Efficient by design
Tools that fit constrained machines: small binaries, clear APIs, and no dependency stacks that outgrow the problem.
Research · Tools
Tools for building intelligence where compute is scarce
Research software we build and share: frameworks and utilities for training, experimenting, and shipping models without assuming a hyperscale stack.
How we approach tools
The same bar as the rest of our research: local, accessible, and useful enough for real work on ordinary machines.
Open to inspect
Code and ideas others can read, fork, and teach from, so progress in low-resource settings is shared, not locked away.
Built for builders
Frameworks and utilities for researchers and engineers shipping models, demos, and systems that have to run locally.
Part of the stack
Tools sit beside data, models, inference, and benchmarks, so the path from experiment to deployment stays coherent.