File uploads
The Chat Agent lets you attach files directly to a chat message. Once attached, the model can read and reason over the file's contents — describing an image, summarising a document, analysing a spreadsheet, or reviewing a code file — without any copy-paste required.
Attaching a file
Click the paperclip (attach) icon in the message input bar, then select one or more files from your device. Attached files appear as thumbnails or file chips above the text field before you send. You can remove an attachment by clicking the × on its chip.
Alternatively, drag and drop files directly onto the chat window — a drop zone will appear to confirm the target.
Supported file types
The Chat Agent accepts the following categories of files:
- Images — JPEG, PNG, GIF, WebP, and SVG. Sent directly to vision-capable models for analysis, description, and reasoning over visual content.
- PDFs — text is extracted from each page and passed to the model as readable content. Scanned PDFs with no embedded text layer may produce limited results.
- Plain text and code —
.txt,.md, and common source code extensions (.py,.js,.ts,.json,.yaml,.csv, and others). The file content is included verbatim in the conversation context. - Documents — Microsoft Word (
.docx) and PowerPoint (.pptx) files. Text and slide content is extracted and passed to the model.
How the model uses attached content
Images
When you attach an image and your selected model supports vision, the image is sent alongside your text message. The model can describe what it sees, answer questions about the image, compare multiple images, or extract text visible in the image (OCR). The accuracy of visual reasoning varies by model — newer, larger models generally perform better.
Documents and text files
Text is extracted from the file and included in the conversation context as readable content. The model can summarise, quote, answer questions about, translate, or rewrite the extracted text just as it would any other message content.
File size limits
Each individual file must be 25 MB or smaller. You can attach up to 5 files per message. Total attachment size across all files in a single message must not exceed 50 MB. These limits apply regardless of your plan.
Model availability
File upload support — and in particular image (vision) support — depends on the model you have selected for the conversation. Not every model can process every file type.
Vision support varies by model
Image attachments are only processed by models that declare vision capability — for example, GPT-4o, Claude 3 and Claude 3.5 series, and Gemini 1.5 Pro. If you attach an image to a conversation using a text-only model, the image will be ignored and you will see a warning in the chat. Switch to a vision-capable model to use image attachments. See Choosing a model for a comparison of model capabilities.The attach icon is disabled automatically when the selected model does not support any file input. To re-enable it, switch to a model that supports file uploads from the model selector at the top of the conversation.