From screen to lab bench.
These are the questions we work on now, and where we want to take them.
Agents that carry out real work. Not chat, but tasks done end to end: reading mail, screening tenders, writing and checking code, with a person signing off.
Voice and generative media. Real-time voice that holds a conversation, and generated images and video.
Small models that judge. Fast, cheap models that classify and check, so a large model only reads what matters.
Next: science, chemistry and robotics. The same methods pointed at the physical world: lab work, materials, and robots that learn a task from a description.
What we build.
Each line links to one example on our work page.
Apps for iPhone, iPad and Mac. Native apps from first sketch to App Store release, including real-time voice and AI on the device.
Example: Tonic, on the App Store
Consumer products. Products for everyday people, with onboarding, subscriptions, store pages and the numbers behind them.
Example: Call Santa Claus Now
Prototypes for the public sector. Working prototypes for public challenges: document analysis, evidence and tender screening, with a source kept for every finding.
Example: a location-history prototype for municipal investigations
Tools for developers. Open-source tools for people who build with AI agents.
Example: Context Lens
We take on outside projects when they fit what we explore: public bodies, companies and research partners.
AI that shows its work.
What we build keeps the source of every finding, says when it is unsure, and lets a person overrule it. Work for others keeps its data in the EU.
- Sources.
- Every answer points back to the document or data it came from.
- Uncertainty.
- When a model is not sure, the work goes to a person instead of guessing.
- Control.
- A person signs off on what matters, and can see why the system decided.
Tell us what you are working on.
Based in Haarlem, the Netherlands.