Multi-Channel AI Customer Support for $85/mo
4
Channels
<2s
Avg Response
$85
Cost/mo
80%
AI Resolution
AI Support Agent
AI Shopping Assistant
"Show me trending sneakers"
From incoming message to AI response — across any channel.
Customer sends a message via WhatsApp, Telegram, Email, or Web Chat
Django backend normalizes it into a UnifiedMessage format
Claude Haiku classifies the ticket (billing, technical, account, general)
Escalation check: low confidence, negative sentiment, or human request?
pgvector searches your knowledge base for the top 3 relevant chunks
Claude Sonnet generates a response constrained to KB context only
3-layer guardrails scan for hallucinated prices, policies, or guarantees
Response sent back through the original channel
Incoming Channels
Cloud API v22.0
Gmail OAuth2
Telegram
Bot API
Web Chat
WebSocket
Django Backend
DRF + Django Channels + Daphne
Claude Haiku
Ticket classification — $1/M tokens
Normal Path (80%)
pgvector Search
Cosine similarity — top 3 KB chunks
Claude Sonnet
Response generation — $3/M tokens
Guardrails Scan
3-layer anti-hallucination check
Escalation Path (20%)
Escalation Detector
Confidence < 0.7 | Sentiment | Human request
Handoff Package
AI summary + suggested response
Human Agent
Takes over with full context
Operations
Dashboard
Next.js 15 + Tailwind
Knowledge Base
Upload & manage docs
Analytics
AI vs human resolution
Team Settings
Multi-tenant + roles
Everything you need to run AI-powered customer support in production.
WhatsApp, Telegram, Email (Gmail), and Web Chat — all normalized into a single UnifiedMessage format. One AI brain handles every channel.
Upload your docs (.md, .txt, .pdf). Auto-chunked with smart boundary detection, embedded with OpenAI, and searched via pgvector cosine similarity. AI only answers from your KB.
Claude Haiku for fast ticket classification (billing, technical, account, general). Claude Sonnet for high-quality response generation with strict guardrails.
System prompt constraints + empty RAG fallback + post-generation scan for sensitive patterns (prices, refund policies, SLAs). No fabricated answers.
Three triggers: low AI confidence (<0.7), negative sentiment (weighted keyword scoring), or customer requests a human. Clean handoff with AI summary + suggested response.
Ticket queue with filters, conversation view with AI sidebar, analytics (AI vs human resolution rates), knowledge base manager, canned responses, and channel settings.
Built for teams that want 24/7 support without the $990/mo price tag.
Small businesses replacing expensive support tools like Intercom ($990/mo) or Zendesk ($1,500/mo)
Startups needing 24/7 customer support without hiring a full team
E-commerce stores automating billing, shipping, and product FAQ responses
SaaS companies handling technical support and account questions
Agencies building white-label support solutions for clients
Django + Claude AI + pgvector — battle-tested and self-hosted.
One Docker Compose command gets you from zero to live support agent. Full details in the GitHub README.
git clone https://github.com/codewithmuh/ai-support-agent.git && cd ai-support-agentAdd your API keys: Anthropic (Claude), OpenAI (embeddings), and channel credentials (WhatsApp, Gmail, Telegram)
cp .env.example .envStarts PostgreSQL + pgvector, Redis, Django backend (Daphne), and Next.js dashboard
docker compose up -d --buildGo to localhost:3000/knowledge-base, upload your support docs (.md, .txt, .pdf). The system auto-chunks and embeds them.
Dashboard Settings page — add your WhatsApp Cloud API credentials, Gmail OAuth, and Telegram bot token. All configured via UI.
Send a message on any connected channel. Watch the AI classify, search your KB, and respond. Try triggering an escalation with an angry message.
Deploy the AI Support Agent on your own server. Open source, 4 channels, $85/mo instead of $990+. One Docker Compose command.
Built by CodeWithMuh