Skip to Content
SynapseX ChatFiles and documents

Ask questions about your files and documents

Attach a contract, a spreadsheet or a report to a SynapseX Chat message and ask about what is actually inside it. When you finish this page you will have uploaded a document, asked a question that forces the answer to come from the file, and you will know the exact point where a document becomes too long to be read in full.

The two limits to know first

  • 10 MB per attachment. A larger file is refused in the browser with the message File size exceeds 10MB limit.
  • 60,000 characters of extracted text per document. Past that, the text handed to the model is cut off and explicitly marked as truncated.

The app also checks the file’s real signature against its extension, so renaming report.exe to report.pdf will not get it through.

How do I attach a file?

Click the + button in the composer and choose Add files or photos. You can also paste an image straight into the message box.

The picker accepts these types:

.txt  .md  .pdf  .csv  .tsv  .json  .docx  .pptx  .xls  .xlsx
images (image/*)   audio (audio/*)

What happens to the document?

PDF, DOCX, XLSX, XLS, CSV, TXT, MD and JSON attachments are converted to text on the server and inlined into your prompt. That is why every one of the six SynapseX models can answer about an attachment, not just a vision model: the model receives the document’s text alongside your question.

The extracted text is capped at 60,000 characters. When a document is longer, the text is cut at that point and an explicit truncated marker is appended, so a 400-page PDF is read partially. Two ways to work around it:

  • Split the file into sections and attach the section you care about.
  • Ask about a specific part — a named sheet, a chapter, a date range — rather than “summarise this”.

Worked example: a spreadsheet

Attach an .xlsx or .csv file, then send a prompt that forces the answer back into the data:

From the attached file, list the five largest values in column "Revenue"
and the row each one comes from. Show them as a table, and say which
sheet you read.

A good grounded answer names the sheet, reproduces five values that exist in your file, and cites the row for each. If the reply is generic, or the numbers are not in your file, the model is not using the attachment — re-attach it and ask again referring to the column by its exact header.

Spreadsheets are converted sheet by sheet, so telling the model which sheet you mean is the single most useful thing you can add to the prompt.

Where do uploads count against my plan?

Attachments consume your plan’s file storage allowance. When that allowance is full, an upload is refused with HTTP 413. Free space by deleting old attachments, or move to a plan with more storage — the per-plan storage figures are in Chat plans.

Libraries

Context > Libraries is per-user file storage: files you upload there persist on your account rather than living inside one conversation. Use it as a place to keep documents you re-upload often.

Document storage is what Libraries are documented for. Retrieval over a Library during a conversation is not a behaviour these docs describe or promise, so do not depend on it. When you want an answer grounded in a specific document, attach that document to the message — that path is described above and is the one you can rely on.

Conversations and projectsVoice input