musadiqurrahman/whaagent

多租户WhatsApp人工智能平台

---

概述

WhaAgent 连接 AI 代理到 WhatsApp。Go 网关负责 WhatsApp 协议。Python API 运行基于 LangGraph 的代理。多个 WhatsApp 号码、多个 AI 代理、单一部署。

---

建筑

---

信息流

---

安装

选项A：Docker

Docker 将 Go 网关和 Python API 构建在一个容器中。暴露端口 8000。

选项B：手动设置

先决条件：

• Python 3.12+
• Go 1.21+
• UV包管理器

---

配置

1.环境变量

将.env.example复制到.env，并设置你的LLM API密钥：

2.应用配置

运行交互式设置向导：

这会生成 ~/.whaagent.yaml：

3.链接 WhatsApp

启动服务后，在浏览器中打开二维码页面：

打开http://localhost:8000/api/v1/whatsapp/qr/personal/page，用WhatsApp扫描（设置>关联设备）。

---

特工们

WhaAgent 内置了 12 个代理。每个代理都有系统提示、工具访问，并且可以自定义。

内置代理

---

Overview

WhaAgent connects AI agents to WhatsApp. A Go gateway handles the WhatsApp protocol. A Python API runs LangGraph-powered agents. Multiple WhatsApp numbers, multiple AI agents, single deployment.

User sends WhatsApp message
        |
        v
Go Gateway (whatsmeow) --> Python API (FastAPI) --> AI Agent (LangGraph)
        |                                                  |
        v                                                  v
Response sent back via WhatsApp                    Tools / LLM / Memory

---

Architecture

flowchart TB
    subgraph Clients["Client Layer"]
        WA1["WhatsApp User A"]
        WA2["WhatsApp User B"]
        WA3["WhatsApp User C"]
    end

    subgraph Gateway["Go Gateway"]
        SM["Session ManagerMulti-tenant WhatsApp connections"]
        QR["QR Code GeneratorSSE streaming to browser"]
        EVT["Event HandlerAsync message processing"]
        HTTP["HTTP ClientForward to Python API"]
    end

    subgraph API["Python API — FastAPI"]
        ROUTE["Webhook Router/api/v1/channels/whatsapp/message"]
        AUTH["API Key AuthX-API-Key header"]
        POOL["Tenant Agent PoolLRU eviction, per-tenant isolation"]
        DB["SQLite CheckpointConversation memory persistence"]
    end

    subgraph Graph["LangGraph StateGraph"]
        SUM["Summarize NodeToken threshold, context compression"]
        ROUTER["Router NodeLLM intent classification"]
        AGENT["Agent NodeTool orchestration, ReAct loop"]
    end

    subgraph Agents["Agent Registry"]
        A1["assistantGeneral-purpose orchestrator"]
        A2["developerCode exploration and analysis"]
        A3["news / br-news / es-newsNews aggregation"]
        A4["learning / recipe / youtubeDomain-specific agents"]
        A5["github-pr-reviewerPull request automation"]
        A6["Custom [email protected]"]
    end

    subgraph Tools["Tool Layers"]