
How a Nano Collective project gets built: Spark the idea, draft a whitepaper, 30-day public review, build approval with rationale, then ship. If the proposer disappears, anyone can pick it up. https://nanocollective.org



The Nano Collective is fiscally hosted by Open Source Collective. Every incoming payment and every payout shows up on the public OSC ledger in real time. The transparency is structural, not a promise. https://nanocollective.org



get-md v1.4.0: `node-llama-cpp` is now an optional peer dep. The standard HTML→Markdown path ships light (no model, no native bloat). Opt into the on-device ReaderLM-v2 path only if you want it. https://github.com/Nano-Collective/get-md



When a migration's output doesn't match its schema, json-up's ValidationError hands you the full Zod issue list: path and message for every field that broke. No guesswork about which step wrote the bad shape. https://github.com/Nano-Collective/json-up



Nanocoder custom tools: a markdown file in .nanocoder/tools/ with a YAML schema and a shell body. The model can call it, you confirm, the script runs. No MCP process, no build. https://github.com/Nano-Collective/nanocoder



Build multi-turn chat training data interactively in Nanotune, turn by turn in the TUI. Then `nanotune data validate` catches duplicate examples, broken role alternation, and context-message drift before you burn a run.

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