How routing decides

Routing picks the right model per question. The rules are written to be read - this page is the whole story, and Settings → Routing shows the same rules in the app alongside a live list of recent decisions.

The modes are the consent

• Auto - Offline Only: never touches the internet. The mode name is the promise.
• Auto - Online and Offline: may use online models, by the rules below.
• A specific model: no routing at all - that model answers everything.

Each AI has its own mode, set in its editor.

The rules

1. Health stays home. A question about your health routes to your device's models even in the online-and-offline mode - with a medical model preferred when installed.
2. Current information can go to the web. Questions that need fresh facts can use a web-search model, with cited sources and live progress shown while it researches.
3. Hard questions can escalate. Difficult reasoning, coding, or math can step up to a stronger model - on-device first when your hardware carries it, online when allowed and warranted.
4. Fit-aware picks on device. Among installed models, routing prefers what runs fully on your graphics card.
5. Projects use tool-capable models only - with their own picks for agent work and helper tasks.

The reply always says why

The Model button under every reply names the model that answered, where it ran (on device, online, your server), and the routing reason - with one-click second opinions: Redo on your device or Try this answer online. Never a silent substitution.

Your levers

Settings → Routing:

• On your device: prefer fastest or strongest among installed models.
• Online eagerness: privacy-first to freshness-first.
• Per-category online picks: which online model handles current information, hard coding and reasoning, other hard questions - and, with projects, agent work and helper agents - prices shown up front.
• Working on projects: keep simple side-work on this device, or keep whole project sessions on-device when your hardware fits them.