Poor Richard Almanack Bot
A NOOA agent that answers factual questions from the Poor Richard almanack — a curated set of offline Python reference libraries (ISO codes, unit conversions, business calendars, checksums, …). Every answer is a verified golden value or a computation from one of those libraries — never model memory.
The LLM runs through HF Inference with your own account: on a Space, sign in (the inference-api scope is requested); locally, the app uses your machine's own HF token (HF_TOKEN or hf auth login). Pick a model and provider, ask, and watch the activity feed stream the agent's turns (generated code + execution output) as they happen. Provider auto routes via your HF Inference Providers settings; pin a cheaper one for a model (e.g. Qwen/Qwen3.8-27B currently runs free on OVHcloud) by picking that provider.