An assistant in Docutrain is a document you can talk to. You give it content, Docutrain reads it, splits it into small overlapping sections and indexes them, and from then on the assistant answers from that material. Here is every way to get content in, the limits that apply, and what you see while it processes.

Uploading is done from the dashboard by owner administrators and super administrators. If the upload area is missing from yours, that is a role question for your administrator rather than a bug.

Starting an upload

Go to the documents area of your dashboard. From the dashboard home, Upload document takes you there and opens the upload panel; the tile beside it shows your allowance, for example "3 / 10 used". Find the Train New Assistant card and click Create Assistant.

Four starting points are offered. PDF / Documents takes PDF, DOCX and PPTX; its button reads Choose files, and you can drag files straight onto the card. Paste Text is direct input for quick articles, notes or briefings. Audio Training transcribes lectures, meetings and podcasts, and shows an Upgrade badge if your plan does not include voice training. Web Scrape appears only when web scraping is enabled for your installation.

Whichever you pick, the Train New Assistant window opens with tabs: PDF / DOC, Text, Audio, Web where available, and Form. Two settings above the tabs apply to all of them. Assistant name is how the assistant appears in the dashboard and in chat, up to 200 characters; selecting a file suggests one from the filename. Auto-generate cover image creates a cover from the document's top keywords once training finishes — turn it off and you can add a cover later under Configuration → Images.

One small mercy: once you have started filling in the form, Escape no longer closes the window, so a configured upload cannot vanish to a stray keystroke. Close it with Cancel or the X.

Files, and the questions they raise

The PDF / DOC tab accepts PDF, Word (.doc, .docx), PowerPoint (.pptx), Excel (.xlsx) and Markdown (.md, .markdown). Click Browse or drag & drop. The zone states the formats and the current size limit, for example "PDF · DOCX · PPTX · XLSX · MD up to 40MB". Once a file is selected you see its name, size and — for PDFs — its page count, with Change file if you picked the wrong one. Then Upload Document.

Select a PDF and Docutrain asks What type of PDF is this?, because how it reads the file changes what the assistant can show in answers. Document ("Reports, papers, handbooks — mostly text with some figures") pulls out individual figures, tables and charts and shows them inline when relevant. Slide deck ("A PowerPoint or Keynote presentation exported to PDF") captures every page as a full slide image, so the assistant can present entire slides. Click Continue; Change in the File handling row reopens the choice any time before upload. What that decision affects downstream is covered in images and video.

Page counts are checked before upload. Over 300 pages you get a warning that documents this large often exceed training limits, but you can proceed. Over 500 the upload is blocked: "This PDF has N pages, which exceeds the 500-page limit for training." Either way the fix is the same — split the file, train on the first part, then add each remaining part through Document configuration → Retrain with Add to existing data.

PowerPoint uploads extract slide text only. For the actual slide pictures to appear in answers, export the presentation to PDF and choose Slide deck; Google Slides works the same way. Excel is accepted, but assistants work best on prose. Password-protected and legacy .xls spreadsheets are rejected, and a mostly-numeric one still trains but earns a warning: "This spreadsheet may make a weak chatbot source."

Text, audio, web pages and forms

On the Text tab, name the assistant and paste into the Text content box. A live counter shows characters used out of two million plus a word count; the minimum is 10 characters and 5 words. Click Upload Text.

The Audio tab takes MP3, WAV, M4A, OGG, FLAC and AAC on Business and higher plans; lower plans see "Voice training requires Business or higher" with a View plans button. The recording is transcribed automatically and then trained like any text document, so allow extra time for long files.

The Web tab takes one http or https URL for a publicly accessible page. The assistant name is optional here — leave it blank to use the scraped page title — then click Scrape & Upload.

The Form tab uploads nothing at all. It sends you to the Forms area of your dashboard to build a form whose responses become a chattable knowledge base: see forms that become knowledge.

The limits, plainly

File size is whatever the upload window states, 40 MB in a typical installation. PDFs are warned at 300 pages and blocked at 500. Pasted text is capped at 2 million characters, assistant names at 200. You can have three documents in flight at once, waiting or processing, and they train one after another; a fourth is refused with "Upload queue is full (max 3 documents at a time)."

Separately, each owner group holds a set number of active trained documents: 1 on Free, 10 on Pro, 50 on Business, no limit on Unlimited. At 80 percent an amber notice appears; at the limit a red Document Limit Reached notice disables the upload options until you delete a document or upgrade. The allowance is rechecked as each queued document comes up, so a batch can stop partway with "Document limit reached for your plan." See plans and limits.

Watching it train

The window becomes an Upload successful screen: "Your document is now being processed. You can watch the progress in the dashboard." Click Close and look at the Processing panel.

The pipeline examines the file, extracts its text (transcribing audio, reading scanned pages visually), chunks that text into overlapping sections, generates embeddings so the assistant can find relevant passages, then creates the document with an AI summary and keywords. Embedding generation is usually the longest stage. Small text documents take 30 seconds to a minute, typical PDFs one to two minutes, larger PDFs and presentations two to four, large documents and audio four to eight or more, web pages one to five including the fetch.

Each card shows the title, the current stage, and chips for file size, chunks so far, embedding batch progress ("Batch 3/12"), elapsed time, time remaining, and "Cover generating…". With several documents in the panel the cards sit behind compact tabs with status dots, and the panel follows whichever task is actively training unless you select another tab.

Status labels are Queued (waiting behind one of your own documents), Pending, Processing, Ready, Error, Cancelled, and Stuck — no progress for over five minutes, which adds a Force Retry button. If your PDF contains complex diagrams, an extra reading pass runs and explains the delay on the card. If the document has slides or figures, its gallery builds in the background after the text is ready, so you can start chatting while images fill in.

Unless you change something, a newly trained document arrives with the title you gave it, a unique web address derived from that title, private access limited to your owner group, an AI-generated cover from its top keywords, an AI summary and keywords, and source references hidden in chat answers. All of it is adjustable afterwards in the document settings.

For more than a couple of files, the same window will queue three at a time with per-file settings — that is the batch upload queue.