Local models

A local model runs on your own computer. It is free, your data does not leave your machine, and it works without internet. The trade-off is speed: on a computer without a graphics card, answers can take a minute or more.

Set up

Open Settings, then Local models. Omny shows where you are:

1. Ollama is not installed. Use the download link, install it, and press Check again.
2. Installed but not running. Press Start Ollama. You do not have to: Omny also starts it by itself when you send a message to a local model.
3. Running. You see the models on your computer, and can download more.

Omny starts Ollama for itself and stops it when Omny closes, so nothing keeps running in the background.

Choose a model

Each installed model shows what it can do:

• tools: it can use the agent's tools (read and change files, run commands). Without this, a model can only chat.
• sees images: you can attach images, and it can look at screenshots.
• thinks: it shows its reasoning before answering.

Press Use on a model to switch to it. To get another one, pick a suggestion or type any model name from the Ollama library and press Download. A progress bar shows the download, and you can cancel it.

What works the same, and what does not

Local models get the same agent as cloud models: the same tools, modes, proposals and undo. Omny sizes the context to each request, so a long prompt is not silently cut. The tokens are counted, and the cost is shown as $0.

Web search and MCP servers are services of the model's provider, so they work with Claude and Gemini only.