Upload a PDF, Word doc, Excel sheet, PowerPoint deck, plain text, Markdown, CSV, JSON, or XML file straight into Training, and every employee on your team can search it within minutes. Up to 20 files per run, each up to 72 MB, feed the same shared knowledge base your employees check on every message. No connected app and no crawlable URL required.
Train from File Uploads lets you hand an employee a document instead of a subscription. Drop a PDF, a Word file, an Excel sheet, a PowerPoint deck, plain text, Markdown, CSV, JSON, or XML file into Training, and it becomes part of the knowledge base every employee on your team searches automatically, on every message, without anyone telling them to look.
A run accepts up to 20 files at once, each up to 72 MB. The system reads the text out of each file, checks it against what is already trained so re-uploading an unchanged document gets skipped, and builds searchable knowledge from what is genuinely new. You watch the whole thing in a live session: files found, files processed, files skipped as duplicates, and any that failed, with the credits the run used shown once it completes.
This is the source for the document that only exists on a hard drive, not in an account you can log into: a scanned contract, a one-off spreadsheet a client emailed you, a policy nobody ever put in Notion. Training from an App pulls through a connected tool's live API, and Training from a Website crawls a public URL; File training needs neither. Whatever you could attach to an email, you can train an employee on.
Every file is read by Sistava itself before anything is stored, rather than being handed off to a library and assumed to have worked. That matters because the failure mode it replaces was invisible: a document could be accepted, marked complete, and still teach your employees nothing. Now each file reports what actually happened to it, and a file that genuinely cannot be read, such as a scan with no selectable text in it, is marked as failed with the reason sitting next to it.
You do not have to take the result on faith either. The Knowledge Inspector opens from Training and shows you the documents that made it in, the stats for the run, and a test search box, so you can type the question you expect a customer to ask and see whether the answer comes back out of the file you just uploaded. If it does not, you know before a customer finds out.
Training uses credits, and the run tells you how many it used once it completes, so the cost of teaching your employees something is visible rather than buried in a monthly total. Because unchanged content is recognised and skipped, keeping a document current costs roughly what the change was worth instead of the price of the whole file again.
File training accepts PDF, Word (.docx), Excel (.xlsx), PowerPoint (.pptx), plain text (.txt), Markdown (.md), CSV, JSON, and XML files. Older Office formats like .doc, .xls, and .ppt are not on the list; save or export to the modern format first.
Each run takes up to 20 files, and each file can be up to 72 MB. A file that is empty, in an unsupported format, or too large is rejected before upload with a specific reason, so you know immediately which file to fix rather than finding out after a run fails.
Files are one of four training sources alongside connected Apps, crawled Websites, and pasted Text, and all four write into the same company-wide knowledge base. You do not train each employee separately: upload a file once, and every employee in the organization can draw on it starting with their very next conversation.
Training from files requires the Builder plan or higher, the same gate as the other three sources. If your document changes, upload the new version and start another run; unchanged content from the previous version is skipped automatically, so only the differences get processed and billed.
Not every document has text in it. A photographed page, a scan produced without character recognition, or a PDF that is really just an image has nothing to extract, and Sistava says so instead of quietly reporting success. That file shows as failed with the reason attached, and the run tells you how many items did not make it, so a completed run means the content is genuinely searchable.
The other half of that promise is verification you can run yourself. Open the Knowledge Inspector after a run, look at the documents listed against the session, and use the test search to ask the question you actually care about. If the answer comes back grounded in your file, your employees will answer the same way in a live conversation.
From a file on your machine to knowledge every employee can search
Open Training on the Company Dashboard and pick the Files tile, or upload directly from the Files tile inside a source picker. Select up to 20 files at once; each one is checked against the allowed file types and the 72 MB per-file ceiling before it uploads.
Once uploaded, the platform extracts the text from each file, whether that means reading a PDF's pages, a spreadsheet's cells, or a slide deck's text boxes, and hands it to the same knowledge-building pipeline every other training source uses. Documents that match something already trained are skipped as duplicates; only new or changed content gets processed.
The run shows up as a session with real-time progress, and it queues automatically if another training run is already in flight, so a second file upload never gets rejected because Notion training happened to be running. Once it finishes, the extracted knowledge is immediately searchable by every employee in the company, not just the one you were looking at when you uploaded it.
Upload a signed contract, an employee handbook, or a compliance policy as a PDF and your employees can answer questions against its actual language instead of guessing from a summary you typed up separately.
A client emails over a pricing sheet or a project tracker in Excel. Upload it directly; there is no need to get it into a connected app first just so an employee can reference it.
Upload your standard operating procedures, sales scripts, or brand guidelines as Word or PowerPoint files the day you hire a new employee, so it starts every conversation already knowing how your company operates.
When a document changes, re-upload the new version. The system skips anything unchanged and only processes what is actually new, so keeping your employees current does not mean retraining from scratch.
| Before | After |
|---|---|
| The policy PDF lives on your laptop, so every time somebody asks about it you open the file, find the paragraph, and paste it into the conversation yourself. | The PDF is uploaded once and every employee answers from its actual wording, on every message, without you fetching anything. |
| A document only reaches an employee if it already lives in a connected app or on a public web page, so the one-off spreadsheet a client emailed you never makes it in at all. | You upload the spreadsheet straight from your machine. No connected account and no crawlable URL required. |
| A small edit to a long handbook means processing and paying for the entire document again, so people put off keeping it current. | Unchanged content is recognised and skipped, so refreshing a document costs roughly what changed in it. |
| An upload says it succeeded and you have no way to tell whether the text was really read, until a customer gets an answer that is not in the document. | Each file reports processed, skipped as a duplicate, or failed with the reason, and a test search shows you what was learned before anyone else sees it. |
PDF, Word (.docx), Excel (.xlsx), PowerPoint (.pptx), plain text, Markdown, CSV, JSON, and XML. Legacy formats such as .doc, .xls, and .ppt are not supported; export them to the modern equivalent first.
Up to 20 files per training run, and up to 72 MB per file. If you have more than 20 files, upload the first batch, let it complete, then start a second run; nothing is lost, it just queues.
Yes. File training, along with Apps, Websites, and Text training, requires the Builder plan or higher.
The system checks new content against what is already trained. An unchanged file is skipped automatically so you are never charged twice for the same knowledge. A new version of a document is processed as new content.
No. Training is company-wide. A file you upload becomes part of the shared knowledge base every employee in your organization searches automatically on every message.
Open the Knowledge Inspector from Training after the run finishes. It lists the documents that were processed, shows the stats for that session, and gives you a test search box. Type a question you expect a customer to ask and see whether the answer comes back out of your file.
Only if the file contains real, selectable text. A scan or an image-only PDF has nothing to extract, so it is marked as failed with the reason shown next to it rather than silently reported as complete. Run the scan through character recognition first, then upload the searchable version.
Yes. Building searchable knowledge from a document consumes credits, and the training session shows how many the run used once it completes. Content that matches something already trained is skipped, so re-uploading an unchanged document does not spend credits again.
Train from File Uploads is part of Things that make them smart.
Upload documents, connect Notion or Google Docs, and choose the source content that becomes shared company knowledge. A 6-layer memory system ensures they never forget a conversation, a decision, or a preference. When one employee learns something new, the whole team benefits through shared knowledge.