Somebody on your team answers the same five questions about a long document every week. The document contains the answers. Nobody reads two hundred pages to find one paragraph, so they ask a person instead.
Docutrain removes that step. You upload the document, it is processed — the product calls this training — and what comes out is a chat page where anyone with the link can ask a question in ordinary language and get an answer built from your content, with numbered citations pointing back to the source passages. There is no code to write and no model to configure.
The constraint is what makes it worth using. An assistant answers from the documents it was trained on, not from the open internet and not from the model's own general knowledge. Ask about something the document does not cover and you get a refusal under a heading reading Not in this document, rather than a confident paragraph you would then have to go and check. That is a smaller promise than a general-purpose chatbot makes, and a far more verifiable one. How a grounded answer is built walks through the mechanics.
Three kinds of people
Docutrain has three audiences, and they see very different products.
Askers are the end users: anyone chatting with a trained assistant. They may arrive through a shared link, a widget embedded in a website, an organization's own domain, or Slack and Telegram. For public documents they do not need an account at all, and most of them never see a dashboard.
Owner admins are the people on the other side. They upload documents, brand the experience, manage members, and control who can open what. An owner group's first administrator — the primary admin — also handles billing.
Super admins are Docutrain's own platform operators. They oversee every organization and have sections that ordinary customers never see, such as platform-wide plan management and system messages. Where this series mentions a super-admin-only capability, it says so.
There is a fourth position worth naming: the registered user who belongs to an owner group but is not an administrator. They sign in, browse and chat with the documents their groups have shared with them, and can request access to further groups from My Profile. They have no Home section — signing in takes them straight to Documents. Roles are covered properly in users, roles and invitations.
The vocabulary you will keep meeting
Five words carry most of the product. They appear in the dashboard, in the settings, and throughout the rest of this series.
| Concept | What it is |
|---|---|
| Document | An uploaded file, pasted text, audio recording, or scraped web page that has been trained so people can chat with it. |
| Owner group | The organization that owns documents, branding, members, and a plan. Also called a workspace. A group can be a large team or a single person; having one is what lets you create assistants at all. |
| Collection | A curated set of documents shared through a single link, optionally behind a passcode. |
| Chat | The conversation interface where questions are asked and cited answers come back. |
| Plan | The subscription tier of an owner group — Free, Pro, Business, or Enterprise, whose badge reads Unlimited — which sets limits such as how many documents you can train. |
The plan badge sits in the dashboard top bar for owner admins, next to the account menu. What each tier includes is in plans and limits.
What you can train an assistant on
The upload area opens from Create Assistant and has a tab per content type.
| Content type | Accepted formats |
|---|---|
| Documents | PDF (.pdf), Word (.doc, .docx), PowerPoint (.pptx), Excel (.xlsx), Markdown (.md, .markdown) |
| Text | Pasted or typed directly into the Text tab |
| Audio | .mp3, .wav, .m4a, .ogg, .flac, .aac, transcribed automatically; availability depends on your plan |
| Web pages | A page address, scraped into training content, where enabled |
| Forms | Structured form responses |
That last row is easy to miss and quietly useful: you can build a form, collect responses, and make the accumulated responses chattable. See forms that become knowledge. Size limits and the per-format details — PowerPoint slide image extraction, for instance — are in training your first assistant.
How readers actually arrive
Most askers never see a Docutrain home page. They arrive through something an owner handed them:
- A direct document link that opens the chat for one document.
- An owner library link that opens an organization's full set of documents.
- A collection link, sometimes protected by a passcode. See collections.
- A shared conversation link, which opens a read-only copy of somebody else's chat and can usually be continued. See sharing a conversation and exporting an answer.
- An embedded widget inside another website, or a Slack or Telegram bot. See embedding on your website and Slack, Telegram and custom domains.
- An organization's own domain, where the whole experience carries the owner's branding rather than Docutrain's. See branding and landing pages.
If someone does land on Docutrain without a link, the Start page offers two paths: Single document ("Focus on one specific resource") with a Search for a document… box, and Owner group ("Browse an organization's full library") with a Search for an owner group… box. Pressing Ctrl+K, or Cmd+K on a Mac, jumps to whichever search box is active.
Which of those routes a given reader can use depends on the document's Access Level, set per document in the editor: Public, Passcode, Token link, Registered, Owner group, or Owner admins only. Every one of those rules is checked on Docutrain's servers on every request — loading the document, sending a message, fetching a reference — so access cannot be talked around in the browser. Access levels explained covers the trade-offs, including the fact that Public and Passcode are paid-plan features.
You do not need an account to chat with public documents. You need one to open restricted documents shared with you, and to create assistants of your own. Creating a group is free to start, and a group of one is a perfectly ordinary thing to have.
If you are evaluating Docutrain, the honest summary is this: it is good at making a fixed body of material answerable and checkable, and it is deliberately bad at answering anything else. Whether that is a limitation or the whole point depends on what you were going to use it for.