
60 days is the silence limit on a Nano Collective whitepaper. After that it's labelled Paused and anyone can pick it up by adding their name to the file. "No-one's driving this" stops being a worry and becomes a known handoff window.


Every "why was this built" thread on an open source repo dies in someone's DMs. At the Nano Collective, whitepaper decisions - yes, no, paused, declined - are written on the page with dates and rationale. Future-you won't have to ask.


Converting a docs site to markdown is not the HTML. It's the crawler: sitemap indexes, recursive globs, runaway URL counts. `get-md --sitemap` walks sitemap.xml, honours include/exclude, caps depth and URL count. One command, one folder of markdown.



Most migration libraries make you pick a side: all sync, or all async, for the whole chain.

`json-up` doesn't. `createAsyncMigrations` lets sync and async `up()` functions sit in the same chain. Only the steps that actually need to await pay the async cost.


600 tokens of system prompt, and a 1B local model never gets to its first file. `nano` profile in Nanocoder cuts it to ~200 by dropping three tools, two prompt sections, and AGENTS.md. The difference between doing work and running out of context on the prompt alone.



Fine-tuning usually means a rented GPU, a YAML config, and a half-day job. Nanotune runs LoRA over MLX on Apple Silicon through guided prompts, then writes a GGUF on the way out. The laptop on your desk is the stack.