Customer chat · AI Brain

The customer chat lets you embed an AI support assistant on your website or app. It answers questions using your knowledge base, reflects your brand personality, and handles multi-turn conversations.

How it differs from the owner copilot

Feature	Owner copilot	Customer chat
Who uses it	You (the business owner)	Your customers
Tools / actions	Yes — can create posts, check analytics, etc.	No — knowledge-only
Knowledge base	Optional (included in context)	Always consulted
Streaming	Yes (SSE token-by-token)	No (full response returned)
Conversation memory	Full — persisted, history loaded	Session-only — no cross-session memory
Personality	Uses your copilot context	Uses your personality config (tone, custom instructions)
Auth	Requires owner JWT	No auth — public-facing

Setting up customer chat

Configure your personality (business name, tone, custom instructions) — this becomes the AI's system prompt
Upload your knowledge base documents (FAQs, product info, policies)
Integrate the customer chat endpoint into your frontend

Chat endpoint

POST /brain/chat/customer
Content-Type: application/json
 
{
  "message": "Do you offer international shipping?",
  "history": [
    { "role": "user", "content": "What are your hours?" },
    { "role": "assistant", "content": "We're open Monday–Friday, 9 AM–6 PM IST." }
  ],
  "contactContext": {
    "name": "Priya Sharma",
    "email": "priya@example.com"
  }
}

Field	Required	Description
`message`	Yes	The customer's current message
`history`	No	Previous turns in this session (for multi-turn support)
`contactContext`	No	Known customer details — injected into system prompt for personalisation

Response:

{
  "response": "Yes! We ship to 50+ countries via DHL and FedEx. Standard delivery takes 7–14 business days and express shipping is available for an additional charge. You can see all shipping rates at checkout.",
  "conversationId": "conv_01j..."
}

How knowledge retrieval works

For each customer message, the AI:

Converts the message to an embedding (same model used during ingest)
Runs a cosine similarity search against all your knowledge chunks
Selects the top 3 chunks with a similarity score above 0.30
Injects those chunks as reference material in the system prompt

The AI is instructed to answer from the knowledge base and say "I don't have that information" when the knowledge base doesn't cover the question.

"use client"
import { useState } from "react"
 
type Message = { role: "user" | "assistant"; content: string }
 
export function CustomerChat() {
  const [messages, setMessages] = useState<Message[]>([])
  const [input, setInput] = useState("")
  const [loading, setLoading] = useState(false)
 
  async function send() {
    if (!input.trim() || loading) return
    const userMsg: Message = { role: "user", content: input }
    const history = [...messages, userMsg]
    setMessages(history)
    setInput("")
    setLoading(true)
 
    const res = await fetch("/api/chat", {
      method: "POST",
      headers: { "Content-Type": "application/json" },
      body: JSON.stringify({ message: input, history: messages }),
    })
    const { response } = await res.json()
    setMessages([...history, { role: "assistant", content: response }])
    setLoading(false)
  }
 
  return (
    <div className="chat-container">
      <div className="messages">
        {messages.map((m, i) => (
          <div key={i} className={`message ${m.role}`}>{m.content}</div>
        ))}
        {loading && <div className="message assistant">Thinking…</div>}
      </div>
      <input
        value={input}
        onChange={(e) => setInput(e.target.value)}
        onKeyDown={(e) => e.key === "Enter" && send()}
        placeholder="Ask a question…"
      />
      <button onClick={send}>Send</button>
    </div>
  )
}

Your Next.js route handler (/api/chat/route.ts) proxies to the Vlozi brain-service:

// app/api/chat/route.ts
import { NextRequest, NextResponse } from "next/server"
 
export async function POST(req: NextRequest) {
  const body = await req.json()
  const res = await fetch(`${process.env.VLOZI_API_URL}/brain/chat/customer`, {
    method: "POST",
    headers: {
      "Content-Type": "application/json",
      "x-tenant-id": process.env.VLOZI_TENANT_ID!,
      // gateway key is server-side only — never expose to browser
    },
    body: JSON.stringify(body),
  })
  const data = await res.json()
  return NextResponse.json(data)
}

What the AI will and won't say

The AI will:

Answer questions covered by your knowledge base
Apply your personality tone and custom instructions
Say "I don't have that information" when the question isn't in the knowledge base
Use the customer's name if you pass it in contactContext

The AI will not:

Access external websites or real-time data
Take actions (it has no tools in customer chat mode)
Remember previous sessions (each session is stateless unless you pass history)
Reveal the system prompt or knowledge base contents verbatim

Conversation history

The history field is optional. If you pass previous turns, the AI maintains context across the conversation. If you start a new session without history, each message is treated independently.

Managing history in your frontend: keep the messages array in state. On each response, append both the user message and the assistant response. Pass the full messages array (excluding the current user message) as history on the next call.

Rate limits

Customer chat requests are rate-limited per tenant. If your volume exceeds the rate limit, you'll receive a 429 response. Contact support to increase limits for high-traffic deployments.