---
type: Agentic Web Standard
title: "Open Responses Specification"
description: "How can an agent or application call LLM inference and server-side tool loops across multiple providers using a single vendor-neutral wire format?"
resource: "https://www.openresponses.org/specification"
tags: ["act", "endpoint", "linux-foundation", "expose-tools-to-agents", "agent-to-agent-messaging"]
status: stable
generated: { by: agenticweb-builder/1.0, at: 2026-10-25T00:00:00Z }
verified: { by: process:agenticweb-engine, at: 2026-10-24T00:00:00Z }
stale_after: 2026-11-23T00:00:00Z
sources:
  - id: ref-1
    resource: "https://www.openresponses.org/specification"
    title: "Open Responses launched in January 2026 under the Linux Foundation with OpenAI, Hugging Face, Vercel, OpenRouter, Ollama, vLLM, SGLang, and LM Studio as founding adopters."
    last_modified: 2026-01-15T00:00:00Z
  - id: ref-2
    resource: "https://github.com/openresponses/openresponses"
    title: "github.com/openresponses/openresponses hosts the normative OpenAPI 3.1 schema and conformance test suite for the /v1/responses agentic loop endpoint."
    last_modified: 2026-01-15T00:00:00Z
---

# Open Responses Specification

Open Responses Specification is an agentic-web standard (endpoint — /v1/responses) at step 5, Act, governed by Linux Foundation; its status is "Published specification & conformance suite under Linux…". It answers: "How can an agent or application call LLM inference and server-side tool loops across multiple providers using a single…?" Replacing or extending OpenAI Chat Completions API, implementation status is multi-implementation (OpenAI, Hugging Face, Vercel, OpenRouter, Ollama…), last verified 2026-10-24.

# Overview

**Open Responses** is an open inference and agentic-loop API specification launched in January 2026 under the Linux Foundation, seeded by OpenAI’s Responses API (`POST /v1/responses`) with founding support from Hugging Face, Vercel, OpenRouter, Ollama, vLLM, SGLang, LM Studio, LiteLLM, and Anyscale [^ref-1][^ref-2].

Unlike legacy `chat/completions` endpoints that exchange stateless message arrays, Open Responses defines polymorphic **items** (`message`, `function_call`, `function_call_output`, `reasoning`), structured **semantic streaming events** (`response.created`, `response.output_item.added`, `response.output_text.delta`, `response.completed`), and multi-turn server-side tool execution loops (`previous_response_id`, hosted MCP tool calls, and client-side function calls) in a vendor-neutral OpenAPI 3.1 contract [^ref-1][^ref-2].

# Examples

```http
POST /v1/responses
# parse: JSON response object with object="response" or streaming response.* SSE events
# on absent: HTTP 404 or non-Open-Responses JSON payload
```

# Knowledge Graph

* **Lifecycle Step**: [5 Act](/category/act/index.md)
* **Pairs with**: [Model Context Protocol](/okf/protocols/mcp.md), [Agent2Agent Protocol](/okf/protocols/a2a.md)

[^ref-1]: Open Responses launched in January 2026 under the Linux Foundation with OpenAI, Hugging Face, Vercel, OpenRouter, Ollama, vLLM, SGLang, and LM Studio as founding adopters. (https://www.openresponses.org/specification)
[^ref-2]: github.com/openresponses/openresponses hosts the normative OpenAPI 3.1 schema and conformance test suite for the /v1/responses agentic loop endpoint. (https://github.com/openresponses/openresponses)