OpenResponses Python

OpenResponses Python provides a robust, typed interface for interacting with LLMs that adhere to the Open Responses standard. It includes a high-performance async client, Pydantic V2 models, and utilities for building compliant providers.

✨ Features

• Standardization: Strict Pydantic V2 models for InputText, MessageItem, ReasoningItem, ToolCallItem, ensuring type safety and validation.
• High Performance: Built on httpx with full async/await support and efficient streaming.
• Semantic Streaming: First-class support for streaming reasoning (thinking) and text deltas separately.
• Provider Agnostic: Seamlessly switch between OpenRouter, OpenAI, Ollama, LM Studio, and HuggingFace TGI.
• Developer-First: Fully typed, thoroughly tested, and ready for production.

📦 Installation

Install via pip:

pip install openresponses-python

We recommend using uv for modern dependency management:

uv add openresponses-python

🚀 Quick Start

Here's how to send a request to a compliant provider (e.g., OpenRouter or a local proxy).

1. Simple Request

import asyncio
from openresponses.client import AsyncOpenResponsesClient

async def main():
    # Initialize client (defaults to local proxy if base_url not set)
    # Ensure you have a provider running (see Examples below)
    client = AsyncOpenResponsesClient(base_url="http://localhost:8001")

    response = await client.create(
        model="deepseek/deepseek-r1",
        input="Explain semantic streaming.",
    )

    # Access the output items directly
    for item in response.output:
        if item.type == "message":
            print(f"Answer: {item.content}")

if __name__ == "__main__":
    asyncio.run(main())

2. Semantic Streaming

Stream reasoning ("thinking") and the final response in real-time.

stream = await client.create(
    model="deepseek/deepseek-r1",
    input="Why is the sky blue?",
    stream=True
)

async for event in stream:
    if event.event == "response.reasoning.delta":
            print(f"🧠 {event.data['delta']}", end="", flush=True)
    elif event.event == "response.text.delta":
            print(f"🤖 {event.data['delta']}", end="", flush=True)

🔌 Supported Providers

This package includes proxy examples to adapt popular providers to the Open Responses standard.

🛠️ Development

We use uv for dependency management and make for task automation.

git clone https://github.com/uday/openresponses-python.git
cd openresponses-python
make install      # Sync dependencies
make test         # Run tests
make lint         # Run linters
make docs-serve   # Preview documentation

🔮 Future Scope

In alignment with the Open Responses standard, we are planning the following enhancements:

• Expanded Provider Ecosystem: Native integrations for Anthropic, Google Vertex AI, and AWS Bedrock.
• Framework Adapters: Drop-in compatibility with LangChain and LlamaIndex for seamless orchestration.
• Observability: Built-in OpenTelemetry tracing for full visibility into requests and stream events.
• Enterprise resilience: Advanced retries, circuit breakers, and failover strategies.
• Smart Caching: Semantic caching strategies to reduce latency and API costs.
• CLI Tools: A comprehensive CLI for testing providers and debugging stream deltas.

🤝 Contributing

We welcome contributions! Please see CONTRIBUTION.md for guidelines.

📄 License

This project is licensed under the MIT License.