chat --status=private-beta

European AI chatbots, grounded in your content.

Hosted in Germany on Hetzner. Inference on Mistral. Never trained on your data. Every conversation, token, and audit entry visible in your dashboard.

Flat monthly fee, no per-message surprises. Bring your own Mistral key on BYOK — 79 €/mo, unlimited messages. Verified academic tenants get 25% off every paid tier.

principles

sovereign · grounded · auditable
Sovereign

Hetzner DE · Mistral FR

European hosting, European inference. Mistral and every sub-processor are bound not to train on your data.

Grounded

GitHub · PDFs · your library

Sync your Markdown repo or upload PDFs we OCR. The chatbot answers from your own material — grounded in your library, not the open web.

Auditable

Every conversation · Every token

Operator-grade dashboard: every reply, every model call, every audit-log entry visible to the operator who runs the bot.

data flow

EU data residency

Every request stays inside the European Union. No transfer to the US or any other third country.

Visitor

Your website

Anonymous to us — no cookies, no profiling.

Platform · DE

Hetzner · Falkenstein

Postgres with row-level isolation per tenant.

Inference · FR

Mistral · Paris API

EU endpoint. No training on your data.

↑ Everything inside the dashed border above is inside the EU.

try it · live chat

Live

Ask it anything about Oshu. It's the same widget you'd ship to your site. Nothing mocked, nothing prerecorded.

oshu · loading

booting widget.js · connecting to Oshu · stand by

status

private beta

Access is by invitation while we stabilise the platform. Email oshu@beyondsimulations.com with a sentence about what you want to build — a product chatbot, an internal Q&A, a course assistant, anything. Invite to live embedded agent typically takes under 30 minutes.

observability

your dashboard view

The chat widget on this page is the real product. Open it from the launcher to see what your visitors see. The view below is what you see in the dashboard — every reply, with model, latency, tokens, and library tool calls.

course-assistant.example.edu Live
user
Derive the gradient of f(x) = x² sin(x).
May 13, 2026, 04:24 PM
assistant · mistral-small-latest
By the product rule, f′(x) = 2x·sin(x) + x²·cos(x).
May 13, 2026, 04:24 PM 184 tok 340ms 1 source
document_library ×1
Messages · 14 d +38 %
4 821
Apr 30 May 13
Model split last 30 d
  • mistral-large 62 %
  • mistral-small 28 %
  • codestral 10 %

The dashboard tracks every conversation: message volume, model mix, latency, token spend and a per-agent audit log. Visitors never see any of this — they just see the widget.

capabilities

01

For Mittelstand and SaaS teams — product, docs, and support

Sync Markdown from your GitHub repos, drop the widget on your site, and the chatbot answers from your own material — grounded in your library for accurate, on-topic replies. Every conversation visible in your dashboard. Sovereign by default: Hetzner DE + Mistral FR + binding no-training across the stack.

02

For agencies and consultants — many client sites, one bill

One workspace, one bill, up to fifty chatbots. Each agent has its own persona, library, and embed snippet. Same sovereign stack for every client; same dashboard to audit every conversation across the portfolio.

03

For universities and lecturers — one bot per course

Build one chatbot per lecture, ground it in your slides and notes. Replies are grounded in the source PDFs you upload; KaTeX renders math inline. The EU sovereignty stack is Mistral Paris + Hetzner Falkenstein + GmbH on the AVV. Drops into Canvas, ILIAS, OpenOlat, Stud.IP via the rich-text editor; Moodle via your sysadmin (native LTI 1.3 on the roadmap). 25% off every paid tier for verified academic tenants.

04

For technical founders — bring your own Mistral key

BYOK at 79 €/mo flat: every chat reply, every PDF you upload, every Markdown sync passes through your own Mistral key. Unlimited messages, unlimited library, per-agent model selection. The dashboard still shows you every conversation, token, and audit entry.

features

GitHub Markdown sync

Point an agent at a repo and folder. New commits are pulled into the library on a schedule.

PDF upload with OCR

Drop in PDFs and we OCR them into Markdown for the agent’s library. Scanned documents work too.

LaTeX and math rendering

Inline and block KaTeX so equations, derivations, and scientific notation render correctly.

Grounded answers

The chatbot answers from your library — replies stay grounded in your own material instead of guessing from the open web.

Streaming replies

Replies stream token by token over SSE. Library searches may add a few seconds before the first token.

Theming

Brand colours, radius, fonts, and labels are configurable per agent. The widget looks like part of your site.

Embed layouts

One script tag drops the agent in. Pick floating button, inline panel, full page, or iframe per embed.

Operator dashboard

Every conversation, message volume, latency, model usage, and a per-agent audit log — all in the same workspace.

AVV available

Art. 28 DPA optional — signed bilaterally on request before you onboard personal data. Free.

No training on your data

Neither we nor Mistral train on data submitted via the API. Every plan, managed or BYOK.

12-month auto-purge

Chat logs delete themselves after a year. Wipe earlier any time from the dashboard.

EU endpoints only

Hosted at Hetzner Germany, inference via Mistral's EU API endpoint. No third-country transfer.

plans

simple pricing

Start free on the Free tier. Hobby at 9 € adds 1 500 messages and 250 pages; Solo at 29 € gives you three bots and 5 000 messages; Studio at 59 € is for teams running up to ten embedded agents; Scale at 199 € fits multi-product teams and agencies with 50 bots and 40 000 messages a month. BYOK at 79 € flat removes our message and library caps entirely — you bring your own Mistral key and we route every inference + OCR call through it. We provision tenants by hand for now — email us and you are live in under 30 minutes.

Free

0 €

1 bot · 100 msgs · 50 pages

  • Embed widget, full theming
  • Base and Premium model tiers
  • No card required

Free entry tier. 100 messages a month, then a hard cap, no overage.

Start free →

Hobby

9 €/ mo

1 bot · 1 500 msgs · 250 pages

  • One agent, full theming
  • Base and Premium model tiers
  • Grounded, LaTeX, GitHub sync

For a personal site, a small project, or a single course.

Email us →

Solo

29 €/ mo

3 bots · 5 000 msgs · 1 000 pages

  • Three agents on three sites
  • Base and Premium model tiers
  • Grounded, LaTeX, GitHub sync

For small businesses and active hobbyists.

Email us →

Studio Most popular

59 €/ mo

10 bots · 10 000 msgs · 5 000 pages

  • Up to ten agents per workspace
  • For docs, support, onboarding
  • Same hard-cap, no overage

For teams running several embedded agents.

Email us →

Scale

199 €/ mo

50 bots · 40 000 msgs · 25 000 pages

  • 50 agents per workspace
  • Multi-product teams, agencies
  • Centralised billing

For multi-product teams and agencies.

Email us →

BYOK

79 €/ mo flat

50 bots · unlimited msgs + library

  • Use your own Mistral key
  • Unlimited msgs + library
  • Per-agent model picker

Flat platform fee. Mistral bills you directly for inference and OCR.

Email us →

Custom

from 299 €/ mo

negotiated

  • Managed deployment by our team
  • Self-hosted or white-label
  • Tailored caps and SLAs

Fully managed: we set up the agents and libraries, deploy the widget, and operate the stack. Email us to scope.

Email us →

At 10 000 messages / month: 59 €/mo on Studio (10 000 cap, flat) versus ~140 €/mo on Chatbase Standard (12 000 message credits, 5 bots) or ~460 €/mo on Chatbase Pro (40 000 credits). Flat fee, hard cap, no per-message billing.

Prices are net, excl. VAT. 19% German VAT is added at checkout for German customers. EU businesses outside Germany with a valid VAT-ID: reverse-charge (no VAT). Non-EU: no VAT.

next steps

Already invited?

Sign in with the email address tied to your workspace.

Need access?

Tell us a bit about your use case and we will provision a workspace for you.