§01 New
- Local models. + › Local model lists the models on your Mac and popular ones to download (Qwen 3.5, Gemma 4, gpt-oss and more), with each one's size and whether it runs well on your Mac's memory. Pick one and Start: it opens in Claude Code's terminal, on that model instead of the cloud.
- Ollama, set up for you. If Ollama isn't installed, one click downloads Ollama's signed build into ~/.local (no password) and runs it on your Mac only, with the 64K context coding agents need. An Ollama app you already run works as it is.
- The whole Cadence experience. Local agents get a dot on the toolbar, their questions in the Inbox, titles, your rules, and Talk ("start gemma locally on my project…").
- Settings › Agents › Local models shows Ollama's status and lets you download and remove models.
Local models are free and private, and less capable than Claude: start them on smaller tasks.