Loading
MYG Labs is our research and engineering arm. We build predictive systems with the best product architecture available, contribute the tools we rely on back to open source, and build the proprietary intelligence that lets a smaller company operate like an enterprise.
A lot of what we build is for ourselves first. The best of it we open-source, so the wider community can read it, run it, and build on it.
Urfael is our open-source, self-hosted AI assistant: an always-on local brain that runs on your own machine, on the flat-rate Claude subscription you already have, with no API key and nothing listening on a network port. Built security-first, it allowlists who can reach it and sandboxes every autonomous action fail-closed. It is voice-capable, but what matters is how little it could ever break: it ships with the proof attached, a benchmark that runs the real attack classes that compromised other agents in the wild and lands 10 out of 10. It is where we prove the patterns we later bring into client work.
The tools we depend on get developed in the open, commit by commit.
Readable architecture, no black boxes, designed to be forked and bent to your needs.
What we learn building for clients flows back into work anyone can use.
Our predictive engines are built on the best product architecture we can engineer: systems that model your operation, simulate it forward with swarm intelligence, and surface the decision before you have to make it. They are built to handle genuinely complex infrastructure, and to become load-bearing once they prove it.
We spawn swarms of lightweight agents, each modelling a different version of your operation, and let them run thousands of scenarios in parallel. You see how the outcomes diverge before you commit to any single one of them.
The very best product architecture, applied to prediction. Engines built to ingest your data, learn its shape, and forecast what happens next across demand, pricing, risk, and capacity, with the evidence attached.
We use AI to design algorithms and systems that meet genuinely complex infrastructure demands, then wire them into how you already run, so the engine becomes a primary component of management and organisation rather than a dashboard you check.
Once integrated, the engine is an operational necessity, not a vendor you rent. It lets a smaller company compete with enterprise players and, given time, stop renting the tools it depends on entirely.
When an engine is good enough, it stops being software you bought and becomes part of how the organisation runs. Management leans on it. Planning assumes it. That is the point: a smaller company gets to compete with the enterprise on the same footing, and given time, stops renting the tools it depends on and runs on systems it owns.
If you want a predictive system that becomes part of how your business runs, this is where it starts.