
How a Nano Collective project actually comes to life: Spark, Whitepaper draft, Public review (30 days), Build approval, Build and ship. Merging the whitepaper is not endorsement. The build call is its own gate.

https://nanocollective.org



Sponsorship funds the Nano Collective. It does not buy roadmap influence, early access to unshipped work, exclusivity, slot caps, or sponsor-directed work targeting a specific contributor. Same rule for every sponsor.

https://nanocollective.org



get-md's chunkMarkdown splits at heading boundaries, keeps a heading trail on each chunk, and supports overlap so RAG answers that straddle a boundary still surface. The RAG half of the pipeline. https://github.com/Nano-Collective/get-md


Each `up` function in json-up is typed from the previous step output. Rename a field in v2 and downstream references break at compile time. Type safety travels with the chain, not just the schemas. https://github.com/Nano-Collective/json-up


Nanocoder's main agent can call the `agent` tool up to five times in a single response. The built-in `explore` subagent and others fan out in parallel, each with its own progress row, so research doesn't stall the main thread.

https://github.com/Nano-Collective/nanocoder



Nanotune's loop: run `nanotune benchmark`, open the markdown report, find the failed tests, add training examples targeting those failures with `nanotune data add`, retrain. Each pass closes a gap the benchmark actually measures.

https://github.com/Nano-Collective/nanotune
