Skip to content
Braver New WorldBNW docs

Sharing a model

If Ollama, LM Studio, or llama.cpp runs on your computer, the node can find its models and share one in a single step:

Terminal window
bnw share models                               # models served here, and whether they're shared
bnw share model gemma3:4b --access contacts    # publish, connect, answer automatically, offer
bnw share model gemma3:4b --access anyone --max-per-hour 10   # sharing again updates it
  • What sharing does. It publishes the model as an inference.model capability and connects it to the runtime’s OpenAI-compatible chat endpoint (egress local). It sets the autopilot to auto (or --mode ask) with 30 calls per person per hour, sets who may use it, and signs a week-long offer that renews itself. It is all the steps below, with defaults.
  • Detection. The node probes only the runtimes’ fixed loopback addresses (127.0.0.1:11434, :1234, :8080), with short timeouts and bounded replies. Ollama models that list capabilities without completion, such as embedding models, are skipped.
  • Only local endpoints. One-step sharing accepts only an http chat endpoint on this computer’s loopback interface; pass --url for a runtime on another port. Remote or paid endpoints go through the full setup below, which asks about egress and secrets.
  • In the web client. My capabilities → Share a model on this computer does the same, with a choice of who can use it.