About
Making AI smarter, safer, and more efficient.
We're an AI research lab doing the hard work that makes intelligent applications possible. We design protocols, generate training data, research efficient architectures, and build products that prove it works.
Most of our research is published openly — because the future of AI should be built together.
What we do
Four pillars, one mission.
Three lines of research, and one commitment under all of them: publish the work.
Protocol Research
Designing foundational protocols that enable AI-native applications — Virtual DOM, Virtual Browser, W0rkTree, and more.
- Virtual DOM & Virtual Browser
- W0rkTree & Model Context
- Open specifications for all
Synthetic Data Generation
Creating high-quality synthetic datasets that improve model performance. Better training data means smarter, more reliable AI systems.
- Code generation datasets
- UI/UX training data
- Billions of tokens, validated
Model Efficiency
Researching efficient architectures, fine-tuning methods, and inference optimization. Making models smaller, faster, and more capable.
- Tool-use-preserving distillation
- 50x context compression
- Less compute, better results
Open Research
Under all threeMost of our research is published openly. Papers, datasets, model weights, and protocols — available for the community to build upon.
- Public research registry
- Open model weights
- Reproducible pipelines
Research that ships
Our products are our proof.
Each one began as a testbed for a protocol. If the research doesn't survive a real product, the spec isn't finished.
Milestones
A busy year, publicly logged.
Apr 2026
W0rkTree goes open source
The Rust version control core ships publicly under GPL 3.0, spec included.
May 2026
Tavoc reaches 1.0
The Rust-native code platform goes stable after a year of betas.
Jun 2026
Nokuva enters public beta
The AI-native design editor opens to all teams, protocols built in.
Jul 2026
Research registry goes live
Every spec, dataset, and paper — catalogued in one public registry.