chatflow-agent

Lightweight, async multi-agent framework with native handoffs for real-world channels.
Supports Google Gemini, Google Gemma (Local), xAI Grok, OpenAI (GPT-4o), and Anthropic Claude.
Connect autonomous swarms directly to WhatsApp, Telegram, Webhooks, and CLI.

Framework multiagente asíncrono y liviano con transferencias nativas (handoffs) para canales reales.
Compatible con Google Gemini, Google Gemma (Local), xAI Grok, OpenAI y Anthropic Claude.
Conecta equipos de agentes autónomos a WhatsApp, Telegram, Webhooks y Terminal.

English Documentation

Overview & Architecture

chatflow-agent is designed for developers who want the multi-agent power of OpenAI Swarm without being locked into a single provider, combined with turnkey connectors for messaging platforms like WhatsApp and Telegram.

[ WhatsApp / Telegram / Webhook / CLI ]
                   │
                   ▼
         ┌───────────────────┐
         │      Runner       │ ◄─── Session Memory (per-user history)
         └─────────┬─────────┘
                   │
         ┌─────────┴─────────┐
         ▼                   ▼
  ┌──────────────┐    ┌──────────────┐
  │ Triage Agent │───►│  Spec. Agent │ (Autonomous Peer Handoff)
  │ (Gemini/Grok)│    │ (Local Gemma)│
  └──────────────┘    └──────────────┘
         │                   │
         ▼                   ▼
    [@agent.tool]       [@agent.tool]

API Keys & Configuration

You can provide API keys using any of the following 3 methods:

1. Direct in Python Code (Easiest)

Pass your API key directly when instantiating the Runner or Agent:

# Pass to Runner (used by all agents with that provider)
runner = Runner(starting_agent=my_agent, api_key="AIzaSyYourGeminiKey")

# Or pass directly to a specific Agent:
grok_agent = Agent(
    name="GrokSpecialist",
    provider="grok",
    api_key="xai-your-key-here",
    instructions="..."
)

2. Using a `.env` File

Create a .env file in your project root:

# Google Gemini (Get free key at https://aistudio.google.com/)
GEMINI_API_KEY="AIzaSy..."

# OpenAI (https://platform.openai.com/api-keys)
OPENAI_API_KEY="sk-..."

# xAI Grok (https://console.x.ai/)
XAI_API_KEY="xai-..."

# Anthropic Claude (https://console.anthropic.com/)
ANTHROPIC_API_KEY="sk-ant-..."

Then in your script:

from dotenv import load_dotenv
load_dotenv()

3. Via Terminal Environment Variables

Windows (PowerShell):
```
$env:GEMINI_API_KEY="AIzaSy..."
```
Linux / macOS (Bash/Zsh):
```
export GEMINI_API_KEY="AIzaSy..."
```

Provider Credential Reference Table

Provider	Where to get Key	Environment Variable	In-Code Parameter	Free Tier Available?
Google Gemini	Google AI Studio	`GEMINI_API_KEY`	`api_key="..."`	Yes (Generous free tier)
Google Gemma (Local)	Ollama	None needed	`provider="ollama"`	100% Free & Offline
xAI Grok	xAI Console	`XAI_API_KEY`	`api_key="..."`	Pay-as-you-go
OpenAI	OpenAI Platform	`OPENAI_API_KEY`	`api_key="..."`	Pay-as-you-go
Anthropic Claude	Anthropic Console	`ANTHROPIC_API_KEY`	`api_key="..."`	Pay-as-you-go

Installation

Install only what you need:

# Core framework (Gemini, Gemma, Grok, OpenAI, Claude, CLI)
pip install chatflow-agent

# With WhatsApp Webhook connector (FastAPI + Uvicorn)
pip install "chatflow-agent[whatsapp]"

# With Telegram Bot connector (python-telegram-bot)
pip install "chatflow-agent[telegram]"

# Complete bundle (All channels & dev tools)
pip install "chatflow-agent[all]"

Step-by-Step Quickstarts

A. Cloud LLM Quickstart

Save as bot.py and run with python bot.py:

import asyncio
from chatflow_agent import Agent, Runner

# Define your agent
support_agent = Agent(
    name="SupportBot",
    model="gemini-2.5-flash",  # Or provider="openai", model="gpt-4o-mini"
    instructions="You are a helpful customer support agent for a retail store.",
)

# Register business tools using Python decorators and type hints
@support_agent.tool
def get_order_status(order_id: str) -> dict:
    """Look up shipping and tracking status for an order."""
    return {
        "order_id": order_id,
        "status": "Out for delivery",
        "carrier": "FedEx",
        "eta": "Today before 6:00 PM",
    }

async def main():
    # Pass api_key directly or set GEMINI_API_KEY in environment/.env
    runner = Runner(starting_agent=support_agent)

    response = await runner.run_async(
        session_id="user_session_101",
        user_message="Hi, can you check the status of my order #FDX-8821?",
    )
    print(f"[{response.active_agent_name}]: {response.content}")

if __name__ == "__main__":
    asyncio.run(main())

B. Local Offline Quickstart (Google Gemma)

Run 100% locally on your machine with zero API costs and total privacy using Ollama:

Start Ollama with Gemma: ollama run gemma2:9b
Run this script:

import asyncio
from chatflow_agent import Agent, Runner

# Connects to http://localhost:11434/v1 with no API key required
local_agent = Agent(
    name="LocalAnalyst",
    provider="ollama",
    model="gemma2:9b",
    instructions="You analyze confidential financial reports locally.",
)

@local_agent.tool
def calculate_vat(subtotal: float, rate_percentage: float = 21.0) -> dict:
    """Calculate VAT tax and total amount."""
    vat = subtotal * (rate_percentage / 100.0)
    return {"subtotal": subtotal, "vat": round(vat, 2), "total": round(subtotal + vat, 2)}

async def main():
    runner = Runner(starting_agent=local_agent)
    response = await runner.run_async(
        session_id="local_user",
        user_message="Calculate VAT for a $450 invoice.",
    )
    print(f"[{response.active_agent_name}]: {response.content}")

if __name__ == "__main__":
    asyncio.run(main())

C. Multi-Agent Swarm with Handoffs

Agents can autonomously delegate tasks to specialists:

import asyncio
from chatflow_agent import Agent, Runner

# 1. Specialist: Technical Support
tech_agent = Agent(
    name="TechSupport",
    model="gemini-2.5-flash",
    instructions="You diagnose hardware and software issues.",
)

@tech_agent.tool
def run_diagnostics(device_id: str) -> str:
    """Checks device telemetry."""
    return f"Device {device_id}: All sensors nominal. Firmware v2.1 up to date."

# 2. Specialist: Billing & Invoices
billing_agent = Agent(
    name="Billing",
    model="gemini-2.5-flash",
    instructions="You handle invoices, subscriptions, and refund requests.",
)

# 3. Receptionist (Frontline) with handoffs to specialists
concierge = Agent(
    name="Reception",
    model="gemini-2.5-flash",
    instructions="Greet customers and transfer to TechSupport or Billing as required.",
    handoffs=[tech_agent, billing_agent],  # Swarm handoff capability
)

async def main():
    runner = Runner(starting_agent=concierge)

    # The model detects it is a technical query and hands off to TechSupport automatically
    resp = await runner.run_async(
        session_id="client_77",
        user_message="My device DEV-42 is blinking red. Can you run diagnostics?",
    )
    print(f"Active Agent: {resp.active_agent_name}")  # Output: TechSupport
    print(f"Response: {resp.content}")

if __name__ == "__main__":
    asyncio.run(main())

Channel Connectors

WhatsApp Channel (Meta Cloud API)

Run a production webhook server compatible with Meta WhatsApp Business Cloud API:

from chatflow_agent import Agent, Runner
from chatflow_agent.channels import WhatsAppChannel

agent = Agent(name="WhatsAppConcierge", instructions="Answer customer inquiries.")
runner = Runner(starting_agent=agent)

channel = WhatsAppChannel(
    verify_token="my_custom_webhook_secret",  # Verification token configured in Meta App
    access_token="EAA...",                    # Meta Permanent/System User Token
    phone_number_id="109876543210987",        # WhatsApp Phone Number ID from Meta Dashboard
    fallback_message="We are experiencing a temporary delay. Please try again shortly.",
    unsupported_media_message="Currently I can only process text messages.",
)
channel.attach(runner)

if __name__ == "__main__":
    # Exposes GET /webhook (verification handshake) and POST /webhook (incoming messages)
    channel.run(host="0.0.0.0", port=8000)

Testing locally? Use a tunnel like ngrok (ngrok http 8000) or Cloudflare Tunnels to provide Meta with a public HTTPS URL (https://your-domain.ngrok-free.app/webhook).

Telegram Channel

from chatflow_agent import Agent, Runner
from chatflow_agent.channels import TelegramChannel

agent = Agent(name="TelegramBot", instructions="You assist Telegram users.")
runner = Runner(starting_agent=agent)

channel = TelegramChannel(
    token="123456:ABC-DEF1234ghIkl-zyx57W2v1u123ew11",  # From @BotFather
    fallback_message="Sorry, a temporary issue occurred. Please retry in a few moments.",
)
channel.attach(runner)

if __name__ == "__main__":
    # Runs async polling; automatically handles /start, /reset, and typing indicators
    channel.run()

Interactive Terminal (CLI)

Test swarms in your terminal with colored chat output:

from chatflow_agent import Agent, Runner
from chatflow_agent.channels import CLIChannel

agent = Agent(name="TerminalAssistant", instructions="Answer user questions concisely.")
runner = Runner(starting_agent=agent)

channel = CLIChannel(session_id="dev_test")
channel.attach(runner)
channel.run()

Production Resilience & Concurrency

ChatFlow includes built-in safeguards engineered specifically for real-world messaging traffic:

Per-Session Concurrency Lock (asyncio.Lock): When a user sends multiple messages in rapid succession (e.g., three WhatsApp voice transcriptions or quick texts in 2 seconds), an async lock guarantees FIFO execution. Turns are processed in strict sequential order, preventing race conditions, overlapping tool executions, or corrupted conversation histories.
Meta Anti-500 Error Shield: Meta's webhook infrastructure retries delivery aggressively if your endpoint returns HTTP 500. WhatsAppChannel catches any upstream LLM outages or rate limits, returns HTTP 200 to Meta to prevent retry storms, dispatches the configured fallback_message to the user, and logs clean diagnostic details.
Unsupported Media Handling: Audio voice notes, photos, and PDF files are safely intercepted with unsupported_media_message without throwing unhandled exceptions or disrupting ongoing chat sessions.

Session Memory & Persistence

Every conversation turn is tracked by SessionContext:

Sliding Window Pruning: Prevents LLM context overflow by keeping the last max_messages (default: 50).
Active Agent State: If a handoff occurs (e.g., from Reception to Billing), subsequent messages from that user automatically continue talking to Billing.
Resetting: Call session.clear() or send /reset on Telegram to start fresh.

Guía Completa en Español

Configuración de API Keys (Claves de Acceso)

Puedes configurar tus credenciales de cualquiera de estas 3 formas:

Opción 1: Directo en tu Código Python (La más fácil)

Pasa tu clave directamente al instanciar el Runner o el Agent:

# Pasándola al Runner (la usan todos los agentes de ese proveedor):
runner = Runner(starting_agent=mi_agente, api_key="AIzaSyTuClaveDeGemini")

# O a un agente específico:
agente_grok = Agent(
    name="Grok",
    provider="grok",
    api_key="xai-tu-clave-aqui",
    instructions="..."
)

Opción 2: Usando un Archivo `.env` (Recomendado en Producción)

Crea un archivo .env en la raíz de tu proyecto:

# Google Gemini (Obtén tu clave gratis en https://aistudio.google.com/)
GEMINI_API_KEY="AIzaSy..."

# OpenAI (https://platform.openai.com/api-keys)
OPENAI_API_KEY="sk-..."

# xAI Grok (https://console.x.ai/)
XAI_API_KEY="xai-..."

# Anthropic Claude (https://console.anthropic.com/)
ANTHROPIC_API_KEY="sk-ant-..."

Y en tu código Python:

from dotenv import load_dotenv
load_dotenv()

Opción 3: Variables de Entorno en la Terminal

En Windows PowerShell:
```
$env:GEMINI_API_KEY="AIzaSy..."
```
En Linux o macOS:
```
export GEMINI_API_KEY="AIzaSy..."
```

Tabla Comparativa de Proveedores y Claves

Proveedor	Dónde obtener la clave	Variable de Entorno	Parámetro en Python	¿Capa Gratuita?
Google Gemini	Google AI Studio	`GEMINI_API_KEY`	`api_key="..."`	Sí (Muy generosa)
Google Gemma (Local)	Ollama	No requiere clave	`provider="ollama"`	100% Gratis y Offline
xAI Grok	xAI Console	`XAI_API_KEY`	`api_key="..."`	Pago por uso
OpenAI	OpenAI Platform	`OPENAI_API_KEY`	`api_key="..."`	Pago por uso
Anthropic Claude	Anthropic Console	`ANTHROPIC_API_KEY`	`api_key="..."`	Pago por uso

Ejemplo Rápido: De 0 a Funcionando en 2 Minutos

Crea un archivo mi_asistente.py:

import asyncio
from chatflow_agent import Agent, Runner

# 1. Definir el agente con sus herramientas
soporte = Agent(
    name="SoporteClientes",
    model="gemini-2.5-flash",
    instructions="Eres un asistente cordial de atención al cliente.",
)

@soporte.tool
def consultar_stock(articulo: str) -> dict:
    """Consulta la disponibilidad y precio de un artículo en inventario."""
    catalogo = {
        "laptop": {"stock": 5, "precio": "$1,200"},
        "teclado": {"stock": 18, "precio": "$45"},
        "mouse": {"stock": 30, "precio": "$25"},
    }
    return catalogo.get(articulo.lower(), {"stock": 0, "precio": "No disponible"})

# 2. Ejecutar la conversación
async def main():
    # Puedes pasar tu api_key aquí directamente si no usas variables de terminal:
    # runner = Runner(starting_agent=soporte, api_key="AIzaSy...")
    runner = Runner(starting_agent=soporte)

    respuesta = await runner.run_async(
        session_id="cliente_whatsapp_1",
        user_message="Hola, ¿tienen stock de la laptop y a cuánto está?",
    )
    print(f"[{respuesta.active_agent_name}]: {respuesta.content}")

if __name__ == "__main__":
    asyncio.run(main())

Para ejecutar:

python mi_asistente.py

Ejemplo: Servidor de WhatsApp en Producción

Crea servicio_whatsapp.py:

from chatflow_agent import Agent, Runner
from chatflow_agent.channels import WhatsAppChannel

agente_ventas = Agent(
    name="VentasWhatsApp",
    model="gemini-2.5-flash",
    instructions="Ayudas a los clientes a cotizar y realizar compras.",
)

runner = Runner(starting_agent=agente_ventas)

# Conector oficial para Meta Cloud API con resiliencia de produccion
servidor_whatsapp = WhatsAppChannel(
    verify_token="tu_token_verificacion_meta", # Configurado en Meta Developers
    access_token="EAA...",                     # Token de acceso de Meta
    phone_number_id="102938475610293",         # ID del numero de WhatsApp Business
    fallback_message="Disculpa, estamos experimentando una demora temporal. Por favor intenta en unos momentos.",
    unsupported_media_message="Por el momento solo puedo procesar mensajes de texto.",
)
servidor_whatsapp.attach(runner)

if __name__ == "__main__":
    # Levanta el webhook en http://localhost:8000/webhook
    servidor_whatsapp.run(host="0.0.0.0", port=8000)

Resiliencia y Concurrencia en Produccion

Candado de Concurrencia (asyncio.Lock): Si un cliente envia 3 mensajes seguidos en WhatsApp o Telegram, se encolan y procesan en estricto orden FIFO por usuario. Nunca se mezclan turnos ni se corrompe el historial.
Escudo Anti-500 en WhatsApp: Si la IA tiene una microcaida o se agota la cuota del proveedor, el webhook responde HTTP 200 a Meta (evitando bombardeos de reintentos) y le envia al usuario un mensaje de contingencia amigable (fallback_message).
Filtro de Mensajes Multimedia: Audios, fotos y documentos son interceptados con un aviso claro (unsupported_media_message) sin interrumpir la sesion.

Author & Community

Created and maintained by Gustavo Baranda.

If you find chatflow-agent useful for your projects, consider giving it a ⭐ on GitHub!
Contributions, issues, and feature requests are always welcome.

License

Distributed under the MIT License.

Name		Name	Last commit message	Last commit date
.github/workflows		.github/workflows
examples		examples
src/chatflow_agent		src/chatflow_agent
tests		tests
.gitignore		.gitignore
LICENSE		LICENSE
README.md		README.md
pyproject.toml		pyproject.toml

Navigation Menu

Repository files navigation

chatflow-agent

English Documentation

Table of Contents

Overview & Architecture

API Keys & Configuration

1. Direct in Python Code (Easiest)

2. Using a `.env` File

3. Via Terminal Environment Variables

Provider Credential Reference Table

Installation

Step-by-Step Quickstarts

A. Cloud LLM Quickstart

B. Local Offline Quickstart (Google Gemma)

C. Multi-Agent Swarm with Handoffs

Channel Connectors

WhatsApp Channel (Meta Cloud API)

Telegram Channel

Interactive Terminal (CLI)

Production Resilience & Concurrency

Session Memory & Persistence

Guía Completa en Español

Configuración de API Keys (Claves de Acceso)

Opción 1: Directo en tu Código Python (La más fácil)

Opción 2: Usando un Archivo `.env` (Recomendado en Producción)

Opción 3: Variables de Entorno en la Terminal

Tabla Comparativa de Proveedores y Claves

Ejemplo Rápido: De 0 a Funcionando en 2 Minutos

Ejemplo: Servidor de WhatsApp en Producción

Resiliencia y Concurrencia en Produccion

Author & Community

License

About

Releases

Packages

Contributors

Languages

Latest commit

History

Folders and files

Repository files navigation

chatflow-agent

English Documentation

Table of Contents

Overview & Architecture

API Keys & Configuration

1. Direct in Python Code (Easiest)

2. Using a .env File

3. Via Terminal Environment Variables

Provider Credential Reference Table

Installation

Step-by-Step Quickstarts

A. Cloud LLM Quickstart

B. Local Offline Quickstart (Google Gemma)

C. Multi-Agent Swarm with Handoffs

Channel Connectors

WhatsApp Channel (Meta Cloud API)

Telegram Channel

Interactive Terminal (CLI)

Production Resilience & Concurrency

Session Memory & Persistence

Guía Completa en Español

Configuración de API Keys (Claves de Acceso)

Opción 1: Directo en tu Código Python (La más fácil)

Opción 2: Usando un Archivo .env (Recomendado en Producción)

Opción 3: Variables de Entorno en la Terminal

Tabla Comparativa de Proveedores y Claves

Ejemplo Rápido: De 0 a Funcionando en 2 Minutos

Ejemplo: Servidor de WhatsApp en Producción

Resiliencia y Concurrencia en Produccion

Author & Community

License

About

Topics

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages

2. Using a `.env` File

Opción 2: Usando un Archivo `.env` (Recomendado en Producción)