OpenAI SDK

Building an Agent

If you prefer to build manually or want to understand the configuration details, follow the guide below.

Base URL and AnotherAI API Key Setup

AnotherAI provides a unified API that routes requests to various AI providers. To use AnotherAI instead of calling OpenAI directly, you need to:

Change the base URL - This redirects API calls from OpenAI's servers to AnotherAI, which then routes them to the appropriate provider while adding observability features
Configure your AnotherAI API key - This enables access to AnotherAI's features

For cloud-hosted AnotherAI, use your AnotherAI API key:

import openai

client = openai.OpenAI(
    base_url="https://api.anotherai.dev/v1",  # AnotherAI cloud endpoint
    api_key="aai-***",  # Your AnotherAI API key
)

For self-hosted AnotherAI, point to your local instance with your AnotherAI API key:

import openai

client = openai.OpenAI(
    base_url="http://localhost:8000/v1",  # Local AnotherAI instance
    api_key="aai-***",  # Your AnotherAI API key
)

Metadata

Agent Identification In order to distinguish between different agents in AnotherAI's web view, include an agent_id in your agent's metadata.

completion = client.chat.completions.create(
    model="gpt-4o-mini",
    messages=[{"role": "user", "content": "Analyze the sentiment of this product review"}],
    metadata={
        "agent_id": "product-review-sentiment", # recommended for observability
    }
)

Workflow Identification

If your agent is part of a workflow, it's recommended to include a trace_id and workflow_name in your metadata. You can read more about workflow set up here.

completion = client.chat.completions.create(
    model="gpt-4o-mini",
    messages=[{"role": "user", "content": "Analyze the sentiment of this product review"}],
    metadata={
        "agent_id": "product-review-sentiment",
        "workflow_name": "review-analysis-pipeline",
        "trace_id": "trace-123e4567-e89b",  # Unique ID for this workflow instance
    }
)

Input and Output Design

1. Input Variables

If there is variable content in your prompts, use Jinja2 templates to separate static prompts from dynamic content:

completion = client.chat.completions.create(
    model="claude-3-5-sonnet-20241022",
    messages=[{
        "role": "user", 
        "content": "Analyze the sentiment of this product review: {{review_text}}"
    }],
    extra_body={
        "input": {
            "review_text": "This product exceeded my expectations! The quality is amazing..."
        }
    },
    metadata={
        "agent_id": "product-review-sentiment",
        "workflow_name": "review-analysis-pipeline",
        "trace_id": "trace-123e4567-e89b",
    }
)

2. Structured Outputs

Structured outputs aren't required, but they're highly recommended, especially for agents that have multiple output fields.

from pydantic import BaseModel
from enum import Enum

class Sentiment(str, Enum):
    positive = "positive"
    negative = "negative"
    mixed = "mixed"

class SentimentAnalysis(BaseModel):
    sentiment: Sentiment
    explanation: str  # Why this sentiment was determined

completion = client.chat.completions.parse(
    model="gpt-4o",
    messages=[{
        "role": "user", 
        "content": "Analyze the sentiment of this product review: {{review_text}}"
    }],
    response_format=SentimentAnalysis,
    extra_body={
        "input": {
            "review_text": "This product exceeded my expectations! The quality is amazing..."
        }
    },
    metadata={
        "agent_id": "product-review-sentiment",
        "workflow_name": "review-analysis-pipeline",
        "trace_id": "trace-123e4567-e89b",
    }
)

result = completion.choices[0].message.parsed