5. Upload and maintain knowledge sources

Add documents, FAQs and website content to give your agent the information it needs. After an import, check the extracted content and test a question before relying on the new source.

Open Knowledge → Sources → Add source. Staff can maintain sources; creating the knowledge base itself requires an Admin.

File formats

Ordinary File uploads have a maximum size of 20 MiB. Extraction also has content complexity limits: two million characters and 50,000 document elements. A file can therefore be rejected even when its byte size is below the upload limit.

Extension	Preparation and extraction	Common problem
.pdf	Use selectable text. Text is extracted and converted for retrieval.	Image-only scans have no OCR; form-field content is not extracted.
.doc, .docx	Use readable paragraphs, headings and tables. Embedded documents are not imported.	Protected, damaged or unsupported document content can fail extraction.
.txt	Supply UTF-8 text with clear topic boundaries.	Wrong encoding or empty content.
.md	Supply UTF-8 Markdown with meaningful headings.	Broken organisation or empty content.
.csv	Use File for general tabular content; use FAQ import for question-and-answer rows.	Treating an arbitrary CSV as an FAQ spreadsheet.
.rtf	Supply a readable Rich Text Format document.	Unsupported or unreadable content.
.xlsx	Supported through FAQ spreadsheet import, not ordinary File upload.	Multiple worksheets or a general-purpose spreadsheet supplied as FAQs.

Before uploading, check the parts most likely to need preparation:

Scans and images: provide a readable text version; images and diagrams cannot replace written explanations.
Protected files: there is no password-entry workflow.
Tables: inspect the extracted text to confirm that column relationships remain clear.

Document upload

Select File and choose the document.
Select Upload.
Review the displayed pages, credits and size. Files costing more than one credit require a separate Confirm action; one-credit files proceed.
Wait for processing to finish.
Open View content and compare the content with the original.
Test a question that depends on the uploaded information.

The displayed page count is based on extracted text tokens, not the document's printed pagination. The upload preview expires after 15 minutes. If you leave it too long, select and upload the file again.

FAQ spreadsheets

Choose FAQ → Import spreadsheet. Use XLSX or UTF-8 comma-separated CSV with exactly the required Question and Answer headers, in either order and any letter case. Put headers in the first nonblank row. Each subsequent entry needs both values; each trimmed field can contain at most 4,000 characters.

Use the provided sample when preparing a new sheet, and check these limits:

Maximum 500 entries and 500,000 bytes; blank rows are ignored.
XLSX imports use the first worksheet and saved cell values. Formulas are not recalculated.
CSV answers containing commas or line breaks must be quoted.

Review FAQs to import and the error count. Invalid entries are counted and skipped. Choose Replace matching FAQs or Skip matching FAQs before selecting Import. Questions match by their complete text after trimming; similar wording is not a match. FAQ imports spend no credits but have daily limits and use knowledge storage.

Website imports

For one page, choose URL, enter Page URL, optionally set Title, and select Add. This costs one credit. Use a public page that contains the information you want to import; a link alone does not grant access to private content.

For multiple pages, choose Domain, enter Site URL and select Preview. Each preview costs one credit. Review the discovered URLs and crawl cost, then select Confirm. A domain source takes at most 100 pages and is available on Pro and Business. A preview is an estimate of discovered pages, not proof that every page will be extracted successfully. Review the processed pages and any failures afterwards.

Refresh and replacement

Use the source's Re-sync action to queue another import of a supported file, URL or domain source. URL and domain sources do not automatically follow every website edit. Refresh deliberately after the source changes. A resync can spend credits again; do not repeatedly retry an inaccessible page without correcting the cause.

To replace a file with different bytes, add the new file as a source, inspect and test it, then deactivate or delete the old source. The resync action reprocesses the stored file rather than asking for a replacement upload.

Failed imports

Failed processing does not automatically refund credits already spent when the job was accepted. If the old source must not remain usable during a refresh, turn Active off. Re-enable it only after you verify the new result.