Ollama Runtime

Runs models on your own machine through Ollama. It serves the embedding models Vectors uses to build an index, and the Ollama models listed as AI cores in LLM Capabilities. It has no panel; a status bar item shows whether the engine is up.

What you can do

• Embed your text with a model on this machine, so Vectors can build a semantic index without sending your files to an embedding service.
• Chat with an Ollama model by making it the active core in LLM Capabilities.
• Install Ollama from inside the app, with the install running in a terminal you can read and interrupt.
• Pull a model from inside the app, with the download visible in a terminal.
• Remove a model from Ollama's store to free the disk it takes.
• Check from the status bar whether the engine is running.

Getting started

1. Install Ollama yourself, or open LLM Capabilities, choose Add Ollama Model under AI Cores, and use Install Ollama there.
2. Pull an embedding model. From a terminal that is ollama pull nomic-embed-text.
3. In the Vectors settings, set the embedding source to local, the model to nomic-embed-text, and the dimension to 768.

Once Ollama is installed, an Ollama item appears in the status bar. Click it to see the models on this machine, or to open AI Cores when the engine is stopped.

Requirements

• The desktop app.
• Ollama on the machine. On macOS the in-app install uses Homebrew, and on Linux the vendor's install script. Otherwise it points you to the download page.
• LLM Capabilities and Ollama Library, for the in-app install and for chat through AI Cores.

Good to know

• Ollama starts at launch when nothing is answering. Turn off the auto-start setting to be told instead.
• Stopping only stops an engine this extension started. One you started yourself is left alone.
• Starting a model never downloads one. A missing model is refused with the fix.
• Starting the engine, installing Ollama, and pulling or removing a model all go through Studio's approval rules when an assistant or a flow asks for them.
• There is no reranking or transcription, because Ollama exposes no such route.
• The engine is reached at 127.0.0.1, port 11434, unless you change it in settings.