Documentation

BolChat Platform Guide

Everything you need to build, optimize, and deploy your AI support agent — from first upload to production.

Quick Start

Getting Started

Go from zero to a live AI agent on your website in under 10 minutes. No coding experience required.

BolChat works in four simple steps. First, you create a knowledge base by uploading your company documents. Then you build an agent that uses that knowledge to answer questions. Next you customize how the widget looks. Finally, paste a single script tag and go live.

1

Create a Knowledge Base

Upload PDFs, DOCX files, and FAQs. BolChat converts them into searchable AI knowledge vectors.

2

Build Your Agent

Name it, write a system prompt that defines personality and rules, and link your Knowledge Base.

3

Customize the Widget

Colors, shapes, icons, teaser messages, and quick replies — match your brand perfectly.

4

Deploy to Your Website

Copy the <script> tag from the Deploy tab, paste into your HTML — your agent is live.

How It Works — The Platform Flow

Your documents are chunked into smaller pieces and converted into mathematical vectors. When a visitor asks a question, the AI searches these vectors for the most relevant context, then generates a precise answer using only the matched content. This is called RAG (Retrieval Augmented Generation) — it prevents the AI from making things up.

📄 Documents → 🧮 Vectors → 🧠 Agent → 💬 Widget → 👤 Visitors

You can always refine your agent after deploying. Use the Analytics tab to see which questions stumped the bot, then update your knowledge base to fill gaps.

Make sure documents are up-to-date before deploying. The AI will confidently cite outdated information if stale docs remain in your knowledge base.

Knowledge Base

Building a Powerful Knowledge Base

Your knowledge base is the foundation. Better data = better answers, fewer hallucinations, happier customers.

Preparing Your Documents

The quality of your AI agent depends entirely on the quality of data you feed it. Clean, well-structured documents produce accurate, reliable answers. Messy data produces inconsistent, sometimes incorrect responses.

PDF

Best — clean, structured

DOCX

Good — preserves headings

TXT

Fast — minimal formatting

Aim for 5–50 pages per document. Too large = noisy chunks where details get lost. Too small = missing surrounding context the AI needs for complete answers.

Remove headers, footers, page numbers, and watermarks from PDFs before uploading. Repetitive elements pollute every chunk and degrade search accuracy.

What to Upload — Content Strategy

Think of your knowledge base as the agent's brain. If information isn't in the KB, the agent can't answer questions about it. Prioritize these:

  • Product and service descriptions with detailed specs and pricing — include all tiers and plans
  • FAQs your support team answers daily — highest-value docs because they directly match user queries
  • Onboarding guides, how-to manuals, and step-by-step tutorials
  • Company policies — shipping times, return windows, refund criteria, privacy, terms of service
  • Your actual website copy — product pages, landing pages, about page. The bot should know everything your site says
  • Case studies and testimonials — helps the bot sell by sharing relevant success stories
  • Competitor comparison documents — 'Why choose us over X' content handles competitive questions professionally

What NOT to Upload

  • Internal HR documents or employee handbooks (unless the bot serves internal employees)
  • Raw spreadsheets, financial reports, or data tables — AI cannot reliably parse tabular data
  • Documents that are primarily images, scanned PDFs, or infographics — OCR extraction is limited
  • Outdated product information or sunset feature docs — the AI will confidently cite old info
  • Confidential documents you wouldn't want exposed to website visitors

The AI treats all uploaded documents as ground truth. If you upload a 2022 pricing sheet alongside current 2024 pricing, the bot may cite either. Always remove obsolete documents.

Content Quality & Chunking Tips

BolChat splits your documents into smaller "chunks" for vector search. How you structure content directly impacts retrieval quality.

1

One Topic Per Document

Don't merge pricing, shipping, and returns into one file. Dedicated docs per topic produce precise, focused answers.

2

Use Clear Headings (H1/H2/H3)

The AI uses headings to understand structure. Well-organized docs dramatically improve retrieval accuracy.

3

Write in Q&A Format

Structure content as questions and answers when possible. It directly mirrors how visitors ask.

4

3–5 Sentences Per Paragraph

Keep paragraphs focused on a single idea. Walls of text mix multiple topics, reducing accuracy.

Keeping Your Knowledge Fresh

A knowledge base is never "done" — it's a living system. Follow this maintenance schedule:

  • Monthly review: Delete outdated docs, re-upload updated versions. Check if products, services, or prices changed.
  • After every product launch: Upload new docs before you go live — your bot should be ready from day one.
  • Version your files: Name them with dates (pricing-2024-q4.pdf, return-policy-v3.pdf) to spot stale content easily.
  • Monitor Analytics: Frequent 'I don't have enough information' responses = your KB has gaps that need filling.

Multilingual Knowledge

BolChat auto-detects visitor language and responds accordingly. However, response quality depends on your KB:

  • Upload documents in each language you support. English KB + Spanish query = on-the-fly translation, which works but isn't as precise.
  • Avoid machine-translated documents. Google Translate output introduces awkward phrasing the AI replicates verbatim.
  • Prioritize your top 2–3 languages. If 80% speak English and Hindi, complete your KB in both first.

Measuring Knowledge Quality

📊 Fallback Rate

If more than 15% of responses are fallbacks, your KB has significant gaps. Target under 5%.

🧪 The 20-Question Test

Before publishing, ask 20 common customer questions. If it can't answer 18+, iterate on your KB.

Upload your actual website copy as a document. Visitors ask about things they see on your landing page — the bot should know everything your site says.

Agent Manager

Creating High-Performance Agents

Your agent is the personality and brain of your chatbot. The system prompt is the single most important config.

Writing Effective System Prompts

The system prompt defines who your agent is, how it speaks, what it knows, and what it should never do. A great prompt = a great agent. A vague prompt = a generic, unhelpful bot.

❌ Bad — Too Vague

"You are a helpful assistant."

✅ Good — Specific & Bounded

"You are a customer support agent for [Company Name]. You specialize in helping users with product questions, pricing inquiries, and order tracking. Rules: - Always be polite, concise, and professional - Only answer using information from your knowledge base - If you don't know something, say 'Let me connect you with our team at support@company.com' - Never discuss competitor products or provide legal/medical advice - Keep responses under 150 words unless the user asks for details - Use the user's language (auto-detect) Tone: Friendly and approachable, like a knowledgeable colleague."

System Prompt Checklist

Every effective prompt answers these five questions:

1

Who is the agent?

'You are a [role] for [Company].' Define the role clearly.

2

What should it do?

List tasks: answer questions, track orders, book demos, qualify leads.

3

What should it NOT do?

'Never give refunds', 'Don't discuss competitor pricing'.

4

How should it sound?

Formal/casual? Emoji or not? Short or detailed? Match your brand.

5

What if it doesn't know?

'Say: Let me connect you with our team' — never make things up.

Prompt Templates for Common Use Cases

🛒 E-Commerce Support

Handles orders, returns, shipping, product recs. Temp: 0.2. Best with FAQ-style KB covering return policy, shipping times, product catalog.

💻 SaaS Onboarding

Guides new users through setup. Temp: 0.3. Works best with step-by-step tutorials and getting-started guides.

🎯 Lead Qualifier

Asks qualifying questions before offering demos. Temp: 0.4. Include pricing tiers, feature comparison, case studies.

📚 FAQ Bot

Strictly answers from KB, zero improvisation. Temp: 0.1. Highest confidence. Upload only verified FAQ documents.

Temperature & Confidence Tuning

These two settings control creativity vs. accuracy. Getting them right is the difference between a reliable agent and an erratic one.

🌡️ Temperature (0.0–1.0)

Lower = consistent and factual. Higher = varied and creative.

  • 0.1–0.3: Strict, factual. Best for support/FAQ bots.
  • 0.4–0.6: Balanced default for most agents.
  • 0.7–1.0: Creative. Only for casual/marketing bots.

🎯 Confidence Threshold

How sure the AI must be before using a KB chunk.

  • 0.8+: Fewer hallucinations, more fallbacks.
  • 0.6–0.7: Balanced. Recommended start.
  • 0.3–0.5: More answers, risk of inaccuracy.

Start with Temperature 0.3, Confidence 0.7. Test 20+ questions. Too rigid? Increase temp slightly. Hallucinating? Increase confidence.

Conversation Design

Greeting, quick replies, and teaser are your agent's first impression. Make them count.

❌ Generic Greeting

"Hello, how may I assist you today?"

✅ Specific & Helpful

"Hi! 👋 I'm [Company]'s AI assistant. I can help with product questions, pricing, and order tracking. What can I help you with?"

  • Quick replies: Add 2–3 common starting points — 'Show pricing', 'Track order', 'Talk to sales'. Reduces friction dramatically.
  • Teaser message: Use something business-specific — '📦 Need help tracking your order?' converts better than 'Chat with us'.
  • Keep greetings under 30 words. Long greetings intimidate visitors and reduce engagement.

Multi-Agent Strategy

Different pages serve different purposes. Use separate agents optimized for each:

Sales Bot

Pricing pages. Warm tone. Lead capture ON.

Support Bot

Help center. Strict KB-only. Professional.

Onboarding Bot

Post-signup. Step-by-step. Friendly.

Test 20+ conversations before publishing. Test edge cases, multilingual queries, and tone consistency. Bad prompts = bad first impressions.

Widget

Widget Customization

Make the widget feel native to your website with shapes, colors, animations, and smart behaviors.

1

Brand Color & Icon

Match your brand. 8+ icons: brain, sparkles, bot, chat, zap, help, headphones, message.

2

Launcher Shape

5 shapes: Circle, Rounded, Pill (with text), Square, Flower (organic blob).

3

Teaser Message

Intercom-style speech bubble from the launcher. Fires once per session for proactive engagement.

4

Quick Replies

Up to 3 clickable pill buttons above the input. Reduce visitor friction immediately.

Premium Effects (Optional)

🪟 Glassmorphism

Frosted glass with blur

🌈 Gradient

Rotating brand gradient

🚀 Entrance

Slide, scale, or bounce

The pill shape with text has the highest click-through rate. Combine with a teaser message for maximum engagement.

Integration

Integration Guides

Add BolChat to any website in under 2 minutes. Works with every framework and CMS.

Go to the Deploy tab in your dashboard. Copy your API key and Agent ID. Replace the placeholders below with your actual values.

Plain HTML

Paste before the closing </body> tag:

HTML
<script
  src="https://api.bolchat.tech/static/widget.js"
  data-key="YOUR_API_KEY"
  data-agent="YOUR_AGENT_ID"
  data-api-url="https://api.bolchat.tech"
></script>

Next.js (App Router)

Add to your root layout.tsx:

TypeScript — app/layout.tsx
import Script from "next/script";

export default function RootLayout({ children }) {
  return (
    <html lang="en">
      <body>
        {children}
        <Script
          src="https://api.bolchat.tech/static/widget.js"
          data-key="YOUR_API_KEY"
          data-agent="YOUR_AGENT_ID"
          data-api-url="https://api.bolchat.tech"
          strategy="afterInteractive"
        />
      </body>
    </html>
  );
}

React (Vite / CRA)

Inject the script in a useEffect:

TypeScript — App.tsx
import { useEffect } from "react";

function App() {
  useEffect(() => {
    const script = document.createElement("script");
    script.src = "https://api.bolchat.tech/static/widget.js";
    script.setAttribute("data-key", "YOUR_API_KEY");
    script.setAttribute("data-agent", "YOUR_AGENT_ID");
    script.setAttribute("data-api-url", "https://api.bolchat.tech");
    script.async = true;
    document.body.appendChild(script);
    return () => { document.body.removeChild(script); };
  }, []);

  return <>{/* your app */}</>;
}

Angular

Add to src/index.html before </body>:

HTML — src/index.html
<script
  src="https://api.bolchat.tech/static/widget.js"
  data-key="YOUR_API_KEY"
  data-agent="YOUR_AGENT_ID"
  data-api-url="https://api.bolchat.tech"
></script>

Vue.js

Add to index.html before </body>:

HTML — index.html
<script
  src="https://api.bolchat.tech/static/widget.js"
  data-key="YOUR_API_KEY"
  data-agent="YOUR_AGENT_ID"
  data-api-url="https://api.bolchat.tech"
></script>

WordPress & Shopify

WordPress: Go to Appearance → Theme Editor → footer.php, paste before </body>. Or use a plugin like "Insert Headers and Footers".

Shopify: Online Store → Themes → Edit Code → theme.liquid, paste before </body>.

After adding the script, hard refresh (Ctrl+Shift+R) to clear cache. The widget uses Shadow DOM for complete CSS isolation — zero conflicts with your site.

Analytics

Analytics & Insights

Use real conversation data to continuously improve agent performance.

The Analytics dashboard provides real-time metrics on visitor interactions. Use this data to identify knowledge gaps, refine prompts, and measure ROI.

1

Track Sessions

Monitor daily/weekly/monthly session counts and message volumes. Spot peak hours.

2

Identify Gaps

High fallback rates = topics your KB doesn't cover. Fill those gaps.

3

Review Conversations

Read actual chats weekly. Find great answers to reinforce and struggles to fix.

Continuous Improvement Cycle

  • Week 1: Deploy and collect data. Don't optimize yet — just observe patterns.
  • Week 2: Review conversations. Find top 5 unanswered questions. Add those topics to your KB.
  • Week 3: Refine system prompt based on tone/style. Adjust temperature if too rigid or creative.
  • Monthly: Delete outdated KB docs, re-upload current versions, run the 20-question test again.

Set a Monday calendar reminder to review Analytics. Consistent weekly reviews compound into dramatically better performance.

Best Practices

Do's & Don'ts

Golden rules from hundreds of successful BolChat deployments.

📚 Knowledge Base

Do's

  • Upload clean, well-structured PDFs with clear headings
  • One focused document per topic
  • Write content in Q&A format when possible
  • Review and update monthly
  • Name files with dates (pricing-2024-q4.pdf)
  • Test with 20 real questions before publishing

Don'ts

  • Upload scanned images or visual PDFs
  • Merge unrelated topics into one document
  • Use machine-translated documents
  • Leave outdated product info in KB
  • Upload confidential documents
  • Skip testing and go live untested

🧠 System Prompts

Do's

  • Be specific about role, personality, and boundaries
  • Include company name and key product names
  • Define fallback behavior for unknown questions
  • Set word limits for responses
  • Test 20+ conversations before publishing

Don'ts

  • Use vague prompts like 'Be helpful'
  • Let the agent make up answers
  • Forget to define tone and style
  • Skip testing — bad first impressions stick
  • Change prompts in production without testing

🎨 Widget Design

Do's

  • Match brand colors to your website exactly
  • Use specific teaser messages per page
  • Add 2–3 quick reply buttons
  • Use pill shape with text for higher CTR
  • Test on mobile — 60%+ of chats are mobile

Don'ts

  • Use default colors that clash with your site
  • Enable every effect at once — keep it clean
  • Write vague greetings like 'How can I help?'
  • Ignore mobile responsiveness
  • Place the widget on useless pages
FAQ

Frequently Asked Questions

Quick answers to the questions we hear most.

How long does setup take?
Under 10 minutes. Upload docs, create an agent, customize the widget, paste one script tag. Refine later using analytics.
Can I use BolChat on multiple websites?
Yes. Create different agents for different sites, each with unique KB, prompt, and widget config.
Does BolChat work with SPAs?
Yes. The widget loads asynchronously with Shadow DOM for CSS isolation. Works with React, Angular, Vue, Next.js.
How many languages are supported?
50+ with automatic detection. For best results, upload KB docs in your target languages.
What if the bot can't answer?
It falls back to your configured response ('Let me connect you with our team'). Conversations are still logged for gap analysis.
Is my data secure?
Yes. TLS 1.3 in transit, encrypted at rest. Each tenant has fully isolated data. No cross-org sharing.
Can I see what visitors ask?
Yes — the Conversations page shows full transcripts. Analytics provides aggregate metrics.
How does billing work?
Flexible plans based on message volume. All plans include unlimited agents, KBs, and widget customization.

Ready to build your AI agent?

Get started in minutes — no credit card required.