
Bounty terms belong in the issue, not the invoice. Nano Collective's Economics Charter says it: deliverable, acceptance criteria, and amount written down before work starts. Nothing open-ended. Nothing retrospective. https://nanocollective.org



Sponsorship buys a logo on the wall, not a vote on the roadmap. No input on what gets built. No input on who gets paid. No exclusivity, no early access to unshipped work. Same rule for $5 tiers and bespoke deals. https://nanocollective.org



get-md v1.5.0 ships a pluggable LLM backend. Set llm.sdkProvider to openai-compatible, anthropic, google, or local-llama, and --use-llm routes through it. API keys support ${ENV_VAR} substitution, defaults via ${VAR:-fallback}. https://github.com/Nano-Collective/get-md



json-up defaults to _version, but the `key` option lets you point at any field. Data tracked as version, schemaVersion, or anything else migrates in place, no renames required. https://github.com/Nano-Collective/json-up



Nanocoder ships three tool-calling paths: native function calling, an XML fallback, and a JSON fallback. If output is malformed, the parser repairs it instead of failing the turn, so small or older models still work end to end. https://github.com/Nano-Collective/nanocoder



`nanotune export` ends a fine-tune: fuse the LoRA adapter, convert to GGUF, quantize to q4_k_m, q8_0, q4_k_s, or f16. The output runs in llama.cpp, Ollama, or any local runtime. https://github.com/Nano-Collective/nanotune
