[ about ]

About Oshu

Oshu started as a chatbot I built for a single course. At the University of Hamburg Business School in 2025 I put together a small Socratic tutor for the Applied Optimization module. It pointed students at the lecture material and asked the kind of questions that nudged them toward an answer instead of handing one over. A grant from the Claussen-Simon-Stiftung let me hire a student assistant, and the prototype grew into an open framework: a Rust layer that adds telemetry, semester-scoped pseudonymous IDs, rate limiting, and an embeddable widget on top of OpenWebUI as the chat backend. The code is open source on GitHub. By the winter semester it was running in more than eight courses at the university.

Oshu came out of that work, no longer tied to the university. The research framework still does what it was built for, but the operational bar to deploy it is high. To run it you need OpenWebUI, Postgres, a Rust toolchain, the upstream API keys correctly configured, and someone on call when one of those moves. That works for a researcher with grant funding and a student assistant. It doesn't work for a lecturer with one course, an agency with a dozen client sites, or a company that wants a grounded AI bot on its docs site next week.

Oshu is the hosted version of the same idea, on a fully European pipeline. Hosting is Hetzner Falkenstein. Inference is Mistral Paris. No US clouds anywhere in the request path. Mistral's sub-processor agreement binds them not to train on customer conversations, and every downstream contract has the same kind of binding. Pricing is a flat monthly fee with a hard cap. No per-message billing, no surprise overage. If you prefer, you can bring your own Mistral key on the BYOK tier and pay Mistral directly for inference. Library management happens in Oshu, so there's no separate backend to operate. Replies link back to the source file they came from. Operators get a full dashboard: per-agent analytics, conversation history, and an audit log. The widget supports separate bots per site with separate libraries and branding. The company-side paperwork (DPA, sub-processor list, TOMs) is already in place, so signing Oshu doesn't turn into a six-week procurement loop.

About the name. Oshu is the Japanese word for Europe: Ōshū, 欧州. Europe itself got its name from somewhere else. The continent is named after Europa, a Phoenician princess from the eastern Mediterranean, whose myth blurs into the older Babylonian goddess Ishtar. In the original story, a bull (Zeus) carries her across the sea. In ours, she's the one leading him around — you'll meet them both elsewhere on the site, and yes, she's in charge. Oshu does the same trick in another language: a European product named from the other side of the world.

Naming this was, honestly, harder than I expected. Every short, memorable word in software and AI is already a company, a domain, a Series A, or all three. I worked through dozens of candidates in half a dozen languages and ran competitor and trademark checks on each. Most collided. Oshu survived. Sat next to the Europa myth and the illustrations, it gives the brand some depth. That's where I landed.

Oshu is built by BeyondSimulations, a small Hamburg company. The contracts and the invoices come from BeyondSimulations. Oshu is the product you actually use.