chat-adapter-whatsapp

WhatsApp (Meta Cloud API) adapter for chat-py. Python port of upstream packages/adapter-whatsapp.

Install

uv add chat-py chat-py-adapter-whatsapp chat-py-adapter-state-memory

Along with the core SDK and any chat-py-adapter-state-* (use chat-py-adapter-state-redis / -state-pg in production).

Environment variables

Minimal example

import asyncio

from chat import Chat
from chat_adapter_state_memory import MemoryState
from chat_adapter_whatsapp import create_whatsapp_adapter


async def main() -> None:
    chat = Chat(
        user_name="my-bot",
        adapters={"whatsapp": create_whatsapp_adapter()},
        state=MemoryState(),
    )

    @chat.on_new_mention
    async def _on_mention(event):  # pragma: no cover - example only
        await event.thread.post(f"Hi {event.message.author.user_name}!")

    # Plug `chat.handle_webhook("whatsapp", body, headers)` into your HTTP
    # framework of choice (FastAPI, Starlette, aiohttp, ...).


asyncio.run(main())

Webhook handling

WhatsAppAdapter.handle_webhook answers both:

1. GET — Meta verification challenge. The adapter checks hub.mode == "subscribe" and hub.verify_token against the configured verify token, then echoes hub.challenge back with status 200. Pass the request URL via the url= keyword so the adapter can read the query string.
2. POST — event delivery. The raw body is verified with HMAC-SHA256 against the App Secret using :func:hmac.compare_digest. Mismatched signatures produce 401.

Each method returns (status, headers, body) matching the rest of the chat-py adapter family.

Cards & interactive messages

Cards with reply-button actions (1-3 buttons) render as an interactive button payload. Anything else falls back to a WhatsApp-flavoured markdown message (*bold*, _italic_, ~strike~, ```code```). Reply button IDs encode the action via chat:{json} so callbacks round-trip through decode_whatsapp_callback_data.

Limitations / parity notes

• No edit / delete. WhatsApp Cloud API does not support editMessage / deleteMessage; both raise chat.NotImplementedError, matching upstream's Error throws.
• No history fetch. fetch_messages always returns {"messages": []} — the Cloud API has no message history endpoint.
• No typing indicator. start_typing is a no-op.
• Single-DM threads. Every conversation is 1:1, so the channel ID is the thread ID.

Test

uv run pytest packages/chat-adapter-whatsapp

Upstream

https://github.com/vercel/chat/tree/main/packages/adapter-whatsapp