Attachments
Give an agent a screenshot, a spec or a file to work from. Up to eight files per message.
Add a file
Attach files to any agent’s message with the paperclip, by dropping them on the message box, or by pasting a screenshot.
What you can attach
| File | Up to | What the model gets |
|---|---|---|
| Images (PNG, JPEG, GIF, WebP) | 8 MB | The image, if the model can see images |
| PDFs | 20 MB | The text. A scanned PDF with no text says so |
| Text and code files, including SVG | 1 MB each | The contents |
Models that can see
OpenAI’s GPT-4o, 4.1, 5 and o-series models, Claude, Gemini, most vision models on OpenRouter, and local vision models such as LLaVA or Qwen-VL can see images.
DeepSeek and other text-only models get the file name instead, and the reply notes that the image was left out.
What agents remember
Text and PDF contents stay in the agent’s history for later messages. Images are only sent with the message they’re attached to. Attachments go to the provider you chose, like the rest of the message.