
Sponsorship funds the work, it does not direct it. That is the rule: no influence over what gets built, who gets paid, or which contributors are prioritised. https://nanocollective.org



Every Nano Collective project README opens with the same single sentence and the same link to the collective. The wording is exact, on purpose, so four different projects still read as one community.

https://nanocollective.org



get-md v1.5.0: the new `llm` config block routes `--use-llm` through a chosen backend. Local ReaderLM-v2 by default, or any openai-compatible / Anthropic / Google endpoint, with ${ENV_VAR} for keys.

https://github.com/Nano-Collective/get-md



json-up's `key` option points the version tracker at any field you choose: `version`, `schemaVersion`, whatever your data already uses. Adopt json-up without renaming on-disk data. https://github.com/Nano-Collective/json-up



Nanocoder runs your data, models, and processing on your machine. Pair it with a local model and the whole agentic loop stays off the network. Local-first, by design. https://github.com/Nano-Collective/nanocoder



Trained a LoRA with nanotune? `nanotune export` fuses it back into the base model and writes a GGUF file, quantized to f16, q8_0, q4_k_m, or q4_k_s. Drop into llama.cpp or Ollama. Pre-built binaries, no compile step.

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