ollanet docs

Introduction

ollanet finds Ollama hosts you configure or select, manages models on those machines, and talks to them through a CLI, MCP server, or Node library.

Use the CLI at a terminal, MCP inside an agent, or the Node library in your application. All three work with the hosts you configure. Node 20+ · zero runtime deps. Prompts run on the selected Ollama host. That may be this computer or another host you configured.

Three jobs

  • Discover Ollama on localhost, config, Tailscale, or an optional LAN scan
  • Manage with pull / show / rm / ps on a named machine
  • Use prompt, hash-addressed chats, compare, bench, and the same inventory over MCP

A reachable host is not a trusted host. --lan stays opt-in. See Network for what each command sends.

It is not Open WebUI. No browser chat product: discovery, talk, and continue.

Finetuna

ollanet is the network client. Finetuna can shape a GPU-tuned named variant on the machine that runs Ollama. The pairing is optional. Each tool works alone.

  1. On the host, if you want a named variant: finetuna → something like gemma4-ctx32k
  2. From a client: ollanet scan → ollanet prompt that-host gemma4-ctx32k "…"

You can skip Finetuna and prompt any model already on the host.

Next