Attach Screenshots, PDFs and Scripts to a Chat
October 19, 2026 · InfraScout Team
Some of what you need the AI to look at isn't on a host or in your tenant. It's on your screen. You can now attach screenshots, PDFs and text files to a chat message, and the AI reads them in the same turn it uses to check your hosts and cloud.
Show the AI what you're looking at
Three situations come up all the time. A user sends you a screenshot of an error dialog that has no event ID and no log line you could search for. A vendor's knowledge-base article, saved as a PDF, describes a fix and the conditions it applies to. A colleague hands you a PowerShell script and asks you to run it on the domain controllers, and you want it reviewed first.
Until now, each of these meant retyping the error, pasting excerpts of the article, or pasting the script into the message box and hoping it fit. Now you attach the file itself and ask your question:
This script is supposed to clean up stale computer accounts. Before I run it on
our domain controllers, walk me through what it changes, flag anything risky, and
check which of the OUs it targets actually exist.The AI reads the file and works with it for the whole turn. It can check the OUs a script targets over LDAP, match the version in a KB article against what the inventory says is installed, or look up the error from a screenshot in the event logs of the host it came from, all in the same answer.
Paste, drop or click
Click the paperclip in the composer, paste with ⌘V or Ctrl+V, or drag files onto the chat panel. Paste covers a screenshot on your clipboard as well as files you copied in Finder or Explorer. While you drag, an overlay shows what the chat's model accepts. Each file becomes a chip above the composer with its name, its size and, for images, a thumbnail. A dashed chip holds its place while the file is being prepared, and Send waits until it's ready. A message still needs text, so the AI knows what you want done with the file.
You can attach PNG, JPEG and WebP images, PDFs, and any plain-text file: logs, JSON, YAML, CSV, Markdown, PowerShell, shell or Python scripts, Dockerfiles. The type is decided by what the file contains, not by its extension. A message takes up to five files: images and PDFs up to 5 MB each, text files up to 256 KB each and 512 KB together, and 10 MB in total. Large screenshots are scaled down in your browser before they're sent. If a paste would push your message past its character limit, it becomes an attached pasted-text.txt instead of being cut off.
Text files work on every AI connection. Images and PDFs work on every model in the current catalog, on Anthropic, OpenAI and Azure AI Foundry connections, and the composer offers only what your chat's connection accepts.
One reply, then attach again
A file is context for the reply it's sent with. Within that reply the AI keeps the file in view across every tool call. On later messages it has its earlier answer but no longer the file, so to ask something new about the same document, attach it again. Every sent file shows as a tag in the transcript, and Attach again puts it back in the composer while the browser tab still holds it.
The conversation records only each file's name, type and size, not its content, and your browser holds the file in that tab's memory, not in its storage. The file goes to your connection's AI provider for that reply, under the provider's own retention terms. On a reply that carries files, the AI's web access is narrower: on Anthropic connections it doesn't fetch pages unless the connection restricts web access to an allowlist, and on OpenAI and Azure AI Foundry (OpenAI) connections hosted web search isn't offered for that reply.
If a send is refused, for example because a file is too large, your text and files come back to the composer. When the provider rejects a file, such as a PDF with more pages than the model accepts, the error names that file.
A New chat button you can find
Starting a fresh conversation is now easy to spot. New chat is the filled primary button at the top of the chat sidebar, and it expands to a full-width button with its shortcut when the sidebar opens. The tab strip has a labelled + New chat button. While the chat view is open, ⌘⇧O on macOS or Ctrl+Shift+O elsewhere starts one from the keyboard. The ended-chat banner now says Start a new chat, which is what it does.