AI Tools

Every tool here runs as a Program-as-Weights neural program inside your browser via WebAssembly. Your input is never transmitted — inference happens on your own CPU. The only network traffic is the one-time download of the weights themselves: the shared interpreter from Hugging Face, and each tool's program from programasweights.com. Both are cached on your device afterwards.

The interpreter is GPT‑2 124M, which is small. These tools are good at fuzzy judgement calls and will sometimes be wrong — check anything that matters. Tools that produce structured output are verified in the browser with real parsers, shown beneath each result.