We build AI agents for cyber defence. But sending the data behind it to non-European providers? Not an option for us.
So we build it ourselves. Our colleague Jonathan L. joined the podcast with Nico Puhlmann and Najat Diamante to talk about what that looks like in practice and what we’ve learned along the way.
We believe European tech companies should ship more and demand less. Lukas Brabanski put it well recently: https://lnkd.in/eS5PvhUj
That’s what we’re trying to do: build our own infrastructure, share what we learn, and occasionally pull back the curtain on what works and what doesn’t.
What we covered on the podcast
- AI SOC chatbots for the pre-analysis of security alerts
- automatic enrichment of IPs, hostnames, and events
- human-in-the-loop against hallucinations
- local coding models for sensitive development environments
What we run ourselves
- our own AI servers with 4× GPUs
- models such as Qwen 2.7 and DeepSeek
- an open-source stack with Open WebUI, LibreChat, and MCP
- local coding models at around 100 tokens/second — fast enough for productive use by several developers in parallel
If that sounds interesting, tune in on Spotify: https://open.spotify.com/episode/0tSlcz5i60AQXc0VCatzy6?si=uXhGfNtGSjO3-N1G7vJr-Q&nd=1&dlsi=a91bfeb47b2c4d13