ambertrace 0.1.0

Python SDK for tracing LLM calls (OpenAI, Anthropic, Google Gemini) to AmberTrace observability platform

AmberTrace Python SDK

Lightweight, zero-code LLM observability for Python. Trace every API call to OpenAI, Anthropic, and Google automatically.

Installation

# Core SDK
pip install ambertrace

# With specific provider support
pip install ambertrace[openai]
pip install ambertrace[anthropic]
pip install ambertrace[gemini]

# All providers
pip install ambertrace[all]

Requirements: Python 3.8+

Quick Start

import ambertrace
from openai import OpenAI

# Initialize once at startup
ambertrace.init(api_key="at_...")

# Use your LLM SDK as normal — tracing happens automatically
client = OpenAI()
response = client.chat.completions.create(
    model="gpt-4",
    messages=[{"role": "user", "content": "Hello!"}]
)

# Flush before exit
ambertrace.flush()

Supported Providers

Provider	Models	SDK
OpenAI	GPT-5, GPT-4, GPT-4o, GPT-4o-mini, o1, o3	openai>=1.0.0
Anthropic	Claude Opus 4.5/4.6, Sonnet 4.5, Haiku	anthropic>=0.18.0
Google	Gemini Pro, Flash, 2.0, Gemma	google-generativeai>=0.3.0

Async Support

import asyncio
import openai
import ambertrace

async def main():
    ambertrace.init(api_key="at_...")
    client = openai.AsyncOpenAI()
    response = await client.chat.completions.create(
        model="gpt-4",
        messages=[{"role": "user", "content": "Hello!"}]
    )
    await ambertrace.flush_async()

asyncio.run(main())

Configuration

ambertrace.init(
    api_key="at_...",            # Required (or AMBERTRACE_API_KEY env var)
    environment="production",     # Environment tag for filtering
    debug=False,                  # Enable debug logging
    timeout=5.0,                  # Network timeout in seconds
    enabled=True,                 # Enable/disable tracing
)

API Reference

• ambertrace.init() — Initialize SDK and start tracing
• ambertrace.enable() / ambertrace.disable() — Toggle tracing at runtime
• ambertrace.is_enabled() — Check if tracing is active
• ambertrace.flush() — Block until pending traces are sent
• ambertrace.flush_async() — Async version of flush
• ambertrace.shutdown() — Flush, disable, and clean up

Links

• Documentation
• Portal
• GitHub

License

Apache 2.0