A refusal beats a guess.
Estelle says when it cannot verify something, and it means it. A tool that answers every question is a tool somebody has to check behind, which puts the team back where it started.
We’re building the software factory AI coding agents work in.
Models write code faster than any team can read it. The missing piece isn’t a better agent. It’s the factory around it: somewhere to build, a way to ship that a team can trust, and a line back from production when something breaks.
Our story
Estelle started as a tool we built for ourselves, because we were shipping a large codebase with coding agents and the agents kept being confident about code that was not there.
A function that never existed, a signature that had moved, a decision the team made months ago that nobody wrote down — all of it stated with the same confidence as the parts that were right.
Reading every diff closely is the obvious fix. It doesn’t scale, and it’s the part people skip when they are tired.
So we built the other half: Estelle reads the repository, answers from what is actually in it, and blocks any change that references something it cannot find.
We run our own work through it every day, and when it refuses us we fix the code rather than the gate.
What we believe
Say what you measured.
Every number we publish has a script behind it in the repository. We don’t publish one we cannot reproduce. When a result moves the wrong way we say so, and the limits of the test go in the same paragraph as the number. We write about that in our research.
Your code is yours.
We do not train on what is in your repositories. You bring your own model and key, so the code goes to your provider under your terms. The long version is on how Estelle improves itself.
A human merges.
Estelle proposes a change, shows the evidence behind it and waits. It does not deploy, and it does not push to production. What we have in place today, and what we do not, is written out on security.
Where we are
Fate Labs is the trading name of Honour Systems Inc., incorporated in Ontario and working out of Toronto, and we’re a small team.
Nobody sits between a customer and the people writing the code. Write to khai@fatelabs.ca and an engineer answers.
Join us
We’re hiring in Toronto across engineering, deployment and go-to-market, and we take interns. Every role asks two written questions instead of a cover letter, because what someone has actually built tells us more than a template does.