> Documentation index: [/llms.txt](/llms.txt). Every page as one file: [/llms-full.txt](/llms-full.txt). # Work with files in a chat Work with files in a chat [#work-with-files-in-a-chat] Attach files to a message, or let the agent search your Library, and get the files the agent makes back as downloads. Attach files to a message [#attach-files-to-a-message] Do one of these in the composer: * Select **+**, then **Attach a file**, and choose one or more files. * Drag files onto the chat. * Paste files into the composer. Each file shows above your message while it uploads. Send once every file shows its size; the send button stays disabled while an upload runs. Select **×** on a file to remove it. The composer with Spring campaign results.csv attached, showing its size Any file type can be attached, and the composer sets no size limit. Attachments are yours: only you, and people you share the chat with, can open them. What the agent reads, and how [#what-the-agent-reads-and-how] The agent always gets the file. What differs is whether the model looks at it directly or the agent opens it with code in its sandbox. | File | What happens | | ---------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------- | | Images: .jpg, .png, .gif, .webp | The model looks at the image, up to the size limit below. | | PDF | The model reads short PDFs directly, up to the limits below. Longer ones, and password-protected ones, are opened with code. | | Word (.docx, .doc) and Markdown | Converted to text for the model. A very long document is opened with code instead. | | Spreadsheets (.xlsx, .xls), .csv, PowerPoint (.pptx, .ppt), .svg | Always opened with code, so the agent can calculate with the data. | | Anything else | Opened with code, when the agent can read the format. | Size limits for an image or PDF the model reads directly, by the model's provider: | Provider | Image | PDF | | -------------------------------- | ------------------- | ------------------- | | Gemini | 90 MiB | 20 MiB, 100 pages | | OpenAI (GPT) | 20 MiB | 8 MiB, 25 pages | | Anthropic (Claude) | 4 MiB | 8 MiB, 20 pages | | Models served through OpenRouter | Never read directly | Never read directly | Your deployment can serve a model through a route with lower limits. A file over the limit is not lost: the agent opens it with code, but it cannot see the picture or the page layout. If the agent seems to ignore an image, try a Gemini, GPT or Claude model. Use files from the Library [#use-files-from-the-library] Library files belong to the workspace, so everyone in it and their agents can use them. See [Manage files and folders](/docs/platform/library/files-and-folders). | To | Do this | | -------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------- | | Let the agent search the Library | Select **+**, then **Search in files**. The strip above the composer shows **All files** and **Searching all files in the workspace**. | | Search only some files | In the list that opens, check files or folders, or **Select all**. The strip then shows **Searching the selected files only**. | | Point at one file or folder | Type **@**, choose **Files**, and pick it. See [Mentions and commands](/docs/platform/chat/mentions-and-commands). | | Start a chat from the Library | In **Library**, check files and select **Chat with Files**. | Search in files open above the composer, listing Brand voice guidelines.md and Spring campaign brief.md, with All files and Searching all files in the workspace Select **×** at the end of the strip to stop searching the Library. An agent that has Library files in its own knowledge shows **From the selected agent · always included** instead, and those files cannot be removed for one chat. Open and download files the agent makes [#open-and-download-files-the-agent-makes] Files the agent creates appear under its reply, one row per file. The end of a reply with the generated clicks_by_channel.png below it, with a download button and Save to Library | To | Do this | | ---------------------- | ---------------------------------------------------------------------------------------- | | Preview the file | Select its name. The preview opens beside the chat. MP4, WebM and MOV videos play in it. | | Download it | Select the download arrow on its row, or in the preview. | | Copy it to the Library | Select **Save to Library**. | The buttons at the top of the preview save the file to the Library, download it, refresh it, show or hide **Your files**, share the chat, open the file in a new tab and close the preview. **Your files** lists every file in the chat, grouped as **Templates**, **Uploaded**, **Documents**, **Images** and **Archived**; empty groups are hidden. A chart previewed beside the chat, with the Your files panel listing the uploaded CSV and the generated image Save a file to the Library [#save-a-file-to-the-library] Select **Save to Library** on the file's row. The file is copied into the Library's **Output** folder, in a topic folder picked for it; the button then reads **Saved**. * **Save to Library** works once the reply has finished. * The Library takes only the file types it supports. For any other file, the button explains why and you can download it instead. * If the topic folder already has a file with that name, choose **Replace** or **Keep both**. Where saved files land, and the folder rules, are in [Manage files and folders](/docs/platform/library/files-and-folders).