Overview

wakit supports connecting AI agents that automatically respond to incoming WhatsApp messages. Create lightweight built-in agents or connect external agents via standard protocols.

Architecture

1. An incoming message arrives via the WhatsApp webhook
2. A database trigger fires the agent-client Edge Function
3. The agent-client builds conversation context and calls the agent API
4. The agent responds and the response is saved as a new message
5. An outgoing trigger sends the response via the WhatsApp Cloud API

Agent types

Built-in agents

Configure directly in the dashboard with:

• Provider (OpenAI, Anthropic, Google, Groq)
• Model
• Instructions (system prompt)
• Tools (MCP, SQL, HTTP, Calculator)

External agents

Deploy your own agent as a service and connect via:

• Chat Completions — OpenAI-compatible API
• A2A — Google's Agent-to-Agent protocol

Configuration

type AgentExtra = {
  mode?: "active" | "draft" | "inactive";
  api_url?: "openai" | "anthropic" | "google" | "groq" | string;
  api_key?: string;        // default: provider env var
  model?: string;          // default: gpt-5-mini
  protocol?: "chat_completions" | "a2a";
  instructions?: string;   // system prompt
  tools?: ToolConfig[];
  max_messages?: number;
  temperature?: number;
  max_tokens?: number;
};

Set mode to "active" to enable the agent. Messages will only be routed to the agent when it's active.