Local models on your Macs are the default. When you want a model you can't run locally, connect Anthropic, OpenAI, xAI, Moonshot, or Z.AI, and switch between them from the chat.
With a local model, each request goes to a model server on one of your Macs. With a provider, the conversation you send goes to that provider's API, under your own account or key. r1o never switches to a provider on its own: if the one you picked isn't connected, the request stops with an error.
Type "/model", or ask the agent to switch, and a config card opens in the conversation. Pick a provider, set the model, and apply; the next message uses it. No trip to the settings page.
With an OpenRouter key, some jobs, such as summarising a web page or describing an image, can go to a smaller model while your main model keeps the conversation. Each of those calls shows which model and endpoint it used.
Every model server asmi finds on your cluster, ready to pick
Connect a provider with your account, or paste an API key
Point at OpenRouter or your own server with a base URL and key
A new provider or model takes effect on the next message
Web extracts, vision, and compression can each use their own model
If a provider is not connected, the request stops instead of rerouting