Claude Code interviews you, then builds your WhatsApp bot
It's a system of prompts that turns a 20-minute conversation into a deployed customer-service agent—webhooks, memory, deduplication and all.

What it does
WhatsApp AgentKit is not a template you copy. It is a set of instructions that Claude Code reads to interview you about your business, then write, test, and deploy a complete WhatsApp agent. You answer ten questions; Claude handles the server, webhook verification, conversation memory, and provider glue for Zernio or Meta Cloud API. The project is written in Spanish and aimed at LATAM builders who do not want to wrestle with Facebook app reviews or HMAC signatures.
The interesting bit
The README is admirably honest about costs and architecture. It explains that most of the work in a WhatsApp agent is not the LLM but the plumbing—webhooks, retries, deduplication—and it solves that upfront. The agent immediately acknowledges webhooks to avoid provider retries, then processes asynchronously; it deduplicates by event ID and verifies HMAC-SHA256 signatures before touching a message. That kind of defensive design is usually learned the hard way.
Key highlights
- Claude Code conducts a ten-question interview about your business, then generates the entire codebase including prompts, memory layer, and provider adapters.
- Supports both Zernio (faster setup, no Facebook app needed) and direct Meta Cloud API (more control, more configuration).
- Per-customer conversation history persists across sessions, and the agent refuses to hallucinate business details it does not have.
- Includes a local terminal simulator for testing before you expose a public webhook.
- Open source under MIT, with explicit cost estimates for Claude token usage at different message volumes.
Caveats
- The project is built for Spanish-speaking users; all documentation and the interview flow are in Spanish.
- Signature verification is optional at startup to ease local testing, which means a misconfigured production deploy could accept injected webhooks.
- The agent converses and collects data, but the README explicitly notes it does not yet execute the resulting actions itself.
Verdict
Small business owners and developers in LATAM who need a customer-facing WhatsApp presence without learning Facebook’s developer dashboard should look here. If you already have a robust FastAPI backend and a verified Meta Business app, this is mostly well-documented glue code you could replicate yourself.
Frequently asked
- What is Hainrixz/whatsapp-agentkit?
- It's a system of prompts that turns a 20-minute conversation into a deployed customer-service agent—webhooks, memory, deduplication and all.
- Is whatsapp-agentkit open source?
- Yes — Hainrixz/whatsapp-agentkit is open source, released under the MIT license.
- What language is whatsapp-agentkit written in?
- Hainrixz/whatsapp-agentkit is primarily written in Python.
- How popular is whatsapp-agentkit?
- Hainrixz/whatsapp-agentkit has 502 stars on GitHub.
- Where can I find whatsapp-agentkit?
- Hainrixz/whatsapp-agentkit is on GitHub at https://github.com/Hainrixz/whatsapp-agentkit.