An AI agent for WhatsApp that acts in your systems

A typical WhatsApp AI chatbot stays inside the chat and only answers questions. When a customer asks to change a delivery address or check an order, the bot explains who to write to.

The Lelemon agent carries out the action. It looks up the order in your system, updates the address, creates the exchange label and tells the customer what was done. The conversation ends with the problem solved.

It is built for startups that already serve customers on WhatsApp: e-commerce, marketplaces, SaaS and fintech. The customer gets it all done without leaving the chat.

You connect your WhatsApp Business number with a setup link and the agent answers from that same number. Your customers keep writing where they always have.

To act in your systems, the agent uses your API, your MCP server or signed webhooks. You do not migrate data or swap tools. You register each action, such as looking up an order or opening a ticket, and grant it to the agent one at a time. Whatever you did not grant, the agent cannot do.

• Your APIEach action points to one of your endpoints, and you define the parameters it takes.
• Your MCP serverIf you already expose tools over MCP, you connect them to an agent from the panel.
• Signed webhooksYour systems receive signed notices, for example when a case moves to a person.

Refunds, deleting data and changing a plan are not actions to leave unattended. You mark them as sensitive and the agent stops before running them. The request waits for approval from the person you choose, and only then continues.

When a case needs human judgment, the agent hands it off. Your team gets the conversation with a summary, so nobody has to ask the customer to repeat what they already said.

In the Lelemon panel each conversation is a case, with teams, routing and an inbox. Whoever takes a case pauses the agent and gives it back when done. If your team works in Claude, they can handle cases over MCP too.

Every action is recorded in the case history: what the agent did, in which system, with what data and who approved it. Models run on AWS Bedrock: Claude Haiku for quick replies and Claude Sonnet for work that needs reasoning, and you pick the model for each agent.

Plans start at US$700 a month with 1,000 conversations. A conversation is one thread of up to 10 exchanges within 24 hours, and it costs the same with any model. For more volume there is a 5,000-conversation plan at US$2,500 a month, and with an annual contract, from US$0.35 per conversation.

If you go over your plan, each extra conversation costs US$0.80 on the 1,000 plan and US$0.60 on the 5,000 plan. Meta bills WhatsApp separately, directly to you.