ConvoZen WhatsApp Agent CLI: Build and Manage from the Terminal

A WhatsApp AI agent that can chat, send media, collect payments, make voice calls and reach customers through approved templates has many moving parts. The ConvoZen WhatsApp agent CLI lets you build and manage all of them from the terminal, inside Claude Code, Codex, Cursor, Antigravity or any tool that follows the Agent Skills spec. Your team states the outcome and the business context. The ConvoZen skills then inspect the setup, ask for any missing fields and call the CLI.

Why WhatsApp Agents Are Hard to Run in Production

The demo is easy. Production is the hard part. Teams that build a WhatsApp AI agent rarely struggle with the idea. They struggle with setup, because every agent begins with a prompt tab, an actions tab, a knowledge base tab and a settings tab, and the WhatsApp channel, templates, testing and routing sit on top of those. The work is divided among different people and screens, and four patterns keep appearing:

• Multi-tab hopping: the prompt is in one tab, the actions in another and the knowledge base in a third, so context is lost each time you switch.
• Manual action wiring: teams paste curl commands into forms, guess the shape of the parameters and test blind in an interface designed for clicking.
• Knowledge base babysitting: documents are uploaded, re-indexed and checked for retrieval, all outside the flow where the agent is written.
• Endless config screens: every voice, handler, channel and variable has its own settings page, and none can be reached from the editor.

ConvoZen Skills and the CLI bring all of this into a single terminal conversation.

One Command-Line System for the Entire WhatsApp Agent

The CLI places a convozen binary on your path. The skills copy ConvoZen instructions into the coding agents it detects, such as Codex, Claude Code and Cursor, and call that binary whenever they need to talk to ConvoZen. In the same conversation, you can list skills, inspect agents, add tools, attach knowledge bases or place a test call. Chat, payments, documents, images, voice and templates stop being disconnected tools and become one WhatsApp system, configured through one CLI.

What a ConvoZen WhatsApp Agent Can Do

ConvoZen agents use the full set of WhatsApp business features, so customers are not limited to plain text.

• Text chat and voice notes: conversations are contextual and multilingual, with memory across sessions. The agent receives, transcribes, understands and replies to voice notes.
  
• Images and catalogues: a customer writes “I’m looking for a sofa set,” and the agent replies in the same thread with product photos, names and prices. In property, the agent shares listing photos or a walkthrough video. It can also send ID verification instructions and onboarding clips.
  
• Location: when a customer shares their live location, the agent understands where they are and responds accordingly, so the reply fits the place and the moment.
  
• In-chat payments: UPI (Unified Payments Interface) is India’s instant payment system, in which customers pay by scanning a QR code with any UPI app. The agent presents the QR code and a “Pay with UPI” option inside WhatsApp, and the completed payment appears in the same thread. The customer no longer has to leave the chat to pay.
  
• Documents: the agent sends brochures, invoices and policies, and it parses the documents customers upload.
  
• Templates and carousels: templates are pre-approved message formats, reviewed and approved for delivery, which lets you contact customers even outside the 24-hour window. They support up to 10 variables for per-customer personalisation, interactive buttons (quick replies, call-to-action links, one-tap phone dials and copy-code), and swipeable carousels of up to 10 cards in one message. A template can also open with a pinned location, which turns a plain notification into a branded message with the address attached.
  
• Voice calls: a message can carry a “Call Me Now” button. One tap takes the customer from text to a real-time AI voice call. Agents handle inbound and outbound WhatsApp calls with duplex audio and human-like speech. Customers grant call permission within the chat and can change it later from the business profile.

How Teams Build and Deploy with ConvoZen CLI

Build where you already work. Setup takes four steps and runs once:

1. Install the CLI on macOS, Linux, Windows Command Prompt or PowerShell. Windows installers automatically select x64 or ARM64.
2. Install the skills with a single npx command, then restart your coding agent so it discovers them.
3. Log in with convozen auth login –email, or open the CLI and type /login. Run convozen init to select the active organisation and project.
4. Start building. Open convozen and type /help to see the available commands, or ask your coding agent for the outcome you want.

npx @convozen/convozen-skills install

convozen > create whatsapp-agent

One documented example prompt reads: “Create a ConvoZen WhatsApp support agent for the project. It should answer order-status questions, collect name and phone number, and hand off when the user asks for a human.” In the terminal demo, the skills load, draft the prompt and wire the tools. A test conversation then passes, and the agent goes live.

Use Cases Across Industries

ConvoZen WhatsApp agents use the same building blocks across industries: media, documents, payments, templates and voice. What changes is how each team combines them. The ConvoZen deployments below show the same patterns in production.

• PropTech: a renter or buyer receives property photos and a walkthrough video in the same thread where they ask their questions, and a “Call Me Now” button moves them to a live AI voice call when a text reply isn’t enough. NoBroker Builders, which handles roughly 1 million calls a month, used ConvoZen voice agents to extend outreach across the long tail of buyers. It saw an 8% uplift in total property visits per month and revived 12.5% of rejected customers.
• Retail and D2C: a shopper who asks for a product category gets catalogue cards with photos, names and prices, or a swipeable carousel of up to 10 products. The same agent answers order-status questions, collects name and phone number, and hands off to a human on request. Pilgrim, the beauty brand, runs a ConvoZen WhatsApp AI agent for support. It recorded a 34% decrease in agent transfer rate, a 73% increase in bot resolution rate and a CSAT of 4.25. Nilesh Kambli, Sr. VP, Customer Experience at Pilgrim, said: “We’ve seen agent transfers drop drastically.” He added: “For the first time going into a sale, we haven’t increased headcount at all, and that’s a huge win for us.”
• BFSI and insurance: the agent sends invoices and policy documents in the chat, parses customer uploads, shares ID verification instructions, and takes UPI payments by QR code without leaving WhatsApp. On the voice side, Jana Small Finance Bank used ConvoZen for outreach such as pre-approved loans and EMI reminders, with a 10% boost in resolution rate and 7% sales growth through conversion gains. Lendingkart moved to 100% automated conversation monitoring across 100+ compliance checkpoints and reported 20% higher conversions.
• Edtech and onboarding: short explainer videos and onboarding clips arrive alongside the conversation, and pre-approved templates reach learners outside the 24-hour window with a reminder or next step. Zell Education used ConvoZen to review every counselling conversation, achieving a 7%+ lead-to-conversion uplift and a 60%+ reduction in manual QA effort.

Production Readiness and Control

Testing happens before customers see anything. Before an agent goes live, the terminal flow runs a test conversation, and teams can also place a test call from the CLI. For templates, Live Preview and QA show exactly how a message renders with sample values, so typos are caught in the studio and not in the inbox. Templates are reviewed and approved for delivery. Routing includes human handoff, so a customer who asks for a person is connected to one.

Ready to Build Your First WhatsApp Agent? Watch the WhatsApp agent CLI work on your own use case. Book a demo