GitHub Copilot can now control desktop apps—even ones with no API
The preview controls GUI-only tools no API can reach. Plus Droid's crash-resilient agent sessions and six Copilot models retiring Oct 19.
Copy markdown
Copilot now clicks through apps with no API
In public preview, GitHub Copilot computer use lets the Copilot CLI (`/computer on`) and the Copilot app on macOS and Windows read, click, type, scroll, and navigate desktop apps for you—aimed at GUI-only, legacy software with no API, CLI, or MCP. Every app is approval-gated, macOS prompts for Accessibility and Screen Recording permissions, and orgs can disable it entirely. Think expense-portal automation or end-to-end GUI tests without writing a single selector.
Factory's Droid gets crash-resilient sessions
Factory shipped Droid CLI v0.233.0 and Desktop v0.190.0 on Oct 3: you can now resume a session and append new messages in one command, and long script output is trimmed from the middle so the agent keeps the start and end it needs instead of a hard cut-off. Also fixed—cut-off replies, oversized requests, malformed tool arguments, and session history after reload. Fewer dead agent sessions on long jobs.
Six models leave GitHub Copilot on Oct 19
Mark the date: Gemini 3.7 Flash, GPT-5.5, GPT-5.4, GPT-5.4 mini, GPT-5 mini, and Grok 4.5 are deprecated across every Copilot surface—chat, inline edits, ask and agent modes, and code completions—on October 19. Replacements (Gemini 3.8 Flash, GPT-5.6 Sol/Luna, Grok 4.6) auto-enable for Business and Enterprise unless admins pinned the defaults. If your tooling hard-codes any of these, switch before the cutoff.