Build autonomous multi-agent swarms for WhatsApp, Telegram, and Webhooks in seconds without heavy dependency chains.
Construye equipos de agentes autonomos para WhatsApp, Telegram y Webhooks en segundos sin lidiar con dependencias monstruosas.
pip install chatflow-agentEnglish Documentation | Guía en Español | PyPI Package | Report Issue
-
Ultra-Lightweight Core: Only 3 runtime dependencies (
google-genai,pydantic,httpx). A tiny ~72 KB package footprint that installs in seconds on Docker, serverless, and cloud VPS environments. -
Omnichannel Native: Production-ready connectors for WhatsApp (Official Meta Cloud API), Telegram, Webhooks (FastAPI), and interactive CLI.
-
Multi-Provider Freedom: Switch between Google Gemini, Google Gemma (100% free local via Ollama), xAI Grok, OpenAI (GPT-4o), and Anthropic Claude with a single parameter. Zero vendor lock-in.
-
Production-Hardened Shields: Native per-session concurrency locking (
asyncio.Lock) prevents out-of-order race conditions from fast typers. Webhook anti-500 shield prevents Meta retry bombardment storms during LLM downtime. -
Autonomous Peer Handoffs: Agents delegate conversations dynamically based on customer intent, without rigid state machines or complex graph builders.
import asyncio
from chatflow_agent import Agent, Runner
# 1. Define specialist agents
tech_agent = Agent(name="TechSupport", model="gemini-2.5-flash", instructions="Help with bugs.")
billing_agent = Agent(name="Billing", model="gemini-2.5-flash", instructions="Help with invoices.")
# 2. Receptionist agent with autonomous peer handoffs
receptionist = Agent(
name="Receptionist",
model="gemini-2.5-flash",
instructions="Greet the customer and route them to TechSupport or Billing.",
handoffs=[tech_agent, billing_agent],
)
async def main():
runner = Runner(starting_agent=receptionist)
result = await runner.run_async(
session_id="user_whatsapp_1",
user_message="Hi! I need help with an invoice error on my account.",
)
print(f"[{result.active_agent_name}]: {result.content}")
# Output: [Billing]: I would be glad to help check your invoice. Could you share your account ID?
if __name__ == "__main__":
asyncio.run(main())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]
You can provide API keys using any of the following 3 methods:
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="..."
)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()-
Windows (PowerShell):
$env:GEMINI_API_KEY="AIzaSy..."
-
Linux / macOS (Bash/Zsh):
export GEMINI_API_KEY="AIzaSy..."
| 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 |
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]"
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())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())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())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
app_secret="a1b2c3d4...", # Meta App Secret (App Dashboard -> App settings -> Basic)
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).
Note
API Documentation in Development: For security, FastAPI interactive documentation (/docs, /redoc, and /openapi.json) is disabled by default in production. To enable Swagger UI and OpenAPI schemas during local development or debugging, pass enable_docs=True when instantiating WhatsAppChannel:
channel = WhatsAppChannel(
...,
enable_docs=True, # Enables /docs, /redoc, and /openapi.json for local development
)Tip
Production Rate Limiting & Reverse Proxy: To protect /webhook and /health endpoints against flood attacks, denial of service (DoS), and CPU exhaustion from unauthenticated requests without risking dropping legitimate Meta webhook bursts at the application level, always place Uvicorn behind a reverse proxy in production (such as Nginx with limit_req_zone or Cloudflare Rate Limiting):
limit_req_zone $binary_remote_addr zone=whatsapp_limit:10m rate=10r/s;
location /webhook {
limit_req zone=whatsapp_limit burst=20 nodelay;
proxy_pass http://127.0.0.1:8000;
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
}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()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()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.
WhatsAppChannelcatches any upstream LLM outages or rate limits, returns HTTP 200 to Meta to prevent retry storms, dispatches the configuredfallback_messageto the user, and logs clean diagnostic details. -
Unsupported Media Handling: Audio voice notes, photos, and PDF files are safely intercepted with
unsupported_media_messagewithout throwing unhandled exceptions or disrupting ongoing chat sessions. -
Turn Failure Rollback: If an LLM request times out, rate limits, or crashes mid-turn, ChatFlow automatically rolls back dialogue history to the last consistent turn in memory and SQLite (note: rollback reverts conversational history; external side effects of already executed tools are not rolled back), ensuring clean context without orphaned partial turns.
ChatFlow provides native, zero-dependency persistence to ensure conversations and active specialist agents survive server restarts, worker redeployments, and crashes.
Ideal for testing, local scripting, and ephemeral sessions:
from chatflow_agent import Runner, SessionStore
runner = Runner(starting_agent=triage_agent, session_store=SessionStore(default_max_turns=20))Built on Python's standard sqlite3 library with zero additional external packages. Safely manages Windows file locks and atomic transactions:
from chatflow_agent import Runner, SQLiteSessionStore
# History and active specialist agent state persist across server reboots
session_store = SQLiteSessionStore(db_path="sessions.db", default_max_turns=30)
runner = Runner(starting_agent=triage_agent, session_store=session_store)Key Persistence Features:
-
Autonomous Handoff Survival: If a customer is handed off to
Billing Specialist, the active agent pointer is recorded in SQLite. When the server reboots or workers recycle, subsequent customer messages continue directly withBilling Specialistinstead of resetting to the triage agent. -
Sliding Window FIFO Pruning: Maintains the latest conversation turns while safely discarding older messages in SQLite to avoid LLM context limits.
-
Omnichannel Compatibility: Works seamlessly across WhatsApp, Telegram, Webhook APIs, and CLI.
Ready-to-run reference implementations are available in the examples/ directory:
-
examples/whatsapp_persistent_bot.py: Meta Cloud API WhatsApp bot with SQLite persistence, concurrency locks, and 3-agent handoff hierarchy.
-
examples/telegram_persistent_bot.py: Telegram bot with state persistence across polling restarts.
-
examples/webhook_api_service.py: Async REST API service with CORS,
/api/chatand/api/sessions/{session_id}endpoints for React, Next.js, and mobile frontends. -
examples/local_gemma_cli.py: 100% offline, free CLI assistant powered by Google Gemma 2 via Ollama with persistent terminal sessions.
-
examples/multi_provider_swarm.py: Multi-provider swarm coordinating Gemini, Gemma, Grok, and Claude in a single conversational flow.
-
Nucleo Ultra-Ligero: Solo 3 dependencias de ejecucion (
google-genai,pydantic,httpx). Un paquete diminuto de ~72 KB que se instala en segundos en contenedores Docker, entornos serverless y servidores VPS. -
Nativo Omnicanal: Conectores listos para produccion para WhatsApp (Meta Cloud API Oficial), Telegram, Webhooks (FastAPI) y CLI interactiva.
-
Libertad Multi-Proveedor: Alterna entre Google Gemini, Google Gemma (100% gratis y local via Ollama), xAI Grok, OpenAI (GPT-4o) y Anthropic Claude con un unico parametro. Cero ataduras a proveedores.
-
Escudos Reforzados para Produccion: Bloqueo de concurrencia nativo por sesion (
asyncio.Lock) que previene condiciones de carrera por usuarios que escriben rapido. Escudo anti-500 en webhooks para evitar tormentas de reintentos de Meta durante caidas del LLM. -
Handoffs Autonomos entre Agentes: Los agentes delegan conversaciones dinamicamente segun la intencion del usuario, sin maquinas de estado rigidas ni grafos complejos.
import asyncio
from chatflow_agent import Agent, Runner
# 1. Definir agentes especialistas
tech_agent = Agent(name="TechSupport", model="gemini-2.5-flash", instructions="Ayuda con errores tecnicos.")
billing_agent = Agent(name="Billing", model="gemini-2.5-flash", instructions="Ayuda con facturacion y pagos.")
# 2. Agente recepcionista con handoffs autonomos
receptionist = Agent(
name="Receptionist",
model="gemini-2.5-flash",
instructions="Saluda cordialmente al cliente y derivalo a TechSupport o Billing segun su consulta.",
handoffs=[tech_agent, billing_agent],
)
async def main():
runner = Runner(starting_agent=receptionist)
result = await runner.run_async(
session_id="usuario_whatsapp_1",
user_message="Hola! Necesito ayuda con un error en la factura de mi cuenta.",
)
print(f"[{result.active_agent_name}]: {result.content}")
# Salida: [Billing]: Con gusto te ayudo a revisar tu factura. ¿Me podrias indicar tu numero de cuenta?
if __name__ == "__main__":
asyncio.run(main())chatflow-agent fue disenado para desarrolladores que buscan la potencia multi-agente de OpenAI Swarm sin quedar atados a un unico proveedor, combinado con conectores oficiales listos para produccion en plataformas de mensajeria como WhatsApp y Telegram.
[ WhatsApp / Telegram / Webhook / CLI ]
│
▼
┌───────────────────┐
│ Runner │ ◄─── Memoria de Sesion (Historial por usuario)
└─────────┬─────────┘
│
┌─────────┴─────────┐
▼ ▼
┌──────────────┐ ┌──────────────┐
│ Agente Triage│───►│Especialista │ (Handoff Autonomo entre Pares)
│ (Gemini/Grok)│ │(Gemma Local) │
└──────────────┘ └──────────────┘
│ │
▼ ▼
[@agent.tool] [@agent.tool]
Puedes configurar tus credenciales utilizando cualquiera de estos 3 metodos:
Pasa tu clave 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 especifico:
agente_grok = Agent(
name="GrokSpecialist",
provider="grok",
api_key="xai-tu-clave-aqui",
instructions="Eres un analista de datos.",
)Crea un archivo .env en la raiz de tu proyecto:
# Google Gemini (Obtén tu clave gratuita 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-..."
-
Windows (PowerShell):
$env:GEMINI_API_KEY="AIzaSy..."
-
Linux / macOS (Bash/Zsh):
export GEMINI_API_KEY="AIzaSy..."
| Proveedor | Donde obtener la Clave | Variable de Entorno | Parametro en Codigo | Nivel Gratuito Disponible |
|---|---|---|---|---|
| Google Gemini | Google AI Studio | GEMINI_API_KEY |
api_key="..." |
Si (Nivel gratuito generoso) |
| Google Gemma (Local) | Ollama | No requiere clave | provider="ollama" |
100% Gratis y Offline |
| xAI Grok | Consola de xAI | XAI_API_KEY |
api_key="..." |
Pago por uso |
| OpenAI | Plataforma OpenAI | OPENAI_API_KEY |
api_key="..." |
Pago por uso |
| Anthropic Claude | Consola Anthropic | ANTHROPIC_API_KEY |
api_key="..." |
Pago por uso |
Instala unicamente los componentes que tu proyecto requiera:
# Nucleo base (Gemini, Gemma, Grok, OpenAI, Claude, CLI)
pip install chatflow-agent
`
# Con conector de Webhooks para WhatsApp (FastAPI + Uvicorn)
pip install "chatflow-agent[whatsapp]"
# Con conector de Bot para Telegram (python-telegram-bot)
pip install "chatflow-agent[telegram]"
# Paquete completo (Todos los canales y herramientas de desarrollo)
pip install "chatflow-agent[all]"
Guarda este codigo como asistente.py y ejecutalo con python asistente.py:
import asyncio
from chatflow_agent import Agent, Runner
# 1. Definir el agente
agente_soporte = Agent(
name="SoporteTienda",
model="gemini-2.5-flash", # O provider="openai", model="gpt-4o-mini"
instructions="Eres un asistente cordial de atencion al cliente para una tienda online.",
)
# 2. Registrar herramientas de negocio usando decoradores y tipado estándar
@agente_soporte.tool
def consultar_pedido(pedido_id: str) -> dict:
"""Consulta el estado de despacho y envio de un pedido por su identificador."""
return {
"pedido_id": pedido_id,
"estado": "En distribucion",
"transporte": "Envios Express",
"llegada_estimada": "Hoy antes de las 18:00 hs",
}
async def main():
runner = Runner(starting_agent=agente_soporte)
respuesta = await runner.run_async(
session_id="cliente_101",
user_message="Hola, ¿podrias revisar en que estado se encuentra mi pedido #PED-9921?",
)
print(f"[{respuesta.active_agent_name}]: {respuesta.content}")
if __name__ == "__main__":
asyncio.run(main())Ejecuta agentes de forma local en tu propia computadora, sin costos de API y con total privacidad de datos usando Ollama:
-
Inicia Ollama con Gemma:
ollama run gemma2:2b(ogemma2:9b) -
Ejecuta este script:
import asyncio
from chatflow_agent import Agent, Runner
# Se conecta a http://localhost:11434/v1 sin requerir claves de API
agente_local = Agent(
name="AnalistaLocal",
provider="ollama",
model="gemma2:2b",
instructions="Eres un analista financiero que procesa informacion confidencial de manera local.",
)
@agente_local.tool
def calcular_iva(subtotal: float, tasa_porcentaje: float = 21.0) -> dict:
"""Calcula el impuesto de IVA y el monto total de una operacion comercial."""
iva = subtotal * (tasa_porcentaje / 100.0)
return {"subtotal": subtotal, "iva": round(iva, 2), "total": round(subtotal + iva, 2)}
async def main():
runner = Runner(starting_agent=agente_local)
respuesta = await runner.run_async(
session_id="usuario_local",
user_message="Calcula el IVA de una factura de 550 dolares al 21%.",
)
print(f"[{respuesta.active_agent_name}]: {respuesta.content}")
if __name__ == "__main__":
asyncio.run(main())Los agentes pueden delegar la conversacion a especialistas de forma autonoma segun la intencion del usuario:
import asyncio
from chatflow_agent import Agent, Runner
# 1. Especialista: Soporte Tecnico
agente_tecnico = Agent(
name="SoporteTecnico",
model="gemini-2.5-flash",
instructions="Resuelves incidentes tecnicos de conectividad y servidores.",
)
@agente_tecnico.tool
def diagnosticar_nodo(nodo_id: str) -> dict:
"""Diagnostica el estado de un nodo de red."""
return {"nodo_id": nodo_id, "estado": "Operativo", "latencia_ms": 22}
# 2. Especialista: Facturacion y Cobranzas
agente_facturacion = Agent(
name="Facturacion",
model="gemini-2.5-flash",
instructions="Resuelves dudas sobre cobros, medios de pago y recibos.",
)
# 3. Recepcionista de Primera Linea (Frontline) con handoffs a especialistas
recepcionista = Agent(
name="Recepcion",
model="gemini-2.5-flash",
instructions="Saludas al usuario y derivas a SoporteTecnico o Facturacion segun corresponda.",
handoffs=[agente_tecnico, agente_facturacion],
)
async def main():
runner = Runner(starting_agent=recepcionista)
# El modelo detecta que es un problema tecnico y realiza la transferencia de forma automatica
respuesta = await runner.run_async(
session_id="cliente_77",
user_message="Tengo una caida en el nodo SRV-04. ¿Pueden revisarlo?",
)
print(f"Agente Activo: {respuesta.active_agent_name}") # Salida: SoporteTecnico
print(f"Respuesta: {respuesta.content}")
if __name__ == "__main__":
asyncio.run(main())Servidor de webhook listo para produccion, compatible con la API oficial de WhatsApp Business en Meta Cloud:
from chatflow_agent import Agent, Runner
from chatflow_agent.channels import WhatsAppChannel
agente_ventas = Agent(name="ConciergeWhatsApp", instructions="Atiende consultas comerciales.")
runner = Runner(starting_agent=agente_ventas)
canal_whatsapp = WhatsAppChannel(
verify_token="tu_token_verificacion_meta", # Configurado en tu app de Meta Developers
access_token="EAA...", # Token permanente de System User en Meta
app_secret="a1b2c3d4...", # Meta App Secret (Panel de Meta -> Configuración -> Básica)
phone_number_id="109876543210987", # WhatsApp Phone Number ID del panel de Meta
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.",
)
canal_whatsapp.attach(runner)
if __name__ == "__main__":
# Expone GET /webhook (verificacion de Meta) y POST /webhook (recepcion de mensajes)
canal_whatsapp.run(host="0.0.0.0", port=8000)¿Probando localmente? Usa un tunel como ngrok (
ngrok http 8000) o Cloudflare Tunnels para brindarle a Meta la URL publica HTTPS (https://tu-dominio.ngrok-free.app/webhook).
Note
Documentación interactiva en desarrollo: Por motivos de seguridad, la documentación interactiva de FastAPI (/docs, /redoc y /openapi.json) está deshabilitada por defecto en producción. Para habilitar Swagger UI y los esquemas OpenAPI durante el desarrollo local o debugging, pasa enable_docs=True al instanciar WhatsAppChannel:
canal_whatsapp = WhatsAppChannel(
...,
enable_docs=True, # Habilita /docs, /redoc y /openapi.json para desarrollo local
)Tip
Rate Limiting y Reverse Proxy en Producción: Para proteger los endpoints /webhook y /health contra ataques de inundación, denegación de servicio (DoS) y agotamiento de CPU por peticiones no autenticadas sin riesgo de descartar ráfagas legítimas de webhooks de Meta a nivel aplicación, ubica siempre Uvicorn detrás de un reverse proxy en producción (como Nginx con limit_req_zone o Cloudflare Rate Limiting):
limit_req_zone $binary_remote_addr zone=whatsapp_limit:10m rate=10r/s;
location /webhook {
limit_req zone=whatsapp_limit burst=20 nodelay;
proxy_pass http://127.0.0.1:8000;
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
}Conecta tu enjambre de agentes a un Bot de Telegram con polling asincrono y soporte nativo de comandos:
from chatflow_agent import Agent, Runner
from chatflow_agent.channels import TelegramChannel
agente_bot = Agent(name="TelegramBot", instructions="Asistes a los usuarios en Telegram.")
runner = Runner(starting_agent=agente_bot)
canal_telegram = TelegramChannel(
token="123456:ABC-DEF1234ghIkl-zyx57W2v1u123ew11", # Token provisto por @BotFather
fallback_message="Disculpa, ocurrio un inconveniente temporal. Por favor reintenta en breve.",
start_message="Bienvenido al bot. ¿En que podemos ayudarte hoy?",
)
canal_telegram.attach(runner)
if __name__ == "__main__":
# Ejecuta el bucle asincrono; gestiona /start, /reset e indicadores de escritura de forma automatica
canal_telegram.run()Prueba y depura enjambres de agentes directamente en tu consola con formato legible:
from chatflow_agent import Agent, Runner
from chatflow_agent.channels import CLIChannel
agente_terminal = Agent(name="AsistenteCLI", instructions="Responde de manera concisa y tecnica.")
runner = Runner(starting_agent=agente_terminal)
canal_cli = CLIChannel(session_id="prueba_desarrollo")
canal_cli.attach(runner)
canal_cli.run()chatflow-agent incorpora protecciones nativas disenadas especificamente para el trafico real de canales de mensajeria:
-
Candado de Concurrencia por Sesion (
asyncio.Lock): Si un usuario envia varios mensajes consecutivos en pocos segundos (mensajes rapidos o transcripciones de notas de voz), un cerrojo asincrono garantiza el procesamiento en estricto orden FIFO. Esto evita condiciones de carrera, ejecuciones cruzadas de herramientas y corrupcion de historiales. -
Escudo Anti-500 para Webhooks de Meta: Si tu endpoint devuelve un error 500, los servidores de Meta reintentan agresivamente la entrega, saturando el servidor.
WhatsAppChannelcaptura caidas o limites de cuota de los LLMs, responde HTTP 200 a Meta para detener la tormenta de reintentos, envia al cliente un mensaje amigable (fallback_message) y registra el diagnostico de forma limpia. -
Manejo Seguro de Archivos Multimedia: Audios, fotos y documentos son interceptados con un aviso claro (
unsupported_media_message) sin generar excepciones no controladas ni interrumpir la sesion en curso. -
Rollback ante Fallas en el Turno: Si una solicitud al LLM excede el tiempo de espera o falla a mitad de turno, ChatFlow revierte automaticamente el historial de conversacion al ultimo turno consistente en memoria y SQLite (nota: revierte el historial conversacional; no deshace efectos secundarios externos de tools ya ejecutadas), garantizando un contexto limpio sin turnos huerfanos.
chatflow-agent ofrece persistencia nativa con cero dependencias externas adicionales, garantizando que las conversaciones y los agentes especialistas activos sobrevivan a reinicios del servidor o reciclado de procesos.
Ideal para pruebas, scripts locales o sesiones efimeras:
from chatflow_agent import Runner, SessionStore
runner = Runner(starting_agent=agente_recepcion, session_store=SessionStore(default_max_turns=20))Basado en la libreria estandar sqlite3 de Python. Gestiona de manera segura transacciones atomicas y bloqueos de archivo en Windows:
from chatflow_agent import Runner, SQLiteSessionStore
# El historial y la derivacion al especialista activo se conservan entre reinicios del servidor
session_store = SQLiteSessionStore(db_path="conversaciones.db", default_max_turns=30)
runner = Runner(starting_agent=agente_recepcion, session_store=session_store)Ventajas clave:
-
Persistencia del Agente Activo: Si un cliente fue derivado a
Facturacion, el puntero se guarda en SQLite. Al reiniciar el servidor, los nuevos mensajes continuan directamente conFacturacionsin volver a comenzar en la recepcion. -
Poda por Ventana Deslizante (FIFO): Mantiene acotado el tamano del historial en la base de datos para no desbordar la ventana de contexto de los modelos.
-
Compatibilidad Omnicanal: Funciona exactamente igual en WhatsApp, Telegram, REST APIs y CLI.
En el directorio examples/ encontraras proyectos completos listos para clonar y ejecutar en tu entorno:
-
examples/whatsapp_persistent_bot.py: Bot de WhatsApp Business (Meta Cloud API) con persistencia SQLite, candados de concurrencia y jerarquia de 3 agentes con handoffs.
-
examples/telegram_persistent_bot.py: Bot de Telegram con memoria de sesion y retencion de especialista entre reinicios de polling.
-
examples/webhook_api_service.py: Servicio REST API asincrono con CORS y endpoints
/api/chaty/api/sessions/{session_id}listo para frontends en React, Next.js y apps moviles. -
examples/local_gemma_cli.py: Asistente de terminal 100% offline y gratuito con Google Gemma 2 via Ollama y sesiones persistentes en SQLite.
-
examples/multi_provider_swarm.py: Enjambre multi-proveedor que coordina Gemini, Gemma local, Grok y Claude en un unico flujo conversacional.
Created and maintained by / Creado y mantenido por Gustavo Baranda.
If you find chatflow-agent useful for your projects, consider giving it a star on GitHub!
Si encuentras util chatflow-agent para tus proyectos, considera darle una estrella en GitHub!
Contributions, issues, and feature requests are always welcome.
Contribuciones, reportes de errores y solicitudes de caracteristicas son siempre bienvenidos.
Distributed under the MIT License / Distribuido bajo la Licencia MIT.