Setup guide for configuring your chatbot, connecting knowledge sources, publishing the widget, and understanding what your visitors will experience.
Table of Contents
1. Sign In and Access
Sign in to your account to open the admin dashboard. This is where you configure your chatbot, manage its knowledge, and generate the code used on your website.
2. What Each Tab Does
The admin panel is organized into tabs. Each tab controls a different part of setup or day-to-day operation:
- General: Controls the assistant’s response creativity, cache duration, allowed website URLs, search stop words, and main system prompt.
- Knowledge Files: Stores uploaded documents converted into text knowledge, with optional per-file search collections.
- Knowledge Bases: Connects dynamic sources such as XML feeds, RSS feeds, or GitHub repositories for indexing.
- Routing: Decides which collection is searched first based on keywords, domain rules, collection intention, and score thresholds.
- Widget: Controls how the chatbot looks, which language texts it uses, what users see when no answer is found, and how the chat is embedded on your site.
- Logs: Lets you filter and review saved conversations, then export them as CSV for analysis.
- Profile: Stores your contact details, password change options, and monthly usage information.
3. Configure Basic Behavior
- Response creativity: Use the slider in General to control how focused or varied responses should be. Lower values are more consistent. Higher values are more flexible.
- Cache TTL: Set how long answers remain cached. Use 0 to disable caching. Use the cleanup button if you want to remove expired cache entries manually.
- Allowed URLs: Add one website URL or domain per line for sites that are allowed to load your widget. This setting matters during deployment because an empty list blocks embedded use. If the page is not on your allowed list, or if the allowed list is empty, the embedded widget will not run.
- Search stop words: Add words that should be ignored when the system builds keyword searches.
- Main system prompt: Write the core instructions that define tone, rules, business limits, and how the assistant should use your knowledge.
4. Add Knowledge Sources
- Knowledge Files: Upload PDF, TXT, CSV, XLS, XLSX, DOC, or DOCX files. Uploaded files are converted into text knowledge for indexing.
- Output file naming: You can optionally rename the converted file during upload. The saved knowledge file must end in .txt.
- Edit and structure content: Inside the file editor, you can separate content by paragraph markers, Markdown headings, or blank lines so the search index can split the content more accurately.
- Per-file collections: Each knowledge file can use the default collection or a specific search collection.
- Knowledge Bases: Add live sources from XML, RSS, or GitHub. See the API Reference for XML format details.
- GitHub sources: You can index public repositories without a token, or private repositories with a GitHub personal access token.
- File masks: For GitHub sources, you can limit indexing to selected file extensions such as md, php, json, or txt.
- Incremental updates: Enable this when your source supports modified timestamps and you want faster re-indexing of changed items only.
- Split items: Leave this enabled for documentation and long-form knowledge. Disable it for catalogs or records that should stay as one indexed item.
- Examples of connections for data retrieval: The Knowledge Bases tab includes ready-made script examples for connecting to WordPress, WooCommerce, OpenCart, WHMCS, and general XML export feeds.
5. Set Up Routing and Search
- Collections: Organize your knowledge into collections such as support, products, billing, or orders.
- Routing keywords: Add keywords to collections so matching queries prioritize the right collection.
- Domain routing: You can tell a collection to do nothing special, be ignored for a domain, become the priority for a domain, or be used only when the page domain exactly matches a keyword.
- Collection intention: Use intention types carefully. For example, an order collection can trigger order-related prompting behavior before order lookup.
- Relevance threshold: Raise the threshold for stricter matching. Lower it if you want broader recall.
- Priority order: Reorder collections so the most important sources are checked first.
- Test Routing: Simulate a real user question before publishing so you can confirm the correct collection wins.
- Test Search: Run manual searches against one collection or all collections to see what content the system is likely to retrieve.
Intention Logic (How Routing Decides)
Before knowledge retrieval, the system runs a lightweight intention classification for each user message. The result is one of four labels:
- human: User asks to speak with a human/agent/support.
- order: User asks about their invoices, orders, balance, or other personal order/account financial details.
- eshop: User asks about a product.
- none: Any other general query.
6. Customize the Visitor Experience
- Appearance: Set widget position, theme preset, typing animation speed, rounded corners, colors, header logo, and button icon.
- Multilingual texts: Configure title, input placeholder, send button text, initial greetings, thinking details, start-over confirmation, new conversation button text, and conversation-cleared message for each supported language.
- Initial greeting behavior: You can add multiple greetings for a language. The widget picks one at random, which keeps the opening experience from feeling repetitive.
- No-answer behavior: Customize what users see when the bot cannot find relevant knowledge.
- Support escalation: Choose whether the bot asks for contact details and sends an email to your support address, or simply shows a fallback contact message.
- Compliance texts: Add a disclaimer and an optional privacy-policy link for the visitor-facing widget.
- Business profile data: You can allow the assistant to share business-owned contact or company identifiers when a visitor asks, while still keeping customer-specific data blocked.
- Language behavior: The widget can auto-detect the visitor’s browser language or be forced to a specific language in the embed settings.
7. Publish the Widget
- Generate the widget API key: In the Widget tab, generate the API key and save it before copying your embed code.
- Embed code: Copy the generated script tag and place it before the closing </body> tag on your website.
- Full-page chat link: If you prefer, you can also use the generated full-page chat URL instead of the floating widget.
- Appearance in embed: Use the Include Appearance Settings in Widget option if you want the embed script to carry your saved colors and visual settings.
- Language in embed: You can force a specific language in the embed settings, or leave it on auto-detect.
- Automatic page context: The loader sends the current page URL to the chat system. This helps with routing and allowed-domain checks.
- Recommended go-live test: After embedding, open a real page on your site, confirm the widget loads, confirm the correct language appears, and ask 3 to 5 realistic customer questions.
8. Monitor, Maintain, and Improve
- Logs: Use date filters and quick ranges to review conversation history, then export results to CSV when you need offline analysis.
- Profile: Keep your email, company, phone, VAT, and address current, and change your password from the Profile tab when needed.
- Usage: Review your monthly usage progress in Profile so you can see how close you are to your response limit.
- Indexing workflow: Re-index after major content changes. If automatic indexing is enabled, scheduled sync can keep dynamic sources fresh.
- Best practice for collections: Separate collections by business purpose, such as products, support, billing, or orders, instead of mixing everything together.
- Best practice for source text: Keep source content clean, plain, and direct. Remove unnecessary markup and avoid burying key facts inside long paragraphs.
- Best practice for routing: Start with a small number of clearly named collections, test with real customer questions, then expand routing rules only when needed.
- Best practice for visitor trust: Write friendly fallback messages, add a clear support escalation path, and keep your disclaimer and privacy link current.