Ollama
Ollama runs local models from the command line. It's lightweight and reliable — and it's what vault-rag uses for embeddings, the small model that powers semantic search.
Steps
- Install — download from ollama.com, or
brew install ollama. It runs as a background service. - Pull an embedding model — for vault-rag's semantic search:
(vault-rag's default isollama pull nomic-embed-textqwen3-embedding— pull whichever your index was built with; see connect: vault-rag.) - Pull a chat model — only needed for vault-rag's chat feature:
ollama pull llama3.1 - The endpoint — Ollama listens on port 11434, OpenAI-compatible. See endpoint & model.
Common pitfalls
- Enter the endpoint without
/v1(http://localhost:11434) — the plugin adds it. See the /v1 trap. - The embedding model in your plugin settings must match the one your index was built with — mismatched models give nonsense results.
→ Want it on your phone too? Use it mobile.