doppel-sdk 0.2.0

Official Doppel Python SDK

doppel-sdk — Python

Official Doppel SDK for Python. Wrap your existing LLM client and every call is captured and forwarded to Doppel — no change to your call sites.

Supports OpenAI, OpenRouter (and the OpenAI-compatible family: DeepSeek, Qwen, Grok, Mistral, …), Anthropic, and Google Gemini — across sync and async clients, and streaming and non-streaming calls. Zero runtime dependencies.

Install

pip install doppel-sdk

Usage

from doppel_sdk import DoppelClient
from openai import OpenAI

# Reads DOPPEL_API_KEY from the environment (or pass api_key=...).
doppel = DoppelClient(shadow_model="gpt-4o-mini")

client = doppel.wrap_openai(OpenAI())
client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "Hello"}],
)

Other providers:

doppel.wrap_anthropic(Anthropic())        # messages.create
doppel.wrap_google(genai.Client())        # google-genai (or a GenerativeModel)
doppel.wrap_openrouter(OpenAI(base_url=".../openrouter"))

Async clients (AsyncOpenAI, AsyncAnthropic, client.aio.models) work the same way — the interceptor detects and handles them automatically. Streaming calls are captured too; for OpenAI/OpenRouter the SDK auto-enables stream_options.include_usage so token usage is recorded.

In short-lived processes (serverless, CLI), flush pending deliveries before exit:

doppel.flush()

Configuration

Setting	Argument	Environment variable	Default
API key	api_key	DOPPEL_API_KEY	— (required)
Server URL	server_url	DOPPEL_SERVER_URL	https://api.doppel.in
Shadow model	shadow_model	—	none (chosen in the dashboard)
Debug logs	debug	DOPPEL_DEBUG (1/true)	off

Development

pip install -e ".[dev]"
pytest

Publishing

Pushing a py-v* tag runs the tests, builds, and publishes to PyPI (trusted publishing — no token):

# bump version in pyproject.toml first, then:
git tag py-v0.1.0
git push origin py-v0.1.0