What Is Model Routing in a Desktop AI Assistant?

Model routing means the assistant picks the right AI model for each request instead of sending everything to one fixed model. In Jarvis (getjarvis.eu), a screen-aware desktop assistant for macOS 12+, Windows 10+, and Linux, routing sends careful writing and code to a frontier model, fast general turns to a faster model, and long-context or image tasks to a frontier model. You summon Jarvis with Cmd+/ on Mac or Ctrl+/ on Windows and just ask; the routing is invisible. The benefit is quality and cost balance: hard problems get a strong reasoning model, trivial ones get a fast cheap path, and you never manage it. This is different from ChatGPT or Claude.ai, where you pick one model per conversation. Jarvis abstracts that away across three vendors. It starts with a free plan (40 requests/week), then $16/month, and never trains on your prompts, screenshots, or connector data.

In a single-model app you choose, say, a faster model at the start of a chat and every message goes there, even a one-word factual lookup. Routing flips that: the assistant inspects each message and dispatches it to whichever model is best for that specific job. Jarvis does this across frontier models from Anthropic, OpenAI, and Google. A code question lands on the strongest reasoning model, a 300-page PDF on the widest-context one, and a quick rewrite on a faster one. Because the choice is per-message, a single conversation can quietly use several models, each playing to its strengths.

Desktop assistants face wildly varied requests, summarize this Slack thread, read this Figma screen, draft this Gmail reply, refactor this function. Those have different ideal models. A multimodal screen read benefits from a faster model's vision; a sensitive email benefits from a frontier model's tone control; a fast factual answer benefits from a faster model's speed. Routing lets one floating bar handle all of it well. Without routing, you'd compromise: pick a vision model and lose writing quality, or pick a writing model and lose long-context. Jarvis avoids that tradeoff entirely.

This page is available in the product site but is intentionally excluded from search indexing.

AI models