Ollama Library

Search Ollama's own library from inside the app, find out what a model actually weighs before you download it, and add it as an AI core. Weights files already on your disk go through the same panel.

What you can do

• Search all 200-plus models Ollama publishes, live, instead of picking from a short list somebody wrote by hand
• See a model's real download size, read from the registry ollama itself pulls from, and whether it fits the machine you are on
• Choose which size you want. qwen3 ships in twelve, from 0.6B to 235B, and they are not interchangeable
• Download one into the local ollama store and have it show up as an AI core
• Add a weights file you already have on disk, and be told at the moment you pick it whether anything installed can actually run that format
• Install the ollama engine itself when the machine does not have one yet

Getting started

Open AI Cores, press +, and pick Add Ollama Model. The panel opens beside the list.

If ollama is not installed the panel says so first and offers to install it, with the command shown before it runs. Once the engine answers, the library loads and the search box is live.

Pick a model to see its sizes. Each one carries its exact download size and a verdict about your machine, so a 18 GB model on a laptop with 9 GB free says so rather than failing twenty minutes into the download.

Downloading puts the model in ollama's own store. It then appears in AI Cores, where you set it active. Setting the active core is done there and only there, so there is one answer to the question of where your work runs.

Where the data comes from

The model list and the tag list are read from ollama.com, which is the same catalog the ollama command line reads. There is no bundled copy: a list shipped inside an extension is a list that is wrong by the time you install it.

Exact sizes come from registry.ollama.ai, by adding up the layers in a model's manifest. That is the number of bytes that will land on your disk, not a rounded figure from a description.

If the library cannot be reached, the panel says so and lets you type a model name instead. The name is checked against the registry before anything is downloaded, so a typo is caught immediately rather than after a failed pull.

Requirements

The desktop app, the LLM Capabilities extension for the AI Cores list, and the Ollama Runtime extension, which owns the engine and does the actual downloading. Network access is needed to browse. Once a model is on the machine it runs with nothing leaving it.

What this extension does not do

It does not decide which model answers your work. It adds models; AI Cores routes to them. It does not remove models either, which stays a right-click in that same list.