Skip to main content
You can attach files to a chat message from the + menu’s Attach files item, or by dragging them anywhere onto the chat page — while you drag, the composer shows Drop files here to upload above Release to attach them to this message. (that second line reads Release to attach them to this message too. once the message already has an attachment). Receipt checks each file against an accepted-type list, converts most of them to Markdown, and — depending on the file and the model — either hands it to the model directly or retrieves the relevant part of it on your next question. Pasting 2,000 characters or more of plain text also becomes a .txt attachment rather than inline text; Your first chat covers that. However a file arrives, the send button stays blocked until every attachment has finished uploading.

​
What you can attach

.et is on the accepted-extension list without a corresponding entry on the accepted-MIME-type list — see the validation order below for what that means in practice.

​
Limits

Each file is capped at 10 MB, and you can attach up to 10 files to a single message. Once you’re at the limit, the Attach files item in the + menu relabels itself Attach files (max reached), and dragging more files onto the page shows Attachment limit reached — Remove one file first to upload more.

​
How Receipt decides whether a file is accepted

The order matters, and it isn’t “check the extension first”:
  1. If the browser-reported MIME type matches one of the accepted MIME types above, the file is accepted immediately — the extension is never even looked at.
  2. If the MIME type is present and doesn’t match, but also isn’t empty or the generic application/octet-stream, the file is rejected immediately.
  3. Only when the MIME type is empty or application/octet-stream does Receipt fall back to checking the file extension against the accepted-extensions list.
This is what the .et gap means in practice. Because .et has no MIME type on the accepted list, it can only reach step 3: it is accepted when the browser reports no MIME type for it or application/octet-stream, and rejected at step 2 if the browser reports anything else.

​
Exact error strings

  • File type is not supported for markdown conversion — the MIME/extension check failed.
  • File is empty — shown to you as Uploaded file is empty.
  • File exceeds limit of 10MB — over the 10 MB size cap.
  • Missing file field — the upload request didn’t include a file at all.
A deployment with no markdown conversion worker configured refuses anything that isn’t a direct-text file, with the message: This self-hosted instance only accepts direct text attachments until the markdown worker is configured. The worker is configured with CF_MARKDOWN_WORKER_URL and CF_MARKDOWN_WORKER_TOKEN; see Configuration.

​
What happens after upload

Files whose type is .txt, .md, .markdown, .csv, .html, .htm, or .xml are read inline — Receipt reads the text directly with no conversion step. Every other accepted type (PDF, images, Office formats, OpenDocument, Numbers) is sent to a Markdown conversion worker, which has a 20,000 ms timeout (CF_MARKDOWN_WORKER_TIMEOUT_MS changes it on a deployment you run yourself). If that step fails, you’ll see Markdown conversion timed out or Failed to convert uploaded file.

​
How an attachment reaches the model

  • Images and PDFs are handed to the model natively — but only when the active model declares that it supports image or PDF input. If it doesn’t, they fall back to the same path as everything else.
  • Everything else uses retrieval: the converted Markdown is split into paragraph-aware chunks of roughly 1,600 characters, with 260 characters of overlap between chunks, capped at 140 chunks per document.
On your next question, Receipt embeds your message, searches the attachment’s chunks, and appends up to 8 chunks (12,000 characters) to your message before the model sees it, under a header that begins User-provided attachment context is available below. If nothing comes back from that search — because embeddings are off, because nothing was indexed for that file, or because the search itself failed — Receipt falls back to appending the file’s whole stored Markdown, truncated to 2,000 characters per file, instead. Documents uploaded as organization knowledge are retrieved on the same turn, under a separate header that begins System-provided organization knowledge is available below. Only organization owners and admins can upload them, and they are only retrieved once your organization has at least one knowledge document that is active and finished indexing; see Org Brain.
Embeddings are skipped entirely when your organization has provider keys saved but none of them is an OpenAI key — a non-OpenAI key can’t authenticate the embedding endpoint. The upload still succeeds — you just get the truncated-excerpt fallback instead of ranked retrieval on that file.

​
File uploads aren’t gated by plan

Some parts of Receipt’s own copy suggest file uploads are a paid-plan feature — the upload route still contains a check that would return This feature is available on the Plus plan and above. if it ever denied a request, and the drag-and-drop hint has a locked state that reads File uploads aren’t available / Upgrade to attach files in chat. In this release, that check always allows the request, so every plan can attach files today and neither string can appear. The matching chat error string is listed on Errors and limits. Next step: monitor your runs on the Tasks, Sessions and Computers pages.