"What you sell", in Settings → System → AI over the mesh, lists the
inference servers Routstr forwards to and the models offered from each.
This node's Local Llama is listed by default. "Add a server" takes the
address of any OpenAI-compatible server on the network (llama.cpp, vLLM,
Ollama, NIM; a DGX Spark, say) and an optional API key, checks it answers
with a model list, and adds it. For each model the operator chooses
whether to sell it, the name buyers see, the context and answer limit,
and a price in US dollars per million tokens; the screen previews the
sats a buyer sends up front.
Backend (`routstr_seller.rs`, `routstr.selling.*` RPCs, operator only):
drives Routstr's admin API; Routstr's database stays the one record of
servers, models and prices. Addresses must be http(s) and are refused if
they point at this node, its app network, or link-local addresses. The
context offered is capped at what the server reports, and the local
model's context stays fixed to llama-server's. A model's public ID may
be sold from one server only. Routstr does not hand API keys back, so the
node keeps its own copy (0600) to re-read a server's model list; keys are
never returned or logged.
archy-start keeps the operator's name and answer limit for the local
model and re-states only its context. Image 0.5.0-archy3.
Requires UI/UX UAT approval from the maintainer.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ssmithxandClaude Opus 5.5
12b160b8fb