
Whitepaper proposers go dark sometimes. The collective's answer: anyone can pick up an in-flight whitepaper. If nobody does within 60 days, it moves to Paused. The work does not stall on one person. https://nanocollective.org


Every project the Nano Collective ships has passed the same three checks, in order: privacy-respecting, local-first, open for all. The order matters. They are not a list of nice-to-haves. They are the filter from whitepaper intake to release. https://nanocollective.org


get-md's chunkMarkdown splits a converted doc at heading boundaries and prepends the heading trail to each chunk, so the model knows where it sits. An `overlap` option keeps answers that straddle a boundary in the retrieval set. https://github.com/Nano-Collective/get-md


migrate() reads and writes _version by default, but the `key` option lets you point it at any field name (schemaVersion, revision, v) so the version lives where your data already keeps it. https://github.com/Nano-Collective/json-up


MCP servers through one config: .mcp.json in the project root. stdio, HTTP, or WebSocket per entry, alwaysAllow trims the prompts, env vars override everything so secrets stay out of the commit.

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



nanotune export fuses your LoRA adapter with the base model and writes GGUF in f16, q8_0, q4_k_m, or q4_k_s. Pick the quality-vs-size trade-off for the hardware it'll actually run on.

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