Back to Projects
Client Project • Tricasol

WhatsApp AI Support Agent.

An AI support agent that works on WhatsApp and connects to WHMCS. It can answer customer questions, check domains, manage hosting services, handle invoices, and work with support tickets.

View Repository

Response

Sub-Second

Tools

20+ Actions

Security

Admin Approval

Integration

WHMCS API

The Problem & Solution.

The Problem

Hosting companies receive many repeated support requests. Customers often ask about unpaid invoices, domain availability, passwords, hosting services, and support tickets. Handling all of these requests manually takes a lot of time.

The AI Solution

I built an AI agent that works through WhatsApp and connects directly to WHMCS. Customers can simply send a message and the agent can find information or perform supported actions for them.

How It Works.

When a customer sends a WhatsApp message, it is sent to the backend through Evolution API. FastAPI receives the message and passes it to the AI agent. The agent decides what it needs to do and can use the available tools to get information or perform actions.

[ Customer WhatsApp ]
          │
          │ Message
          ▼
[ Evolution API ]
          │
          │ Webhook
          ▼
[ FastAPI Backend ]
          │
          ├──► [ LangGraph Agent ]
          │
          ├──► [ Conversation Memory ]
          │
          ├──► [ RAG Knowledge Base ]
          │
          └──► [ AI Tools ]
                    │
          ┌─────────┼─────────┐
          ▼         ▼         ▼
      [ WHMCS ]  [ Domains ] [ Tickets ]
          │
          ▼
[ Response sent back to WhatsApp ]

What the AI Agent Can Do.

Domain Management

Check domains, check availability, update nameservers, get EPP codes, enable or disable ID protection, and renew domains.

Hosting Management

View customer hosting products, reset hosting passwords, and generate cPanel login links.

Invoices & Billing

Check unpaid invoices, view customer invoices, and get invoice PDFs.

Support Tickets

Create, read, reply to, close, and reopen support tickets.

Technical Challenges.

1. Conversation Memory

WhatsApp conversations can become very long. Sending the entire conversation to the AI every time uses too many tokens and makes responses slower.

I added context trimming so older messages can be summarized while important information, such as the domain name and customer ID, is kept.

2. Admin Approval for Sensitive Actions

Some actions are too sensitive to let the AI perform automatically. For example, changing a nameserver or resetting a hosting password.

For these actions, the agent asks an admin for approval first. The action only runs after the admin approves it.

3. Fast Responses

The agent uses Groq for fast LLM responses. I also moved longer tasks to background processing so the WhatsApp webhook can respond quickly.

AI Tools.

I built custom tools that allow the AI agent to interact with the company's systems.

search_tricasol_knowledge_basecheck_registered_domainscheck_domain_availabilityupdate_nameserverget_domain_epp_codetoggle_id_protectionrenew_domainget_client_productsreset_hosting_passwordget_cpanel_sso_linkcheck_unpaid_invoicesget_client_invoicesget_invoice_pdfopen_support_ticketreply_to_ticketrequest_admin_approval

Tech Stack.

FastAPIBackend
LangChain & LangGraphAI agent & workflow
WHMCS APIHosting & billing system
Evolution APIWhatsApp connection
Groq Cloud APILLM inference
DockerApplication containers
PostgreSQL / RedisAgent state & memory

My Role.

I built the AI agent and backend system, connected it to WhatsApp and WHMCS, and worked on the AI tools, memory, security checks, and background processing.

PixelActive
Pixel