agent_failover.py

"""AI Agent with automatic failover across providers.

When using SoxAI, you don't need failover logic in your code —
the gateway handles it. But here's how you'd use multiple models
in different agent steps for cost optimization.
"""

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["SOXAI_API_KEY"],
    base_url="https://api.soxai.io/v1",
)


def classify(text: str) -> str:
    """Step 1: Classify intent with a cheap, fast model."""
    response = client.chat.completions.create(
        model="gpt-4o-mini",  # $0.15/1M tokens — cheapest option
        messages=[
            {"role": "system", "content": "Classify the user's intent in one word: question, task, or chat."},
            {"role": "user", "content": text},
        ],
        max_tokens=10,
    )
    return response.choices[0].message.content.strip().lower()


def reason(text: str) -> str:
    """Step 2: Deep reasoning with a capable model."""
    response = client.chat.completions.create(
        model="claude-sonnet-4-6",  # Best quality for reasoning
        messages=[
            {"role": "system", "content": "You are a helpful assistant. Think step by step."},
            {"role": "user", "content": text},
        ],
    )
    return response.choices[0].message.content


def summarize(text: str) -> str:
    """Step 3: Summarize with a balanced model."""
    response = client.chat.completions.create(
        model="deepseek-chat",  # Good quality at $0.27/1M tokens
        messages=[
            {"role": "user", "content": f"Summarize in 2 sentences:\n\n{text}"},
        ],
        max_tokens=100,
    )
    return response.choices[0].message.content


# Run the agent
user_input = "What are the pros and cons of microservices vs monolith architecture?"

intent = classify(user_input)
print(f"Intent: {intent}")

answer = reason(user_input)
print(f"\nFull answer:\n{answer}")

summary = summarize(answer)
print(f"\nSummary:\n{summary}")