
Don't invest weeks on assumptions. The Nano Collective's economics charter is plain language, versioned in public, and explicitly not a legal contract. We'd rather answer your question than have you guess what the fund actually is.



Before you write a whitepaper, ask. The Nano Collective's Stage 0 is a sanity check on purpose: a maintainer tells you yes, redirects you to someone already in the space, or says probably not and why. One week of messaging beats a month of whitepaper nobody backs.


95% of users only need convertToMarkdown. get-md keeps node-llama-cpp an optional peer dep so the standard HTML→Markdown path ships without it. The local ReaderLM-v2 model is 1.12GB - opt in with npm install node-llama-cpp.


Most JSON migration tools validate the final shape and give you a diff with no clue where it broke. json-up runs the declared Zod schema after every up() call, so ValidationError carries the exact version and the Zod issues for it.



You can't un-ask for a refactor. Nanocoder treats sessions like save points - /checkpoint create before-refactor, let the AI try it, /checkpoint load to roll back. Two approaches become cheap to compare.


More training data is rarely the fix for a fine-tune that fails. After `nanotune benchmark`, open the markdown report, read the failed tests table, and add examples targeting those exact prompts. Re-train on a smaller dataset that hits the gap.
