Attachments and Web Search
Give the panel your documents and images, or let it gather live evidence — file types, limits, extraction behavior, and error messages explained.
A deliberation is only as good as its evidence. Dr Moot takes evidence from two directions: attachments bring your private material in, and web search brings live public information in. Every seat on the panel sees the same evidence.
Attachments
Click the paperclip in the composer to attach files to your next prompt.
What's accepted
| Kind | Types | Notes |
|---|---|---|
| Text | .txt, .md, .csv, .json | Used as-is |
| Documents | .pdf, .docx | Text is extracted |
| Images | .jpg, .png, .webp | Requires vision-capable models — see below |
Limits
- 5 files per prompt.
- 10 MB per file.
- Roughly the first 8,000 characters of each file's extracted text reach the panel. For long documents, attach the relevant section — or tell the panel which part matters so the extract that survives is the right one.
How the panel sees your files
Text and document content is extracted at upload time and inserted into every seat's context, clearly labeled as your attached material. Images are passed to the models directly — which means they only help when the models in play can see. If the current seat assignment can't process images, the composer warns you with "Can't see images" before you send. (Which models are vision-capable is controlled by your admin's roster.)
Rules that save you a failed run
- Wait for uploads to finish before sending — the send button won't include files that are still uploading.
- Scanned PDFs don't work yet. If a PDF has no extractable text (it's just page images), it's rejected — export a text-based PDF instead.
- File contents must match the extension. Dr Moot checks the actual bytes, so a renamed file (a
.zipposing as a.pdf) is blocked at upload.
If an upload is rejected
| Message | What happened | Fix |
|---|---|---|
| File too large | Over 10 MB | Split or compress the file |
| No readable text | Scanned/image-only PDF | Re-export as a text-based PDF |
| Content doesn't match type | Bytes don't match the extension | Save the file in its real format |
| Can't process file | Extraction failed | Try converting to .txt or .md |
Web search
Toggle the globe in the composer when the answer depends on the world after the models' training data: prices, releases, current events, live product or policy details, or any claim you want checked against public sources.
How it works:
- Dr Moot runs the search once per run and shares the retrieved evidence with every seat — the whole panel argues from the same sources.
- The search step appears in the deliberation, so you can see what was retrieved.
- The final answer cites sources as a numbered list (
[1],[2], …) with links. Open them — a citation is a pointer, not a proof.
Using both together
They solve different problems, and the strongest runs often use both:
- Attachments = your private context (the contract, the draft, the data).
- Web search = live outside evidence (the market, the news, the docs).
"Here's our proposed vendor contract (attachment) — check its terms against what this vendor currently advertises (web search)" is exactly the shape Dr Moot is built for.
Privacy note
Attachments stay within your workspace and the model calls that process your run. Web search sends only search queries derived from your prompt to the search provider — not your attached files.