Início Rápido

# [min-version: 0.3.0]
# [no-exec: requires real openai api key for live response]
import os
from chatflow_agent import Agent, Runner

# 1. Definir a persona do agente, instruções e modelo de destino
# O chatflow-agent direciona automaticamente "gpt-4o" para o motor OpenAI usando OPENAI_API_KEY
agent = Agent(
    name="Assistant",
    instructions="You are a concise and helpful AI technical assistant.",
    model="gpt-4o",
)

# 2. Inicializar o Runner com seu agente inicial
runner = Runner(starting_agent=agent)

# 3. Executar um turno de conversa
response = runner.run(
    session_id="quickstart_session",
    user_message="Explain async Python in two concise sentences.",
)

# 4. Exibir a resposta
print("Agent Response:")
print(response.content)

python quickstart.py

# [min-version: 0.3.0]
# [no-exec: requires real openai api key for live response]
import os
from chatflow_agent import Agent, Runner, tool

# 1. Definir uma ferramenta com parâmetros tipados e docstrings
@tool(name="calculate_discount", description="Calculate discounted price from original amount and percentage.")
def calculate_discount(price: float, discount_percent: float) -> str:
    """Calculate the final price after applying a percentage discount.

    Args:
        price: The original price in dollars.
        discount_percent: The percentage discount to apply (e.g. 15 for 15%).
    """
    savings = price * (discount_percent / 100.0)
    final_price = price - savings
    return f"Original: ${price:.2f}, Savings: ${savings:.2f}, Final Price: ${final_price:.2f}"

# 2. Anexar a ferramenta ao Agente
agent = Agent(
    name="SalesBot",
    instructions="You help customers with purchase calculations. Always use calculate_discount for math.",
    model="gpt-4o",
    tools=[calculate_discount],
)

runner = Runner(starting_agent=agent)

# 3. Fazer uma pergunta que exija a execução da ferramenta
response = runner.run(
    session_id="sales_session_1",
    user_message="What is the final price for a jacket priced at $120 with a 25% discount?",
)
print(response.content)

# Inspecionar nomes das ferramentas executadas
for tool_name in response.tool_calls_executed:
    print(f"Tool executed: {tool_name}")

import asyncio
import os
from chatflow_agent import Agent, Runner

async def main() -> None:
    agent = Agent(
        name="AsyncBot",
        instructions="Provide concise answers.",
        model="gpt-4o",
    )

    runner = Runner(starting_agent=agent)

    # Turno assíncrono
    response = await runner.run_async(
        session_id="async_session_1",
        user_message="List three key advantages of asynchronous Python.",
    )

    print(response.content)

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

import os
from chatflow_agent import Agent, OpenAIEngine

# Opção 1: Configuração direta do modelo (os motores são resolvidos automaticamente)
openai_agent = Agent(
    name="OpenAIBot",
    instructions="You are powered by GPT-4o.",
    model="gpt-4o",
)

claude_agent = Agent(
    name="ClaudeBot",
    instructions="You are powered by Claude 3.5 Sonnet.",
    model="claude-3-5-sonnet-20241022",
)

gemini_agent = Agent(
    name="GeminiBot",
    instructions="You are powered by Google Gemini.",
    model="gemini-2.5-flash",
)

# Opção 2: Endpoints customizados via OpenAIEngine (Groq, Ollama, DeepSeek)
groq_engine = OpenAIEngine(
    base_url="https://api.groq.com/openai/v1",
    api_key=os.getenv("GROQ_API_KEY"),
    provider="groq",
)

groq_agent = Agent(
    name="GroqBot",
    instructions="You are powered by LLaMA 3.3 on Groq.",
    model="llama-3.3-70b-versatile",
    engine=groq_engine,
)