AnoSys Logger for OpenAI Agents - Python

Automatically capture and send OpenAI API calls (and more) to AnoSys for monitoring, analytics, and observability.

Features

• ✨ Automatic Instrumentation: Automatically captures traces from OpenAI API calls via OpenTelemetry.
• ✨ Manual Instrumentation: Easy-to-use decorator @anosys_logger for logging any function (sync or async).
• ✨ Raw Logging: Send custom JSON data directly to AnoSys for unstructured events.
• ✨ Async & Streaming Support: Fully supports Python's asyncio and streaming response aggregation.
• ✨ OpenTelemetry Standards: Follows LLM / Gen AI semantic conventions.

Installation

Install the package via pip:

pip install anosys-logger-4-openai-agents

Configuration

You need an AnoSys API Key to send data. Set it as an environment variable:

export ANOSYS_API_KEY="your_anosys_api_key"

Or create a .env file in your project root:

ANOSYS_API_KEY=your_anosys_api_key

Usage

1. Automatic Tracing (with OpenAI Agents)

To automatically capture traces from OpenAI agents, register the AnosysOpenAIAgentsLogger as a trace processor. This requires the traceai-openai-agents package.

import os
import contextvars
from agents import add_trace_processor
from AnosysLoggers import AnosysOpenAIAgentsLogger

# 1. Set API Keys
os.environ["OPENAI_API_KEY"] = "your_openai_key"
os.environ['ANOSYS_API_KEY'] = "your_anosys_key"

# 2. (Optional) Setup User Context
current_user_context = contextvars.ContextVar("current_user_context")
current_user_context.set({"session_id": "session_123", "user_id": "user_456"})

# 3. Register the Logger
add_trace_processor(AnosysOpenAIAgentsLogger(get_user_context=current_user_context.get))

# Now, any agent execution will be automatically logged to AnoSys.

2. Manual Logging (Decorator)

Use the @anosys_logger decorator to log inputs, outputs, and execution details of any function.

Synchronous Example

from AnosysLoggers import anosys_logger

@anosys_logger(source="my_sync_function")
def calculate_sum(a, b):
    return a + b

# Calling the function will automatically log data to AnoSys
result = calculate_sum(5, 10)

Asynchronous Example

import asyncio
from AnosysLoggers import anosys_logger

@anosys_logger(source="my_async_function")
async def fetch_data(url):
    await asyncio.sleep(1)  # Simulate network delay
    return {"data": "sample", "url": url}

# Run the async function
asyncio.run(fetch_data("https://example.com"))

Streaming (Async Generator) Example

@anosys_logger(source="my_streaming_function")
async def stream_data():
    yield "part1"
    yield "part2"

# The logger will capture and aggregate the streamed output
async for chunk in stream_data():
    print(chunk)

3. Raw Logging

If you need to log unstructured data or custom events, use anosys_raw_logger.

from AnosysLoggers import anosys_raw_logger

data = {
    "event": "user_signup",
    "user_id": "12345",
    "status": "success"
}

anosys_raw_logger(data)

What Data is Captured?

OpenTelemetry Semantic Conventions

Following the Gen AI standards:

• gen_ai.system: "openai"
• gen_ai.request.model: Model requested (e.g., "gpt-4")
• gen_ai.usage.input_tokens: Prompt token count
• gen_ai.usage.output_tokens: Completion token count
• gen_ai.response.finish_reasons: Why the generation stopped

Additional Metadata

• Request/response messages (optional)
• Execution duration and timestamps
• Error types and full stack traces
• Custom source identifiers

Advanced Usage

Custom API URL or Proxy

If you need to point to a specific AnoSys instance:

from AnosysLoggers import setup_api

setup_api(path="https://custom.anosys.endpoint/api/log")

Troubleshooting

• Missing API Key: Ensure ANOSYS_API_KEY is set in your environment.
• Import Error: Ensure you installed anosys-logger-4-openai-agents.
• Async Errors: Ensure you are await-ing async functions decorated with @anosys_logger.

License

MIT License. See LICENSE for details.