About KRXY

KRXY is an independent experimental research and engineering project. It is not a company, a cloud product, or a claim that decentralised training already works in general.

The project asks whether artificial intelligence can learn and operate across independently run edge devices, starting with Raspberry Pi hardware. The public record is the development journal. This website is the long-term home for that record, for architecture notes, and for citable experiment pages once the evidence is attached.

The journal is explicit about sequence. Part 1 was a plan, written before hardware had been tried. Part 2 records one commissioned Raspberry Pi node. Parts 3 and 4 record identity, heartbeats and the beginning of model-aware scheduling. Part 4 also says a second physical node and distributed-training results belong to later entries. Part 5 measures the answers of three model deployments on that same Pi before choosing a default. It is still one machine, and it is not a training result. Part 6 adds a second, CPU-only Raspberry Pi. Two independent requests could then run at once, roughly doubling batch throughput, but each answer was still generated on one machine, and no training was run.

Running an inference job on a node is not the same as training a model across a network. This site keeps those categories separate: what has been built, what a narrow measurement showed, what is still being investigated, and what remains a longer-term ambition.

KRXY is the public name of the project. Individual contributor profiles can be added later by an administrator. None are published by default, and no registered organisation is implied.