Gemini goes desktop on Mac
Google is beta‑testing a native Gemini app for macOS with 'Desktop Intelligence' to access files and system context, and rolled out tools like Stitch for AI‑driven UI generation — a clear push to own contextual, desktop‑level RAG and design workflows. That shift blurs lines between search, agents and personal productivity and raises new integration and privacy questions. (ad-hoc-news.de) (t3n.de) (googlewatchblog.de) (juliangoldie.com)
Bloomberg reported Google began privately sharing an early macOS Gemini client, internally codenamed "Janus," with external testers on March 19, 2026. (bloomberg.com) App researcher M1Astra’s inspection of the test build revealed a feature labeled "Desktop Intelligence" that, with user permission, can read on‑screen content and pull context from apps such as Calendar. (bloomberg.com) Testers were given a trimmed early build that reportedly supports image, video and music generation, tables and charts, math problem solving and document/media analysis while Google declined to commit to a public release date. (bloomberg.com) On March 18, 2026 Google Labs published the Stitch "vibe design" update, introducing an infinite AI canvas that generates high‑fidelity UI from natural‑language prompts. (blog.google) Reports show Stitch now includes production‑oriented exports and integrations — including Figma and React export paths — plus a new "Antigravity" element to handle logic and backend wiring for generated interfaces. (geeky-gadgets.com) Google’s Gemini 3 developer documentation added parameters this week to control latency, cost and multimodal fidelity, enabling developers to tune model behavior for desktop RAG and productivity integrations. (ai.google.dev) Bloomberg and MacRumors noted the move follows Mac desktop apps from OpenAI (ChatGPT) and Anthropic (Claude), and Google told testers the Janus build "will have only critical features" as it collects feedback. (bloomberg.com)
