The Clear Box Stack
Clear Box builds tools and an open attribution framework for work made by people, AI, or both. We help make contributions easier to credit, disclose, and stand behind.
The Clear Box Principle
AI is changing how work gets made. The response so far has been to hide how it works or try to detect when it was used. We think there's a better answer.
Our clear box infrastructure is open only to you and anyone you explicitly let in, so you can follow every step of your workflows and decide what data, if any, ever leaves. That principle guides everything we build. Transparent to you. Closed to everyone else.
Our Open Work
Clear Box is building a layered attribution framework, a private inference runtime, and the platform that ties them together. All designed around the same commitment: work you can see on infrastructure you control.
A practical attribution framework for recording who contributed to a work and how, across human and AI collaborators alike. DARP is being refined in the open, with feedback welcome and encouraged to achieve community consensus.
Explore DARPA private, local-first runtime for AI models on your own hardware, engineered to fit capable assistants on modest machines and pool several into a swarm when you need more. No cloud dependency, nothing leaving your machine.
Explore commonllamaThe workshop where the stack takes shape. Build frameworks, measure how they perform, and test them under duress. commonllama runs the models underneath, and DARP provides the structure for contribution provenance at every step.
Explore commonFrame† commonFrame is licensed AGPL-3.0 with an added exception. Full exception terms are still being finalized. Check back soon.
An Open Door
Infrastructure you can see into is infrastructure you can learn from. We make private, accountable work possible on hardware you own with workflows you can inspect.
We welcome anyone thinking about what it means to use AI honestly, whether that's in a classroom, a studio, a library, a lab, or a living room. If you're thinking about AI literacy or local-first work, we'd love to build with you.
From the Workbench
Updates
Universities are spending millions to give students AI and millions more on tools to catch them using it. We think the answer is attribution, not detection.
Continue ReadingEvents
