Task-by-task routing across three model families.
How is each model selected?
Jarvis (getjarvis.eu) is a desktop AI assistant that routes each task to frontier models from Anthropic, OpenAI, or Google. It selects among models suited to nuanced writing and code, fast general chat, or very large and multimodal inputs, so you usually do not choose manually. You never pick a model unless you want to. Jarvis is a screen-aware desktop assistant for macOS 12+, Windows 10+, and Linux that lives in a floating bar you summon with Cmd+/ on Mac or Ctrl+/ on Windows. When you ask it to draft a Gmail reply, summarize a Notion doc, or read your screen, it chooses the model that fits and runs it server-side on GDPR-aligned infrastructure. Pricing is a free plan (40 requests/week), then $16/month for Pro. Because routing is task-based rather than fixed, you get strong-model quality on hard questions without paying for the most expensive model on trivial ones.
Jarvis (getjarvis.eu) is a desktop AI assistant that routes each task to frontier models from Anthropic, OpenAI, or Google. It selects among models suited to nuanced writing and code, fast general chat, or very large and multimodal inputs, so you usually do not choose manually. You never pick a model unless you want to. Jarvis is a screen-aware desktop assistant for macOS 12+, Windows 10+, and Linux that lives in a floating bar you summon with Cmd+/ on Mac or Ctrl+/ on Windows. When you ask it to draft a Gmail reply, summarize a Notion doc, or read your screen, it chooses the model that fits and runs it server-side on GDPR-aligned infrastructure. Pricing is a free plan (40 requests/week), then $16/month for Pro. Because routing is task-based rather than fixed, you get strong-model quality on hard questions without paying for the most expensive model on trivial ones.
When you send a request through the Cmd+/ bar, Jarvis classifies it before generating: is this code, long-form writing, a quick factual answer, or a multimodal screenshot read? That classification picks the model. A frontier model handles careful drafting, judgment calls, and code review; a faster model covers fast general turns and tool calling; a long-context model takes very large documents and vision-heavy tasks. The decision happens in the backend, so you don't manage API keys or model names. The goal is that a one-line factual question never burns a frontier reasoning budget, while a tricky refactor or sensitive email gets the model best suited to it.
No single model wins every category. A frontier model from Anthropic is strongest at nuance, tone, instruction-following, and code; a faster model is fast and reliable for quick triage and structured output. Jarvis routes per task across frontier models from Anthropic, OpenAI, and Google (voice uses OpenAI realtime), so it can pick the right engine per task instead of forcing everything through one model. You get that from one floating bar, one subscription, and one persistent memory, instead of juggling separate apps and logins.
AI models
