AIGuard Python SDK

Python SDK for monitoring and logging AI model invocations with AIGuard.

Installation

pip install aiguard

Or install from source:

pip install -e /path/to/python-sdk

Usage

Initialize the client

from aiguard import AIGuard

guard = AIGuard(api_key="your-api-key")

Direct API calls

# Start an invocation
invocation = guard.start_invocation({
    "eventId": "unique-event-123",
    "event": "chat-completion",
    "userId": "user-456",
    "inputMessage": "Hello, how are you?",
    "model": "gpt-4",
    "convoId": "conversation-789",
    "properties": {"temperature": 0.7}
})

# ... make your AI call ...

# End the invocation
completed = guard.end_invocation({
    "eventId": "unique-event-123",
    "output": "I am doing well, thank you!"
})

Manual logging

guard.log({
    "model": "gpt-4",
    "input": [{"role": "user", "content": "Hello"}],
    "output": "Hi there!",
    "latency": 150,
    "tokens": {"prompt": 10, "completion": 5, "total": 15},
    "status": "success"
})

Wrap OpenAI client (automatic logging)

from openai import OpenAI
from aiguard import AIGuard

guard = AIGuard(api_key="your-api-key")
client = guard.wrap_openai(OpenAI())

# All chat completion calls are now automatically logged
response = client.chat.completions.create(
    model="gpt-4",
    messages=[{"role": "user", "content": "Hello!"}]
)

API Reference

AIGuard(api_key, base_url="http://localhost:5000/api")

Initialize the AIGuard client.

start_invocation(data) -> dict

Start tracking an invocation. Required fields: eventId, event, userId.

end_invocation(data) -> dict

End an invocation. Required field: eventId.

log(data) -> None

Fire-and-forget logging of a model call.

wrap_openai(client) -> client

Wrap an OpenAI client for automatic logging.