OpenAI Responses API

Harness Guides uses OpenAI's Responses API as the default OpenAI reference surface.

Official docs:

• Responses API overview
• Responses guide
• Tools guide

Why this is the default here

The Responses API is the right fit for harness documentation because it brings model output, tool use, streaming, long-running work, and conversation continuity into one surface. That maps much more cleanly to agent harness design than teaching everything through legacy chat-completions patterns.

Concepts used across this site

• responses.create for the main loop entrypoint
• input for user and system-side content
• instructions for high-level runtime guidance
• previous_response_id for response continuity
• tools for built-in tools, remote MCP, and custom function calls
• tool_choice when the harness wants tighter control
• parallel_tool_calls when the runtime allows safe concurrency
• stream for incremental model output and tool events
• background for long-running work that should outlive one synchronous wait
• include when the runtime needs richer returned data

Python example

from openai import OpenAI

client = OpenAI()

response = client.responses.create(
    model="gpt-5",
    instructions="You are a coding harness. Use tools when needed.",
    input="Summarize the repository layout and call tools if necessary.",
    tools=[
        {
            "type": "function",
            "name": "read_file",
            "description": "Read a file from disk",
            "parameters": {
                "type": "object",
                "properties": {
                    "path": {"type": "string"},
                },
                "required": ["path"],
            },
        }
    ],
    parallel_tool_calls=True,
)

How to use this page

• Read it before the query loop and tool chapters.
• Use it as the vocabulary guide for every OpenAI-specific example.
• Keep architecture decisions provider-portable, even when the examples use Responses API.