ ██║ ██║ ██║ ██║ ██║ ██╔══██║
██║ ╚═╝ ██║ ██║ ██║ ██║ ██║ ███████╗ ██████╔╝ ██║ ██║
╚═╝ ╚═╝ ╚═╝ ╚═╝ ╚═╝ ╚═╝ ╚══════╝ ╚═════╝ ╚═╝ ╚═╝
$ cat matilda.md
Safe by design. Transparent by default. Approachable by choice.
Matilda is Maincode's flagship AI assistant, built in Australia, from
infrastructure to model behaviour, and built for a higher quality of conversation.
Not another chatbot. Matilda is designed to feel intuitive from the start: clear
in how it responds, deliberate in what it does, and focused on helping you move
through complex work without friction.
// CAPABILITIES
01Deep contextual understanding across long conversations
02Multi-step reasoning with visible, inspectable steps
03Anticipates your needs, but doesn't assume; asks before it acts, checks when unsure
04Learns your workflow over time, on your terms, with clear controls
05Approachable from the first message, no prompt engineering required
06Built and hosted in Australia, aligned with Australian privacy standards
$ cat careers.md
We're a small team doing hard things. If you're the kind of person who builds
at 2am because you can't stop thinking about the problem — we should talk.
Maincode is building AI that works in the real world. We need people who care
about craft, move fast, and don't need permission to ship.
// WHAT WE VALUE
01Ownership. You don't wait to be told what to build
02Gradient. You learn new things quickly
03Range. You keep learning, going deep in any given domain
04Speed. Ship, learn, iterate
05Taste. You care about the details that others skip
06Humility. No ego. No posturing. Mission first
// LOCATION
Melbourne, Australia. On-site. Full-time.
$ ls -lt posts/
loading posts...
$ cat residency.md
A paid 3 to 6 month research residency, built for late-stage PhD students and
standout early-career researchers. You bring the question. We bring the compute,
the infrastructure, the systems, and direct access to the team building
Australian-made AI from first principles.
This is high-agency, autonomous research. You own the project, you set the
direction, you run the experiments. You are the domain expert on your problem,
and you will be treated as one. We are here to remove the constraints and think
alongside you, not to manage the work.
Publish at top venues. Get your hands on state-of-the-art systems. Watch strong
work move from a paper into production.
// RESEARCH AREAS
Maincode is an applied research lab where research meets production systems.
Matilda is our flagship. The research areas below are where our focus sits.
Within them we're open, from pragmatic projects that integrate into Matilda to
high-upside, frontier ideas.
agents
Take action in the real world, reliably
Tool use, planning, memory, and multi-agent orchestration. Turning a model
into a system that can carry out real tasks. The research directly behind what
Matilda does.
Multi-agent Orchestration Tool use Planning Memory
Multi-agent Orchestration Tool use Planning Memory
safety & alignment
Capable systems you can actually rely on
Alignment, interpretability, oversight, and evaluation. The work that decides
whether a powerful system is trustworthy.
Interpretability Oversight Evals Robustness Red-teaming
Interpretability Oversight Evals Robustness Red-teaming
training & algorithms
How models learn, and the algorithms underneath
Training at scale, and the learning algorithms that drive it. New
objectives, optimisation methods, reasoning architectures, and
reinforcement learning for reasoning and planning.
Reasoning RL Optimisation Architectures Objectives
Reasoning RL Optimisation Architectures Objectives
data
What a model learns from decides what it becomes
Curation, filtering, synthesis, and mixture design, from large-scale pipelines
to the statistics of what to train on. This runs from pragmatic data
engineering to the principled end: information-theoretic and causal approaches
to selection, quality verification, and pruning.
Curation Synthesis Mixture design Information theory Pipelines
Curation Synthesis Mixture design Information theory Pipelines
multimodal
Perceiving and reasoning across modes
Vision, language, audio, and beyond. Grounding perception in reasoning, fusing
modalities, and building systems that understand far more than text.
Vision-language Grounding Fusion Perception
Vision-language Grounding Fusion Perception
efficiency
Speed at every level, from the metal to the model
We train and serve on our own GPUs. Make any layer faster, from the kernels
to the model.
Kernels Parallelism Sharding Decoding Quantisation Scheduling
Kernels Parallelism Sharding Decoding Quantisation Scheduling
// WHAT YOU GET
01Publish at top-tier conferences and journals
02Access to our GPU cluster and the training and inference stack we run Matilda on
03A direct line from research to production. Strong work can ship inside Matilda
04Close collaboration with the core team. On-site, hybrid, or remote with visits
// WHO YOU ARE
01A late-stage PhD student, or a standout early-career researcher
02Published at top venues, or clearly on track to
03Comfortable with large-scale experiments and systems-level work
04Driven by the question, not the title
██████╗ █████╗ ███╗ ███╗
╚════██╗ ██╔══██╗ ████╗ ████║
█████╔╝ ███████║ ██╔████╔██║
██╔═══╝ ██╔══██║ ██║╚██╔╝██║
███████╗ ██║ ██║ ██║ ╚═╝ ██║
╚══════╝ ╚═╝ ╚═╝ ╚═╝ ╚═╝
$ cat /dev/late-night
The world is quiet now.
The slack is still. The tickets wait.
But here, in the glow of the terminal,
we are building what morning will find.
No meetings. No metrics. No mercy.
Just the problem, the cursor, and the clock
that stopped mattering two hours ago.
This is where it happens.
Not in the pitch deck. Not in the standup.
Here. Now. In the space between
what doesn't work yet
and what's about to.
The slack is still. The tickets wait.
But here, in the glow of the terminal,
we are building what morning will find.
No meetings. No metrics. No mercy.
Just the problem, the cursor, and the clock
that stopped mattering two hours ago.
This is where it happens.
Not in the pitch deck. Not in the standup.
Here. Now. In the space between
what doesn't work yet
and what's about to.
$ cat status.txt
MODEDeep work
CAFFEINECritical
DISTRACTIONS0
COMMITSPending
COFFEES0
$ type help for lab commands