We're building the operating system for AI engineering teams.
CodeCourier is on a mission to make autonomous AI agents a trustworthy member of every engineering org - shipping production code, on real repos, with the receipts to prove it.
Why we exist.
Engineering teams spend most of their cycles on the gap between an idea and a production-ready pull request: scoping, plumbing, debugging, polishing, reviewing. We believe that gap should belong to an agent - observable, accountable, and operated by the people who own the code. CodeCourier is the platform that makes that real for serious teams.
Agents that ship
Not chat. Not autocomplete. End-to-end agents that pick up an issue, plan, write, test, and open a PR you'd be proud to merge.
Operate with confidence
Branches, sandboxes, replayable runs and persistent learnings. Every action is observable, every output is reviewable.
Built for production
EU-resident data, dedicated sandboxes, fine-grained access control. Made for the teams that take software seriously.
Born from one stubborn question.
After years of shipping product as engineers, we kept asking the same thing: why is the boring 80% of an engineering task still on us, when the model can clearly do it?
The first prototypes were brittle. The agents wrote plausible-looking code that broke under real-world constraints - missing context, no memory, no review trail. So we threw them away and rebuilt from the bottom up around three primitives: contexts, sandboxes, and learnings.
CodeCourier is what came out of that rewrite. An agent platform designed for teams who care about software quality, not demo videos - running on isolated sandboxes, against real repos, with a full audit trail you can hand to your security team.
We're an independent team headquartered in Switzerland, building in the open with our early customers. We treat every PR our agents ship as our reputation. We don't take that lightly.
“Software is the rare craft where the maker is also the customer. We're building the tool we always wished we had.”
How we work.
Six commitments we make to ourselves, our customers and the code we touch.
Build with the agent
Every feature is dogfooded by our own agents shipping our own product. If we wouldn't ship it on our codebase, we don't ship it on yours.
Ship to production
Demos lie. We optimise for the agent's first merged PR, not its first message. Production is the only benchmark that matters.
Earn trust through evidence
Every run is replayable, every action is logged, every decision is auditable. Trust is built one verifiable PR at a time.
Default to openness
Public roadmaps, documented internals, honest changelogs. Closed-source ambition, open-source ethos.
Compound knowledge
Agents that don't learn from the last run are bots. We invest heavily in persistent context, learnings and shared memory so every run makes the next one sharper.
Default to clarity
Clear naming, clear errors, clear interfaces. If a teammate or an agent needs a meeting to understand it, we redesign it.
Quietly compounding.
Small team, big leverage.
A senior, opinionated crew building the agent platform we always wished we had - and hiring the rare engineers who'd ship it with us.
Building CodeCourier from Winterthur. Previously shipped developer tooling, distributed systems and AI products to production. Obsessed with the idea that the best engineering tool is the one that disappears into the work.
We're hiring quietly across engineering, design and devrel. If our principles resonate, get in touch. contact@
Winterthur, Switzerland.
CodeCourier
Schlosstalstrasse 202
8408 Winterthur
Switzerland
Remote-friendly across Europe. We're open to hybrid for senior roles in Zürich, Berlin and Lisbon.
Hire your first AI engineer.
Ship by lunchtime.
5 minutes to onboard. First PR within an hour. Cancel anytime.