Our focus
Neural Automata is the research identity for F2i’s exploration of local neural computation. The work asks how shared rules, distributed state and explicit learning mechanisms can support memory and sequence processing.
The approach is experimental: build a mechanism, test the information flow, keep the controls and publish what the result actually supports. The public work is early-stage research software, not a production AI product or an academic institution claiming credentials it has not earned.
Two sites, distinct jobs
NeuralAutomata.com explains the ideas, research questions, recorded results and limitations. NeuralCellularAutomata.com is the interactive lab, where visitors can run the supplied checkpoint ports and inspect their state.
The lab uses ZIPP, F2i’s runtime project, to execute Python in WebAssembly and submit bounded compute graphs to the browser. The wider software ecosystem is presented at F2i.
Research status
The current public baseline contains a cellular associative-memory model and a separate causal byte-language model. The original lab runs locally with native PyTorch; the browser lab adds a bounded inference port. The websites distinguish recorded CPU reference results from interactive browser behaviour.
Descriptions on this site are tied to research revision 3393d4a and the bundled ZIPP v0.0.18 release. The upstream repositories may move ahead; a site update should verify new checkpoints and claims rather than silently calling every dependency “latest”.
Open implementation, clear attribution
The research repository and ZIPP publish their own licence and provenance information. This website bundle retains the runtime’s Apache-2.0 licence and notices. The field guide credits the external NCA literature rather than presenting it as F2i’s own result.
Explore the research programme, inspect the source repository, or visit F2i.