Documentation/AI Knowledge Base & Training

Knowledge Ingestion & Guardrails

AI Training & Knowledge Base

Teach your Talk2Site AI agent everything about your business. Ground its answers in your website URLs and documents to guarantee 99%+ answer accuracy.

Start Quick Setup Journey Installation Guide Integrations Guide

Knowledge Ingestion Overview

Four components that power reliable, hallucination-free answers

Automated

Website URL Crawling

Our crawler scans your public website pages, cleans HTML tags, and indexes text into vector memory.

PDF / DOCX

Document Ingestion

Upload product brochures, price sheets, manuals, or internal FAQs to provide deep product context.

Guardrails

System Prompt Rules

Set tone, boundary constraints, preferred terminology, and fallback handling to eliminate hallucinations.

Sync

1-Click Retraining

Update content anytime. Retraining re-indexes your changed URLs and documents in seconds.

Training In 5 Simple Steps

Follow this sequence in your agent dashboard to vectorize your business knowledge base.

Provide your public website URL

Enter your domain (e.g. yourcompany.com). The crawler indexes your homepage, product pages, pricing, and FAQ pages automatically.

Follows internal links within your domain automatically

Strips out header/footer boilerplate to focus on valuable page copy

Crawls up to 50+ pages on standard accounts with zero manual work

Upload PDFs, Word docs, and spreadsheets

Have offline catalogs, onboarding documents, or return policies? Upload them directly to your agent's knowledge base.

Supported formats: .pdf, .docx, .txt, .csv, and markdown

Automated document chunking ensures precise passage retrieval

Upload multiple files up to 25MB each simultaneously

Define system prompt & guardrail instructions

Instruct the AI on its personality, brand tone, strict answer limits, and when to ask for visitor contact information.

Set business guidelines: 'Only answer questions based on the provided documents.'

Define lead qualification triggers: 'When visitors ask about custom pricing, ask for their name and email.'

Control tone of voice: Friendly, professional, concise, or consultative

Test responses in the live dashboard preview

Open the Test Agent tab to query the chatbot with common customer questions and verify answer accuracy before going live.

Inspect source citations and verified groundings for every answer

Simulate edge cases, pricing inquiries, and meeting booking requests

Adjust prompt rules or add missing facts immediately if needed

Trigger retraining whenever content changes

Whenever you update prices, launch a product, or modify policies, hit 'Retrain Agent' to sync knowledge immediately.

Delta indexing ensures only modified pages and files are re-processed

Live widget updates automatically without needing to re-embed code

Zero downtime for website visitors during knowledge retraining

Recommended System Prompt

Paste this base template into your Agent Prompts tab.

You are the official AI assistant for [Your Company Name]. Your goal is to answer visitor questions accurately, guide them through our services, and collect lead details for our sales team. Rules: 1. Only answer based on the verified website content and uploaded documents. 2. If you do not know the answer, politely suggest booking a consultation or contacting support at support@company.com. 3. Be friendly, concise, and helpful. Keep responses under 3 sentences whenever possible. 4. When a visitor expresses buying interest or asks for pricing, offer to schedule a call or request their email address.

Tip: Customize the company name, support email, and appointment links to match your brand.

Hallucination Prevention Rules

Talk2Site combines semantic search vector retrieval with strict negative prompting to ensure your bot never invents facts:

Grounding Constraint

The model evaluates document similarity scores before responding.

Honest Fallbacks

If knowledge is missing, the bot directs visitors to your support team.

Instant Retraining

Whenever pricing or terms change, update documents and re-index.

Need custom prompt assistance?Contact Support →

Next: Connect Your Store & Calendars

Now that your AI agent is trained on your content, sync your Shopify or WooCommerce catalog and enable Google Calendar booking.

Go to Integrations Guide Embed Widget