An AI software company, founded by researchers turned engineers.

Yes. In embedded engineering we work to your standards: your repository, CI, linters and code owners. Our pull requests go through your reviewers and your branch protection.

With checks and an engineer in control, yes. Tests, type checks, lint, a dependency audit, a secrets scan and static security analysis run on every pull request, and one of our engineers reads the change and approves it before it merges.

We work with current frontier models such as GPT-6, Claude Opus 5.5 and Gemini 3.1 Pro, with open-weight models such as DeepSeek V4, Qwen3.8 and Gemma 4 on your own GPUs, and with small trained classifiers where an LLM is the wrong tool.

Yes. We build an evaluation set from your real work first, adapt an open model to your data with continued pretraining, fine-tuning and RL post-training, then distil it into the smallest model that holds the score. It runs in your cloud, on your own GPUs or on the device, and you own the weights, the datasets and the evaluation suite.

Yes. Small models run on iPhone, Android, Mac and Windows devices and in the browser with no network, and hand harder requests to a private server when one is reachable. It suits field work, sensitive data and apps that must work anywhere.

Yes, through the AI coding tools we name before work starts, each on business terms that prohibit training on its input. You can veto any of them, and your code is never used to train any model.

Yes. We know how these tools shape a codebase, and we use AI coding tools ourselves. We stabilise what breaks, move secrets out of the repository, put tests on the paths that make money and leave automated checks running in your repository, usually without a rewrite.

A first working slice in two to three weeks, then a demo every week. Every branch also gets a preview link on our domain, so you can click through work in progress.

You own the code from the first commit. We sign an NDA by default, before we see anything, and remove our access within seven days of handover.

Least privilege: read-only access for a technical review, and for build work a named account of our own with write access to branches, never main. Your code lives in your repository, our copy of it on our source-control host, our engineers’ encrypted machines, our agents’ isolated environments and branch previews on our domain; production data stays out of all of them unless the statement of work says so, and a secret found in your repository is reported the same day and rotated within 24 hours. The details are on our security page.

Anything that does not meet the statement of work, reported in writing within 30 days of handover, we correct under it. Embedded or ongoing engineering ends on 14 days’ written notice from either side, and within seven days of the end we revoke our access, delete local copies and confirm both in writing.

Yes. Bring the idea and what you want customers to be able to do. We scope it on a call, show working software every week, and hand over the code, a README, runbooks and architecture notes, so another team can run it without us.

Our working day overlaps working hours in Europe, the Middle East and Asia. You see a demo every week, and work in progress sits on the intake board and on preview links, so you can check where things stand at any hour.

Startups and enterprise teams that need an AI system built, a product shipped, or an AI-built app made safe. We join a team that has engineers, or act as the engineering team for one that does not.

Scope and fee are agreed on a call, once we understand the problem, and set out in a signed statement of work. A change of scope is agreed in writing before the work changes.