Open Responses lets you run a fully self-hosted version of OpenAI's Responses API. It works seamlessly with any large language model (LLM) provider—whether it's Claude, Qwen, Deepseek R1, Ollama, or others. It's a fully-compatible drop-in replacement for the official API. Swap out OpenAI without changing your existing Agents SDK code.
Just run npx -y open-responses init and then:
from openai import AsyncOpenAI
from agents import set_default_openai_client
set_default_openai_client(AsyncOpenAI(base_url="http://localhost:8080/"))
agent = Agent(name="Test Agent", ...)
# ...🔥🔥🔥
From OpenAI docs on Responses API:
The Responses API is our newest core API and an agentic API primitive, combining the simplicity of Chat Completions with the ability to do more agentic tasks. > As model capabilities evolve, the Responses API is a flexible foundation for building action-oriented applications, with built-in tools:
- Web search
- File search
- Computer use
You can read about it in more detail on their announcement blog post.
Tip
This project is developed by the team behind Julep AI, the open-source platform making it easy for data teams to build, deploy, and scale stateful AI agents and workflows. Check us out on github: 
- 🔄 Bring Your Own Model - Compatible with any LLM provider you prefer.
- 🔒 Privacy First - Fully self-hosted, giving you total control over your data.
- 🔌 Easy Switch - Drop-in replacement compatible with OpenAI’s official Agents SDK.
- 🚀 Fast Setup - Get started quickly with Docker or our straightforward CLI.
- 🛠️ Built-in Tools - Supports automatic tool calls like web searches using open-source alternatives.
One simple command to get going:
npx -y open-responses init
# or: uvx open-responses initUsing the Agents SDK
from openai import AsyncOpenAI
from agents import set_default_openai_client
# Create and configure the OpenAI client
custom_client = AsyncOpenAI(base_url="http://localhost:8080/", api_key="YOUR_RESPONSES_API_KEY")
set_default_openai_client(custom_client)
agent = Agent(
name="Test Agent",
instructions="You are a helpful assistant that provides concise responses."
model="openrouter/deepseek/deepseek-r1"
)
result = await Runner.run(agent, "Hello! Are you working correctly?")
print(result.final_output)More examples here.
Using the OpenAI SDK
import { OpenAI } from 'openai';
const client = new OpenAI({
baseURL: 'http://localhost:8080/',
apiKey: "RESPONSE_API_KEY"
});
const response = await client.responses.create({
model: "gpt-4o-mini",
input: "What's the population of the world today?"
});
console.log(response.output[0].content[0].text);More examples here.
import os
from openai import OpenAI
client = OpenAI(
base_url="http://localhost:8080/",
api_key=os.getenv("RESPONSE_API_KEY")
)
response = client.responses.create(
model="gpt-4o-mini",
input="What's the population of the world today?"
)
print(response.output[0].content[0].text)More examples here.
We publish pre-built docker images which you can run using docker compose directly.
mkdir julep-responses-api
cd julep-responses-api
wget https://u.julep.ai/responses-env.example -O .env
wget https://u.julep.ai/responses-compose.yaml -O docker-compose.yml
docker compose up --watchOpen Responses is proudly built by Julep AI—the open-source platform empowering data and ML teams to rapidly create, deploy, and manage stateful AI workflows and intelligent agents at scale.
We’d love your contributions! Open Responses is licensed under Apache-2.0.