github copilot vision and browser tools are generally available — attach images, pdfs, and navigate live pages

GitHub shipped two general-availability releases on July 1, 2026 that expand what Copilot can perceive and what its agents can interact with. they arrived the same day, from the same changelog, and they fit together: one gives Copilot eyes; the other gives its agents hands.

copilot vision: attach images and pdfs to chat

Copilot Vision is now available to all Copilot subscribers — Free, Pro, Pro+, Business, and Enterprise — with no policy changes or admin action required. it's on by default.

the mechanic is direct: drag an image or PDF into the chat prompt and Copilot reasons about what it sees alongside the code in your editor. the practical use cases are immediate — paste a screenshot of a bug, a Figma frame, a database schema diagram, or a PDF spec, and Copilot can reference the visual context rather than asking you to describe it in text.

for users on Business and Enterprise plans, GitHub retains image and PDF attachments for approximately 24 hours to provide the service. after that window, attachments are dropped from GitHub's systems.

surfaces that support vision at GA: Copilot Chat in VS Code, Copilot Chat on github.com, and Copilot CLI image paths. mobile is not listed in the GA announcement.

browser tools: agents that can navigate live pages

browser tools for GitHub Copilot agents in VS Code are also generally available. the capability set: navigate to URLs, click elements, fill forms, inspect page content, capture screenshots, and validate web apps without leaving the editor.

this closes a specific loop that was awkward before: an agent fixing a UI bug could edit files but had no way to verify the change looked right. with browser tools, the agent can open the page, screenshot the relevant component, and report what it sees — or flag a discrepancy between the code change and the rendered result.

the preview period for browser tools ran through the June 2026 VS Code update cycle. the GA release means GitHub is committing to the API surface and the feature is available without any preview flags or policy exceptions.

what this means for the copilot stack

Copilot started as autocomplete. it became a chat assistant. it became an agent that can edit files and run terminal commands. now it can see images and interact with browsers. each of those expansions took roughly a year to move from experimental to GA. the cadence suggests whatever capability comes after browser tools — operating system-level actions, perhaps, or persistent background tasks (GitHub is watching what happens with Claude Code's headless mode and OpenAI's Ona acquisition closely) — is probably 12–18 months out from production-safe GA.

honest take

vision is genuinely useful. the “paste a Figma frame and ask Copilot to implement it” workflow alone reduces a lot of back-and-forth that happens when visual context exists but can't get into the chat. the 24-hour retention window for Business/Enterprise is a reasonable tradeoff given the regulatory environment around sensitive design assets.

browser tools are more narrowly useful today — valuable for web developers doing automated validation, less so for backend or infra work. but as an infrastructure layer for future agent features (screenshot-driven debugging, end-to-end test generation from live pages), the GA baseline matters more than the current feature list.