The assistant's knowledge and sources
What the AI assistant answers from and how you control it: crawl run through your website, schedule, blocked areas, files and Google Drive, approvals, improved answers and knowledge gaps.
The AI assistant answers from your own content and cites it as a source. Which content that is, you decide. Most settings are under /app/projects/{id} in the AI & knowledge tab. An overview of all projects is in the side menu under Knowledge base (/app/kb).
How to build up the knowledge base step by step is described in the product handbook. This page is about what you can set.
What the assistant answers from
- Your website – pages read in by the crawl run, and pages your visitors are currently viewing.
- Your help centre – published articles.
- Files and Google Drive – documents you upload or connect.
- Improved answers – answers you have defined yourself.
- Products – items, variants and categories from the data layer of your pages (section Products the bot knows).
Crawl run through your website
Section Crawl run: during a run, reactive.chat walks your pages and reads the text for the knowledge base, the data layer and the product catalogue. It takes a few minutes and runs in the background.
- Run now – starts within a minute. At most one job runs per project.
- Schedule –
off — only on demand(default),dailyorweekly. The run uses your own bandwidth, so it is off by default. - from which hour – 0 to 23, server time. At night your website is least busy.
Blocked areas
Section Blocked areas: paths on your website the chat never reads. No page text is created there for the knowledge base, the crawl run leaves them out, and the assistant sees neither title nor content there and never cites them as a source. The widget stays visible there; your team chats as usual.
One path per line:
/admin
/konto*
/?action=admin
# comment
/adminblocks/adminand everything below it, also with a language prefix such as/pl/admin.*matches any characters:/konto*also matches/kontoauszug.- With
?the query is compared as well. #at the start of a line is a comment.- 60 patterns at most, up to 200 characters each. A pattern that would match the home page
/is rejected. An invalid line prevents saving, and your account names it.
On saving, reactive.chat removes whatever was already captured from these areas.
There is no selection of individual pages to include. You steer by exclusion.
Excluding areas of a page
The widget reads the visible text of the page the visitor is currently on, so that the assistant can refer to it. You exclude individual areas from this, such as a customer account or a form with personal data:
<div data-rc-private>
… is not read …
</div>
data-rc-ignore works the same way. You can switch off reading the page entirely on the snippet with data-rc-dom-spiegel="false". Then “Show me” is dropped as well (see Product tours). If your project requires consent, the widget only reads the page after the visitor agrees.
Files and Google Drive
Under /app/projects/{id}/kb/files, Files & Google Drive, you upload documents: PDF, DOCX, TXT, MD, CSV and HTML. The assistant cites the file name as the source. Scanned PDFs are read by text recognition (OCR).
Limits per plan:
- Free – 5 MB per file, 50 PDF pages, 20 files per project, 300,000 characters in total, 5 pages of text recognition, no Google Drive.
- Professional – 20 MB, 300 pages, 200 files, 1.5 million characters, 30 pages of text recognition, Google Drive.
- Agency – 25 MB, 500 pages, 1000 files, 3 million characters, 50 pages of text recognition, Google Drive.
You connect Google Drive with read access. You choose up to 50 items: Google Docs, Sheets (the first sheet), Slides and files in the formats listed above. reactive.chat reads them again roughly every six hours. Owners and admins may connect and disconnect. Disconnecting removes the Drive content from all projects of the workspace.
Caution: the assistant can repeat anything you upload to any visitor. Only upload what may be public. Passwords and keys in the text are redacted.
Reviewing and approving knowledge
Under /app/projects/{id}/knowledge-review you decide for individual documents whether and when the assistant may use them:
- Responsible person, rule topic (up to 120 characters) and authoritative value (up to 255 characters), for example “Return period: 30 days”.
- Valid from and through.
- Approval – “Under review – exclude from AI” or “Approve this content”.
The assistant only uses a document managed this way with a matching approval and only within its validity period. If its content changes, it needs a new approval. Your account does not accept two approvals on the same topic with different values for the same period. Changes take effect after about five minutes.
Improved answers
Under /app/projects/{id}/answers you define the right answer for a question: question up to 300 characters, answer up to 3000 characters. It becomes a source of its own that the assistant finds with priority, and it takes effect after a few minutes. You can also get there via Improve answer in the AI log and in the knowledge gaps.
Knowledge gaps
Under /app/projects/{id}/knowledge-gaps you find questions from the last 90 days for which the assistant found no source, which it passed to your team or which got a thumbs down, grouped by topic. For each topic you can:
- Improve answer – create an improved answer,
- Create article – in the help centre, with an AI draft if you like; the draft stays a draft until you publish it,
- Done, Ignore or Reopen.
AI log
Under /app/projects/{id}/ai-log you see every answer from the assistant with its sources and the result of the check. reactive.chat compares numbers, weekdays and statements such as “free of charge” with the sources. If something does not match, the draft is replaced by a quote, a fixed sentence or a handover. The log is a view for checking; nothing can be set there.
Fixed answer to the storage question
If a visitor asks whether the chat is stored, reactive.chat answers with a fixed, checked text instead of a freely worded answer. You influence it via the privacy settings: the link to your privacy policy appears in it, and if deletion via the chat is set to “off”, the note on how the visitor can have their data deleted is missing.
An sabunta ta ƙarshe: