Everything your security officer wants to know.

We use commercial model providers under agreements that exclude customer content from training. Your code is sent to answer questions about your code, and that is the whole of it; your production data is not part of it, because the agents work in the same anonymised environments as our developers. If you want a specific provider excluded, or the work done on infrastructure you nominate, that is a scoping conversation we are happy to have.

The change trail, yes: every change arrives as a commit and a pull request in your own repository, reviewed by a second agent against written criteria, blocked or passed by automated checks, and signed off by a named person. That trail lives in your systems and stays there after we leave. A record of every tool an agent invoked or every byte that left the repository is a different thing, and part of the scoping conversation.

Less than a developer. Every agent run starts in a fresh, isolated sandbox in our Azure environment in West Europe, from a pinned image, and is deleted afterwards. The agent never holds a key: a proxy adds credentials on the way out, and the network is closed except for a list of approved hosts. Each role runs under its own limited identity, which cannot push to main, cannot bypass a branch policy and cannot complete a merge. A change only reaches you as a pull request that one of our engineers reviews and approves. Every run is capped and logged, and any step can be stopped with one switch. The full line is on the how we build page.

Working copies live on managed company equipment for the duration of the engagement and are removed at the end. You revoke access the day we finish. Your code serves your engagement only; afterwards it stays where it belongs, with you.

Start with your production data: our developers and engineers never work with it, whether in Serbia or in Rotterdam. Development, test and acceptance run on anonymised data and databases; production is reached only by the people who need it for operations. So the question of where a developer sits rarely touches personal data at all, and the same holds for our coding agents: they work in the same anonymised environments. What remains, such as operational access and support, is covered in the contract: standard contractual clauses for our Serbian entities, a transfer impact assessment, and a list of sub-processors with name and location. If your policy or your regulator requires EU-only processing, that is part of the scoping conversation.

We do, on exactly the same terms as code typed by a person: the same warranty, the same liability, the same obligation to fix it. In our contracts, code from an agent is simply code. That is the whole reason the gate exists: every line is reviewed before we put our name to it.

You own the code written for you, outright, in your repository. Entry, the foundation underneath it, is open source under Apache-2.0 and stays that way, so it remains free to use for you and for the team after you. The gates and checks we configure for you are yours too.