
Cited answers · EU-hosted · sovereign
Ask Open Source Europe
A public assistant that searches the official OSE documentation and answers with sources you can verify. Fiscal hosting, onboarding, policies, processes. Ask in your own words.
The chat needs JavaScript. Turn it on in your browser to ask a question.
Try
Use it in your own tools — or run it locally
The OSE knowledge RAG is an MCP server that runs on your own machine, so OSE never sees your questions. Wire it into your editor, install the Claude Code or Codex plugin, or clone the repository and run this whole chat locally. Your own agent and model can give better answers than the public demo.
{ "mcpServers": { "ose-knowledge": { "type": "stdio", "command": "uvx", "args": ["ose-knowledge-mcp==0.13.30"] } } }One package, zero install. The whole knowledge RAG runs on your machine.
/plugin marketplace add opensourceeurope/ose-ai /plugin install ose-knowledge@ose-aiYou need uv (https://docs.astral.sh/uv/), because the MCP server starts with uvx. The plugin ships more than the MCP server. It also ships an ose-researcher subagent tuned to query the knowledge RAG efficiently and return concise, cited answers.
codex plugin marketplace add opensourceeurope/ose-aiYou need uv (https://docs.astral.sh/uv/), because the MCP server starts with uvx. After this command, run /plugins in Codex and install ose-knowledge, or run codex plugin add ose-knowledge@ose-ai in the terminal. Then start a new session. You get the MCP server and the ask-about-ose skill.
git clone https://github.com/opensourceeurope/opensourceeurope.org.git cd opensourceeurope.org npm ci ollama pull mistral-small3.2 # the default local model ./scripts/knowledge/run-local.shNeeds uvx, npm, python3 and a running Ollama (https://ollama.com). Starts the MCP server, the agent function and this site on localhost, with no API keys. The first run downloads packages and the embedding model (about 500 MB); after that it runs offline. It uses mistral-small3.2 by default; pull it first, because the script won't download it for you. Already have another model? Point the script at that instead, for example OLLAMA_MODEL=llama3.1:8b ./scripts/knowledge/run-local.sh.
Sovereign by design
The entire request path runs on European infrastructure, from the moment you ask a question to the model that answers it. Nothing leaves the EU.
The AI model
Mistral Medium 3.5 (mistral-medium-3.5-128b) writes the answers. The European provider Mistral builds the model, and Scaleway runs it on its Generative APIs in the EU.
Hosting
This page is part of opensourceeurope.org. The site, the serverless function, the MCP knowledge server and the model inference all run on Scaleway (EU), so the whole stack sits with a single European provider.
Your data
Your conversation lives only in your browser's memory. The chat shares your questions only if you accepted analytics in the privacy settings banner. By default nothing is shared. You can change this at any time with the Privacy settings link. If you accepted, the chat function logs, for each answer, how many calls to the AI model it took and whether the docs answered it. When you close the panel or leave the page, it also logs a short summary of what you asked and what the answer said. An automated step removes contact details, credentials and names from the summary, but it can miss something, so do not type personal data. The logs stay in Scaleway Cockpit in the EU for seven days, and an export job copies them to a private storage bucket in the EU. OSE reads them to find gaps in the docs. They hold no IP address and no full text of your questions.