PradžiaDokumentacija

Knowledge Base

The Knowledge Base is where your chatbot learns. Add content from various sources to make your AI smarter and more accurate.

Content Sources

ChatReact supports multiple ways to add knowledge:

SourceDescriptionBest For
Svetainės nuskaitymasAutomatically extract content from your websiteExisting documentation, blog posts
DUKQuestion and answer pairsCommon customer questions
Core InstructionsSpecific rules and guidelinesBusiness logic, policies
DokumentaiUpload PDFs, docs, etc.Manuals, guides

Knowledge Setup Wizard

Right after you create a chatbot, ChatReact opens a guided knowledge setup wizard with three steps: Scan Website, Upload Documents, and Atlikta. It uses the same website scan and document upload flows described on this page.

You can skip either content step with Vėliau — but without content your chatbot will only give generic answers, so we recommend adding at least one source. You can always come back and add more content from the Knowledge Base later.

Overview Tab

The Knowledge Base page has four tabs: Overview, Instrukcijos, Svetainės, and Dokumentai. Overview is the default landing tab and gives you an at-a-glance dashboard:

  • Žinių bazės būsena — counts of Core Instructions, Improvement Instructions, crawled pages, and documents. Click any card to jump straight to that tab.
  • Vektorinė saugykla — shows whether the vector store is Aktyvi or Neinicializuota, plus the total storage size currently used.
  • How Knowledge Base Works — a short explainer on how Core Instructions and Documents are used together to answer questions.

Website Crawler

Importing Your Website

Website import is a three-step flow on the Svetainės tab: preview your site structure, select pages, then import the content.

  1. Navigate to Žinių bazė in the sidebar and open the Svetainės tab
  2. Enter your website URL (e.g., example.com)
  3. Optionally add Exclude URL patterns — one pattern per line, using * as a wildcard (e.g., /gallery/*)
  4. Click Peržiūrėti puslapius — ChatReact scans your site and lists the pages it finds
  5. Select the pages to import. AI-recommended pages are preselected and marked with a star — click AI Selection at any time to restore that recommendation. Your plan determines how many pages you can select.
  6. Click Importuoti {count} Puslapius (the count is replaced by the number of pages you've selected) to start the import

Crawl Limits

The number of pages you can crawl and crawls per month depends on your plan. See our Pricing page for details.

Crawl Status

Monitor your crawl progress:

  • Laukia: Waiting to start
  • Vykdoma: Currently crawling pages
  • Baigta: All pages processed
  • Nepavyko: Error occurred (check details)
  • Atšaukta: Crawl was stopped before finishing

Crawling runs in the background — you can leave the page and return later to check progress. The progress indicator updates in real-time.

Cancel a Crawl

You can cancel a running or pending crawl at any time:

  • Click the Atšaukti button in the progress bar during an active crawl
  • In Naujausi nuskaitymai, each running/pending job also has a Cancel button
  • If you leave the page during a crawl, the Cancel button is still available when you return

You'll receive a notification when your crawl completes. Click the notification to view results.

Viewing Crawl Results

After an import completes, your pages appear under Nuskaitytos svetainės:

  1. View all imported pages with their titles
  2. Click the eye icon (Peržiūrėti indeksuotą turinį) to see the extracted content rendered as formatted Markdown
  3. Search pages by URL or title to find a specific page quickly
  4. Use the checkboxes to select multiple pages and delete them in one go
  5. Re-crawl individual pages or the entire website

Pages that haven't been refreshed in over 30 days show an age badge (e.g., "45d old") so you can see at a glance when content is due for a re-crawl.

Re-Crawling

To update your knowledge base with fresh content:

  • Single page: Click the refresh button next to any crawled page
  • Entire website: Use the refresh-all button for the whole domain

Re-crawls update the content and refresh the "last crawled" timestamp. You'll receive a notification when complete. You can also cancel a re-crawl in progress.

Refreshing a domain updates its existing imported pages. Run Peržiūrėti puslapius again to discover newly added URLs. Refreshes are started explicitly; there is no recurring website-scan schedule to configure in the dashboard. After updating, use Peržiūrėti indeksuotą turinį and test a representative question in Peržiūra.

Pages that no longer exist on your website (HTTP 404/410, or a page that only says "page not found") are not indexed during a re-crawl. They are marked orange as Nerasta, removed from the knowledge base and no longer used by the chatbot. The website header shows how many pages were not found; click the orange counter to select them all and remove them with Delete selected. If a page comes back, the next re-crawl indexes it again.

Adding a Single URL

You don't need a full crawl to add one page. Click Pridėti URL in the Crawled Websites toolbar, paste the page address, and ChatReact imports just that page.

CMS Integrations

If your website runs on WordPress, connect the ChatReact WordPress plugin for smarter, faster scans enriched with content-type information:

  1. On the Svetainės tab, find the CMS Integrations card
  2. Click Generate Integration Key and copy the key (it's shown only once)
  3. Paste the key into the plugin's Integration tab in WordPress
  4. Use Scan Website to import your content

The WordPress plugin lets you choose which post types are available for scans. Select your content types in the plugin before scanning your site here.

See the WordPress guide for full plugin setup.

Documents

For current pricing, upload your latest price list as a text file or a text-based PDF under Žinių bazė → Dokumentai. For scanned or image-only PDFs, prepare the content manually as text first; do not rely on OCR. Preserve units, tax information and the effective date, wait for processing to finish, then check a chatbot answer against the source. Uploading does not guarantee retrieval or priority over other sources. This manual workaround does not require automatic PDF website crawling.

Upload your own files on the Dokumentai tab:

  1. Drag and drop files into the upload area, or click to select them
  2. Supported formats: PDF, TXT, Markdown, CSV, JSON, DOCX (max 10 MB per file)
  3. Each file shows its processing status while it's being added to the knowledge base
  4. Open or delete uploaded documents at any time

Besides the 10 MB per-file limit, your total document storage is capped by your plan (for example 10 MB total on the Free plan, more on paid plans — see our Pricing page). Once you reach that quota, new uploads are blocked until you delete some documents or upgrade.

Good candidates: product descriptions, FAQ lists, pricing tables, terms of service, support guides, and other text-based documentation.

FAQs

FAQs are the most direct way to teach your chatbot:

Adding FAQs

  1. Go to DUK in the sidebar
  2. Click Pridėti DUK
  3. Enter the question (as a customer would ask it)
  4. Enter the answer — the rich-text editor supports formatting
  5. Optionally assign to a category
  6. Click Sukurti DUK

FAQ Categories

Organize FAQs into categories:

  • Pricing
  • Technical Support
  • Account Management
  • Shipping & Returns

FAQ Suggestions

Generate FAQ suggestions on demand:

  • Generuoti iš Žinių bazės — use the button on the Suggestions tab of the FAQs page
  • Patobulinimų agentas — the agent can suggest FAQs as part of its improvement proposals
  • Manual — create FAQs yourself at any time

Review suggestions and approve or dismiss them. FAQ suggestions are available on paid plans — see our Pricing page. See the FAQ Management guide for details.

Core Instructions

Core Instructions are specific rules your chatbot must follow:

Adding Instructions

  1. Navigate to Žinių bazė in the sidebar
  2. Open the Instrukcijos tab
  3. Click Pridėti pagrindinę instrukciją
  4. Write your instruction clearly in the editor (formatting and links supported)
  5. If your plan includes the Email Assistant (Professional and up), choose Taikoma: Tik pokalbiui or Pokalbis + el. paštas. "Chat + Email" instructions also apply as general rules to every email reply draft; how a specific kind of email is answered is set per category in the Email Assistant. On other plans this choice is not shown and instructions apply to the chat.
  6. Toggle active/inactive as needed

Core Instructions override document content. Use them for critical facts that must always be correct — prices, policies, contact info.

Improvement Instructions

The Instrukcijos tab also has a separate, read-only Patobulinimų nurodymai section. These are created automatically by the Improvement Agent based on chat feedback — you can't edit them directly, but each one links back to the improvement session that created it. See the Improvement Agent guide for details.

Effective Instructions

Good examples:

  • "If a customer asks about pricing, direct them to our pricing page at /pricing"
  • "Never disclose internal metrics or customer data"
  • "For technical issues, always ask for the error message first"

Avoid:

  • Vague instructions: "Be helpful"
  • Contradictory rules
  • Overly complex logic

Smart Content Search

ChatReact searches your knowledge base by meaning, not just keywords. Your chatbot finds and references the right information even when customers phrase their questions differently from how your content is written.


Best Practices

  1. Start with FAQs: Add your top 20 most common questions
  2. Crawl strategically: Target documentation pages, not marketing fluff
  3. Review regularly: Check what questions your bot can't answer
  4. Iterate: Use the Improvement Agent to refine based on feedback

API Access

Documents (upload, list, delete), website crawling (preview, start, check status, cancel), crawled pages, and Core Instructions can all be managed programmatically through the REST API or the MCP Server — useful for syncing content from an external CMS or automating knowledge base updates.


Next Steps