Build local intelligence in public, together.
Cerebral Chips is an open-source platform for people who want to inspect, improve, and share the building blocks of useful local AI.
Proton NPU is public: source, Apache-2.0 contributions, upstream credits and functional RTL evidence
Local intelligence should be shaped by the people who build and depend on it.
Cerebral Chips exists to make the path from local-AI idea to verified system more open. The platform belongs to a community of engineers, researchers, makers, and users who can question it, improve it, and take it into the places where it matters.
Inspect the hardware. Reproduce the run. Improve the next version.
An open hardware reference platform that runs scalar, RVV 1.0 vector and custom INT8 matrix instructions in one bare-metal program, verified in RTL simulation.
The hardware branch includes the integration, bare-metal C examples, verification gates and source-linked records. Original contributions use Apache-2.0, with upstream notices preserved.
Useful systems can precede custom silicon.
Proton NPU is the working reference platform for scalar, vector and matrix integration. The longer-term Proton LPU research explores quantization, memory movement and language-model execution beyond this baseline.
Candidate specifications need status and rationale.
Published architecture material should separate proposals, evaluated options, selected decisions, implemented behavior, and verified evidence.
Implementation follows a reviewable dependency and license policy.
Source publication is intended where ownership, upstream obligations, documentation, build instructions, and validation status are clear.
Results require workloads, methods, and reproducible evidence.
No benchmark is published without the model, configuration, hardware or simulator state, measurement method, limitations, and an inspectable evidence path.
Every useful perspective can help strengthen the shared foundation.
Start with Proton NPU’s contributor guide and a passing baseline. Useful contributions include reproducible bug reports, verification cases, bare-metal kernels, architecture review and clearer documentation. Larger workload and architecture proposals are welcome through the collaboration channel.