Train on documents
A chatbot trained on your own documents
Upload your manuals, price lists, product sheets and terms as PDF, Word, PowerPoint, Excel and other files. intoCHAT reads the text, indexes it, and your AI agent answers visitors from it, in their language, instead of guessing.
Visitor: How often should I descale the espresso machine, and what do I use?
Agent: With hard water, descale about every two months; with soft water, every four. Use the descaling liquid made for the machine, mixed with water as the bottle describes, and run the descaling program from the settings menu. It takes about 20 minutes.
Visitor: The descaling light is still on afterwards. What now?
Agent: The light goes off only after the rinse cycle that follows the program. Fill the tank with fresh water and confirm the rinse when the display asks for it. If the light is still on after that, please contact our service team.
The answers are in your documents, not on your website
A lot of what customers ask is written down somewhere that a website chatbot never sees: the PDF manual, the current price list in Excel, the warranty terms in a Word file, the slide deck your sales team uses. A general-purpose AI model has read none of them, so it either says it doesn't know or, worse, makes something up. Training a chatbot on your documents closes that gap: you upload the files once, and the agent looks up the right passage for every question and answers from it.
Why train your chatbot on documents
Your files become knowledge the agent searches for every question.
The formats you already have
PDF, Word, PowerPoint, Excel, OpenDocument, RTF, EPUB, CSV, plain text and Markdown. You don't have to convert or rewrite anything before you upload it.
Structure is kept
Documents are converted to Markdown with their headings, paragraphs, lists and tables, which helps intoCHAT split them into passages that make sense on their own.
Answers grounded in your text
For every question the agent searches your documents and answers from the passages it finds. When nothing relevant is there, it says what it can't confirm instead of inventing an answer.
Easy to keep current
Upload a new version under the same file name and it replaces the old text of that source. An identical file is skipped, so re-uploading a whole folder is safe.
One document, every language
A German price list can answer a French visitor. The agent replies in the language the visitor writes in, unless you set a fixed reply language.
How to train a chatbot on your documents
Create an agent
Sign up and create an agent. You can add your website too, but documents work on their own: an agent can be trained on files alone.
Upload your files
Open the Knowledge tab, choose Files under Add knowledge, and drop your files onto the upload area, or click it to pick several at once. Files over 4 MB are uploaded with a progress bar first, then read.
Check what was read
Open each file's preview with the eye icon in the source list. It shows exactly the text the agent learns from. Tables and multi-column layouts are worth a look.
Train and test
Click Train agent at the top of the tab. Training continues on the server if you leave. Then ask the Playground the questions your customers ask and see which passages each reply used.
What document training includes
Up to 50 MB per file
Each file can be up to 50 MB and hold up to 4 MB of extracted text. A larger text fails with a message asking you to split the file.
Characters per agent
All of an agent's sources together can hold 100K characters on Free, 500K on Starter, 1M on Pro and unlimited on Enterprise. What counts is the extracted text, not the file size.
Hybrid search over passages
Each document is split into passages of up to about 1,600 characters. Semantic search finds the same meaning in other words, and keyword search finds exact product codes, names and prices.
Priority for the files that matter
Mark a file with the star and it ranks above other matching knowledge and wins when sources disagree. It applies to the next answer, without retraining.
No fine-tuning
intoCHAT doesn't retrain an AI model on your files. It indexes them and hands the model the passages that match each question, so a new upload is used as soon as it is trained.
Text kept, files not
intoCHAT keeps the extracted text, not the original file. Large uploads are deleted from temporary storage once they have been read.
Documents are data, not commands
A line in a file that says "ignore your rules" has no effect. Your knowledge is treated as reference material, and your instructions always outrank it.
Frequently asked questions
Can I train a chatbot on PDFs?
Yes. Upload the PDF under Knowledge → Files and click Train agent. intoCHAT extracts the text, keeps headings, lists and tables, and the agent answers from it. The PDF needs a text layer: there is no OCR, so a scanned PDF fails with a message saying so. Run OCR on it first, or paste the text as a text snippet.
Which file types can I upload?
PDF; Word (.doc, .docx, .docm); PowerPoint (.ppt, .pptx, .pptm, .pps, .ppsx, .ppsm, .pot); Excel (.xls, .xlsx, .xlsm, .xlsb); OpenDocument (.odt, .ods, .odp); RTF, EPUB and CSV; and plain text and Markdown (.txt, .md, .markdown).
How many documents can my chatbot learn from?
There is no limit on the number of files, only on the total text per agent: 100K characters on Free, 500K on Starter, 1M on Pro and unlimited on Enterprise. A large PDF full of photos may hold only a few thousand characters, while a small spreadsheet can hold many more. If an upload would go over the limit, it stops there and the remaining files aren't added.
What is not read from a document?
Text inside pictures, such as charts, diagrams or screenshots; scans without a text layer; and password-protected files. Remove the password and upload the file again. Check the preview after uploading: if something is missing there, the agent can't know it.
How do I update a document?
Upload the new version under the same file name. If the text changed, it replaces the old text of that source and you see "Updated" with the file name; click Train agent to use it. Until then the agent answers from the previous version. A file with a different name is added as a new source, so delete the outdated one yourself. Uploaded files aren't re-read by auto-resync, which checks website pages, Notion pages and Google Drive files.
Do visitors see which document an answer came from?
No. Source links under a reply are shown for website pages only. You can see which passages a reply was given, and how closely they matched, under Why this answer in the transcript on the Conversations tab.
What if two documents say different things?
When matches are close, the agent prefers Q&A pairs, then sources marked as priority, then text snippets, then files. Between sources of the same kind, it trusts the most recently updated one. Remove outdated versions, or mark the right file as priority. On Pro and Enterprise, the Conflicts card lists facts your sources disagree on.
Can I use other document sources besides uploads?
Yes. You can import Notion pages and pick Google Drive files, which can be kept in sync, and write text snippets and Q&A pairs yourself for facts that must be answered one way. All of them count toward the same character limit.
Put your documents to work
Create your agent free, upload your manuals, price lists and terms, train it and ask it your customers' questions in the Playground.
Create your agent freeFree plan, no credit card needed.