Upload a file into a conversation and the assistant can read from it. The file belongs to that conversation only, with no copy in a reusable library. For tools you can reuse across conversations, such as MCP servers and REST APIs, see Connectors.
Uploading a file in chat
Drop a file on the message box, paste it in, or choose Add files or photos from the + menu. What happens next depends on the file type:
- PDF, Word, PowerPoint, plain text, code: vectorised. The file is split into chunks, embedded so they can be searched by meaning, and stored in a vector collection for this conversation. At each turn the assistant retrieves the chunks relevant to your question.
- CSV, Excel, TSV: uploaded as a tabular source. The assistant can run SQL-style queries against it instead of reading the rows as text.
- Images (PNG, JPG and others): run through OCR (optical character recognition) so their text is searchable. The original image is kept too, for models that can read images.
While the file uploads, a chip in the message box shows its progress. Wait for it to finish before you press Enter, or the message goes without the file. If processing fails (unsupported format, corrupt PDF, OCR error), the chip turns red with the reason. Remove it and try again.
The file goes with the next message you send and shows as an attachment on that message. From then on the assistant can use it, whether or not you mention it.
A paste of 15,000 characters or more, or your model's inline limit if that is lower, becomes a .txt source and is attached. A paste of 1,500 characters or 10 lines or more shows as a chip you can click to view or edit, and is still sent to the assistant in full.
Duplicates
If you upload a file you've already attached, PrivateMind links the existing copy without processing it again. The chip finishes straight away and the file attaches to your next message, with no second copy and no extra message in the conversation.
Where next
- Connectors: reusable MCP servers and REST APIs you attach to a chat
- Chat: how attachments feed into the conversation
- Projects: pin a set of connectors to every new conversation