Anthropic models stay in the routing mix.

Claude stays: Jarvis (getjarvis.eu), a desktop AI assistant, keeps routing tasks to Anthropic models. When a task plays to Claude's strengths in writing or reasoning it uses Claude; when another model fits better it switches, with no manual choice. What changes is the surface — a floating bar over any app with Cmd+/ on macOS or Ctrl+/ on Windows, screen awareness, and 30+ connectors like Gmail, Slack, and Linear. The real change versus Claude desktop is that Jarvis is screen-aware and connects to 30+ apps like Gmail, Slack, Notion, Linear, and GitHub, so it acts rather than only chats. GDPR-aligned in GDPR-aligned infrastructure, OAuth tokens AES-256-GCM encrypted, and never trains on your data. A free plan (40 requests/week); Pro $16/month, Unlimited $32/month.

The worry behind this question is fair: people who love Claude desktop do not want to trade away the model to gain features. With Jarvis you do not. A frontier model is part of Jarvis's routing, so the prose quality and reasoning you switched to Claude for are still available. The difference is that Jarvis treats Claude as the best tool for some jobs rather than the only tool for every job. For writing-heavy or nuanced tasks, expect Claude-grade output from the same floating bar you reach with Cmd+/ or Ctrl+/.

Single-model assistants like Claude desktop are excellent until you hit a task another model handles better, fast factual lookups, certain coding patterns, or long-context jobs where frontier models from Anthropic, OpenAI, or Google shine. Jarvis routes per task, so you get Claude's strengths without its blind spots, and without manually swapping tools. You cannot, in Jarvis, pin every single response to Claude the way a Claude-only app guarantees, that is the honest trade. In practice the routing favors quality, so writing tasks land on Claude anyway, and you gain coverage elsewhere.

Comparisons