How to Build a Smart Hebrew WhatsApp Bot · The Full Visual Guide.

Think about your business one week from now. A lead sends a WhatsApp message at 2 AM, receives a response within seconds, and schedules a Zoom meeting in your calendar without you ever being awake. This is not a dream, it is a 30-minute process. All using 5 free-to-generous tools that are the new standard for building agents: Kapso (Meta-approved WhatsApp number), Maton (calendar and Zoom connection), Google AI Studio (Gemini brain), Claude Code (the developer that assembles everything), and Vercel (cloud deployment). In session 3 of the Claude Code workshop, we created such a live bot from scratch. This guide is a freeze of everything we did, with 34 organized screenshots. Follow step-by-step, and at the end you will have a professional WhatsApp bot working for you 24/7.

What you will have by the end of the guide

• A Meta-approved business WhatsApp number · No SIM card to buy, no need to connect your personal number
• A bot that answers customers in Hebrew in your voice, 24/7
• Automatic integration with Google Calendar · Every meeting the bot books goes straight into your calendar
• Integration with Zoom · A meeting link is created and sent to the customer automatically
• Vercel cloud deployment · The bot works even when your computer is turned off
• Full control: Update behavior, add capabilities, change models, at any moment

What you need before starting

If you don't have Claude Code yet, the workshop page has full installation instructions + permissions + personal CLAUDE.md: ai-masterclub.com/claude-workshop

Step 1 · Registering for Kapso and creating a WhatsApp number

✅ You finished step 1. You have an approved business WhatsApp number. So far you have invested 5-10 minutes.

Step 2 · Maton · Connecting the bot to the calendar and Zoom

Step 3 · Gathering all API keys

3a · Kapso API key

3b · Google AI Studio (Gemini) API key

💳 Using Gemini costs pennies. But as the bot handles more interactions, the billing will increase. It is recommended to set a monthly spending limit in your Google Cloud account · This way you won't be surprised.

✅ Now we have 3 keys saved on the side · Kapso, Maton (already saved in step 18), and Gemini. In addition · the Kapso number we created in step 1, and our personal number (for ownership verification). All of this goes into Claude in the next step.

Step 4 · Entering Claude Code · Creating a new folder

AI Master Club opens the door for you: A live session every week, recordings, and a community. 47 ILS per month, cancel anytime.

Step 5 · The skill from the workshop site · The magic

Paste the skill into Claude · Along with the 3 keys

The full prompt to paste into Claude · After the skill is installed

Build me a WhatsApp agent. I already created a number in Kapso, connected tools in Maton, and created an API key in Google AI Studio. Here are the keys: • KAPSO_API_KEY: [Paste the Kapso key from step 22] • MATON_API_KEY: [Paste the Maton key from step 18] • GEMINI_API_KEY: [Paste the Gemini key from step 29] • Kapso number: [The US number you created in step 1] • My personal number: [Your number, for ownership verification only] Use the whatsapp-agent skill I already installed. Work in Plan Mode first, show me the plan, and after I approve, proceed to build and deploy to Vercel.

💡 Claude automatically detects the installed skill and loads the instructions. It will take the 3 keys, create a Next.js project, connect the webhooks, deploy to Vercel, and connect the webhook URL back to Kapso · All by itself. You just approve the plan when it presents it.

🎉 You are done. You have a smart, live, and 24/7 running WhatsApp bot. Send a message to the Kapso number you created · The agent will answer in Hebrew, schedule Zoom meetings, and create Google Calendar events. You can give the number to your customers.

What to do after the bot is live

5 things to do in the first week

• **Send a first message to yourself** · And check that it responds the way you want. If the tone is incorrect, tell Claude 'refine the system prompt in file X. I want a more/less _____ tone'
• **Define who not to answer** · In the system prompt, tell the bot 'if number X or Y sends a message, do not answer'. This way you won't get automated messages to your wife or friends
• **Limit working hours** · You can tell the bot 'between 22:00 and 08:00 respond only with an automated message: I will reply in the morning' if you want it to be clear that it is a bot
• **Add capabilities** · If there is a recurring task the bot doesn't do (sending price quotes, 72-hour follow-up), tell Claude 'add this capability to the bot', and it updates the code and redeploys
• **Monitor conversations** · In Maton you can see a log of all conversations the bot managed. It is worth reviewing them once a week to see where it makes mistakes

💡 Token savings: An agent working 24/7 can burn credits. To save · 1) Use Gemini Flash instead of Pro for most requests (5 times cheaper), 2) Put memory in Supabase so the bot doesn't repeat questions, 3) Set a maximum of 1500 tokens per response.

Common errors + solutions

Quick fix for the 3 most common errors

• **Webhook is not receiving messages** · Make sure in Kapso that the Webhook URL is correct (this is the Vercel URL + /api/webhook). Tell Claude 'check the webhook in Kapso and fix if needed'
• **Gemini answers in English without being asked** · In the system prompt within the project file, add 'always answer in Hebrew, even if the user writes in English or mixes languages'
• **Google Calendar is not receiving events** · Check in Maton that calendar permissions remain active (sometimes Google revokes permissions after 7 days if there is an issue). Open Maton, reconnect, and re-approve all permissions

🚀 **Want guidance along the way?** In the AI Master Club there is an active WhatsApp group of people building bots just like yours. Any problem you encounter, someone has already solved it. Join, share the agent you built, and get ideas for additional agents you can build for your business.

What is the next step?

Frequently asked questions

Kapso · Free US number, and also a basic plan for 1000 messages a month for free. Maton · Up to 100 actions a month for free (Zoom, calendar, email). Gemini · After connecting a credit card, payment is according to actual usage. In practice, thousands of messages a month cost a few shekels. Vercel · Completely free for personal projects. Overall, if the bot receives up to 100 messages a day, you will pay less than 50 shekels a month.

Yes. In Kapso, instead of choosing Digital Numbers, choose Bring your own number. You need a physical SIM (can be bought from Tokman for 30 ILS), and you perform the same approval process with Meta. The bot will work in exactly the same way, just with an Israeli number. The advantage · more professional for Israeli businesses. The disadvantage · an extra step of buying the SIM and activating it.

Yes. Everything I did in this guide you can do. The secret is that Claude is the programmer. You just instruct it · 'Build me a WhatsApp bot', 'Add capability X', 'Fix the bug I saw'. If something doesn't work, you tell Claude 'Here is the bug, fix it'. There is no need to read the code or understand it. Treat Claude like a senior developer working for you.

The exact same process. The only difference is in the prompt you give Claude in step 4. Instead of asking for 'a bot that coordinates meetings', you ask for 'a bot that answers FAQs about my products', or 'a bot that creates price quotes according to template X'. Everything else is identical · Kapso, Maton, Gemini, Vercel. The agents differ in Claude, not in the infrastructure.

Yes, easily. You tell Claude 'Switch the model from Gemini to Claude', and it updates the code, asks for the new API key (from Anthropic or OpenAI), and redeploys. More than that · you can even run the agent with 2 models simultaneously and choose which one works better for each type of question. That is the beauty of working with Claude Code · The structure is completely flexible.

💪 Ready to build more agents?

📎 Concise version of the guide · Without images

Think about it: Every lead that reaches you on WhatsApp at 2 AM gets a response within seconds. Not just a generic response, but actually coordinating a Zoom meeting, creating an event in your Google Calendar, and sending them a link to join. This is not the future. This is today. In this guide, I will show you how to build yourself a smart WhatsApp agent in Hebrew, without writing a single line of code. The skill I created does all the heavy lifting: creates a Next.js project, connects all APIs, deploys to Vercel, and sets up webhooks. You just provide 3 API keys and press Enter.

What the agent does after it is live

• Receives WhatsApp messages through Kapso (WhatsApp Business API)
• Replies in Hebrew in your voice, using a system prompt you control
• Schedules Zoom meetings and automatically creates a Google Calendar event
• Sends the client a ready Zoom link plus a calendar invitation
• Remembers the full conversation history so it never repeats itself
• Runs 24/7 with zero maintenance, on a virtually free Vercel infrastructure

The installation: A single command in the terminal

The skill installation command (to copy into the terminal)

mkdir -p ~/.claude/skills/whatsapp-agent && curl -sL https://www.ai-masterclub.com/skills/whatsapp-agent.md -o ~/.claude/skills/whatsapp-agent/SKILL.md && echo '✅ Skill installed. Open Claude Code and type: Build me a WhatsApp agent'

This skill is also published on GitHub with all files and templates: github.com/amichaiai/whatsapp-agent. The version on the site is the short version (SKILL.md only). For a full installation with Templates and References, it is recommended to run a git clone of the repo.

How to use the skill after installation

What the skill does for you (8 automated steps)

• Checks that you have node, npm and Vercel CLI installed
• Creates a project directory with Next.js 15 and TypeScript ready to use
• Retrieves the necessary IDs from Kapso (phone_number_id and more)
• Connects Maton to Zoom and Google Calendar via OAuth
• Pushes all ENV variables to Vercel via REST API (faster than CLI)
• Performs the first production deployment
• Creates a webhook in Kapso pointing to your new URL
• Verifies end-to-end with a health check and agent test

You must turn off Vercel Deployment Protection in the project, otherwise Kapso will receive a 401 error and fail to call the webhook. The skill explains where to turn this off in 30 seconds. It is a single setting under Settings → Deployment Protection → Disabled.

Who needs this agent

Additional questions

Maton supports all of Google Workspace, Slack, HubSpot, Salesforce, Notion, Linear and more. The skill adds basic tools automatically, and adding more is simple. Tell Claude Code 'add the ability for the agent to send emails' and it will update the code.

Gemini supports 50+ languages. The default in the skill is Hebrew, but you can update the system prompt in lib/prompts.ts so it replies according to the customer language. This is a 3-line update.

In the AI Master club, a live weekly meeting takes place to review such cases together. In addition, there is an active community of business owners who have already built agents and share what works and what does not. You can ask questions in the quiet WhatsApp group.

This guide is not theoretical. I built this agent for myself, it runs and answers clients, and this is what frees me up to work on more important things. When you build it for yourself, you free yourself.