← Todos os guias
Guia do Playbook · Grátis

Use Ollama Models, Chat, and Embeddings Through Your Muse

Verificado 2026-11-18

Through an AnythingMCP custom connector, your muse can talk to open models such as Gemma, Llama, Qwen, Mistral, and DeepSeek on your own Ollama server or Ollama Cloud, so your words go to your server instead of a hosted provider.

What it does

Connected this way, the muse gets 7 tools; it cannot change what is installed on the server. It lists the models on the server with family, size, and format, and shows which are loaded now. It reads one in detail: what it can do, including text, vision, tools, thinking, and embeddings, plus its context length, so the muse can check a model fits the job. It holds a chat with a model, with a system prompt, history, and images for vision models, and answers one plain prompt, including fill-in work on code models. It turns text into embedding vectors.

How to use it

  1. Install Ollama, pull at least one model, and note the server address. Leave the key field empty for your own server.
  2. For Ollama Cloud instead, set the address to the Ollama site and paste a key from its settings page.
  3. In AnythingMCP, install the Ollama connector and paste the address in. In Muse, open Settings, then Connectors, add a custom connector with the address AnythingMCP gives you, and approve.
  4. Ask in plain words: which models are on my server, and ask the small one to summarize this note.

What you need

  • A Muse account that can add custom connectors
  • An AnythingMCP workspace, on Cloud or self-hosted
  • An Ollama server with a model pulled, or an Ollama Cloud account and key

What to watch for

  • AnythingMCP Cloud cannot reach a server on your own computer or home network. Use Ollama Cloud, a publicly reachable server, or a self-hosted AnythingMCP, where the host must be allowed in settings, because private addresses are blocked.
  • Answers are not streamed, and the first call to an unloaded model loads it, which takes from a few seconds to about a minute on large models. A local server costs nothing to run; Ollama Cloud use counts against your plan.
  • A missing model returns a not-found error until you pull it on the server. A model without a capability, such as tools or embeddings, refuses that kind of call; the detail view shows what each model supports. The connector never pulls, copies, creates, or deletes models.
  • This is a custom connector, and Meta does not review custom connectors. You can disconnect it in Settings at any time. The muse prepares, you decide.

Experimentou este guia?

Conte para a comunidade como foi, ou peça ajuda para fazer funcionar.

Fale sobre isso no MuseFM →

Playbook Pro

Quer pular as lições?

Ajude o seu agente a aprender tudo de uma vez. 1,015 habilidades, um passe, $49 por 90 dias.

Obtenha o Pro Pass
Mais guias →