Model Hub

Set up your AI providers once, and let every plugin use them, with the cost of each call recorded.

Every AI plugin asks for its own API key, picks its own model and keeps quiet about what it spent. Model Hub holds your providers and keys in one place. Other plugins ask it for a model by role, such as topic-map.extract, and you decide which models serve that role.

Every call is written to a ledger with its tokens, the listed price and what you were billed. A call that would go over a limit you set is stopped before it is sent. Model Hub also has a small chat panel for the note you are working on, and can let Claude Code, Claude Desktop or Codex use your plugins' tools through a server on your own computer, if you switch it on.

The five words to know#

WordWhat it is
ProviderA place models come from: OpenRouter, Anthropic, OpenAI, Google Gemini, Azure OpenAI, xAI, Mistral, DeepSeek, Groq, Cerebras, Ollama, LM Studio, llama.cpp, LiteLLM, or another server you add.
TierAn ordered list of models with two limits: the most one call may cost, and the most the tier may spend in a month. There are seven built in, from Quick to Frontier.
RoleA job a plugin needs a model for. Each role runs on a tier unless you give it models of its own.
LedgerThe record of every call, kept in a file on your computer.
ToolOne thing a plugin can do when asked, such as adding a book to a shelf. You decide which tools run freely, which ask you first and which are off. See Tools and approvals.

What you need#

  • Obsidian 1.13.1 or later, on desktop. Model Hub does not run on mobile.
  • At least one provider: an OpenRouter, Anthropic, OpenAI, Google, Azure, xAI, Mistral, DeepSeek, Groq or Cerebras account with an API key, or Ollama, LM Studio, llama.cpp or a LiteLLM proxy running on your computer.

Next: Getting started.

Also from Insurgently#

These plugins use Model Hub for their models:

Model Hub is made by a worker-owned workshop, free and open source. About Model Hub