OpenAI redefines the way Agent workflows are built with Agent SDK

Just now, OpenAI officially launched the first set of basic building blocks, aiming to help developers and companies build practical and reliable agent systems.

Over the past year, OpenAI has continued to expand model capabilities, adding new features such as advanced reasoning, multimodal interaction, and new security technologies, all of which have laid a solid foundation for the model to handle complex multi-step tasks.

However, turning these capabilities into production-ready agents remains challenging, often requiring extensive prompt iteration and custom orchestration logic while lacking sufficient visualization support and built-in accessibility features.

To overcome these challenges, they carefully created a set of newly designed APIs and tools to simplify the development process of proxy-based applications:

• New Responses API: Combines the simplicity of the Chat Completions API with the tooling of the Assistants API, designed specifically for building agents.

• Built-in tools: Covering network search, file search, computer use, etc., providing powerful support for agents.

• New Agents SDK: supports orchestration of single-agent and multi-agent workflows, and flexibly responds to various complex scenarios.

• Integrated observability tools: can track and inspect agent workflow execution, making the development process more transparent.

These new tools optimize core agent logic, orchestration, and interaction processes, greatly lowering the threshold for developers to build agent applications. In the coming weeks and months, more tools and features will be launched to further simplify and accelerate the development process of building intelligent applications based on our platform.

Introducing the Responses API

The Responses API is a new API primitive designed to build agents using OpenAI's built-in tools. It combines the simplicity of Chat Completions with the tooling power of the Assistants API.

As model capabilities continue to evolve, the Responses API will provide developers with a more flexible infrastructure for building intelligent applications. With a single Responses API call, developers will be able to use multiple tools and model rounds to solve increasingly complex tasks.

The Responses API will power new built-in tools for web search, file search, and computer usage . These tools are designed to work together to connect models to the real world, making them more useful for accomplishing tasks.

Web Search

Get fast, up-to-date answers from the web, with clear, relevant citations. Web search is available as a tool in the Responses API when using gpt-4o and gpt-4o-mini, and can be combined with other tools or function calls.

const  response  = await openai.responses.create ( {    model :  "gpt-4o" ,    tools : [ {  type :  "web_search_preview"  } ],    input :  "What was a positive news story that happened today?" ,});
console.log (response.output_text) ;

File Search

Easily retrieve relevant information from massive amounts of documents with the improved file search tool. The tool supports multiple file types and features query optimization, metadata filtering, and custom re-ranking to provide fast, accurate search results. Integration can be completed with just a few lines of code through the Responses API.

const  productDocs  = await openai.vectorStores. create ({    name :  "Product Documentation" ,    file_ids : [file1.id, file2.id, file3.id],});
const  response  = await openai.responses.create ( {    model :  "gpt-4o-mini" ,    tools : [{        type :  "file_search" ,        vector_store_ids : [productDocs.id],    }],    input :  "What is deep research by OpenAI?" ,});
console.log (response.output_text) ;

Computer Use

You can build agents that can complete tasks on computers using the computer usage tools in the Responses API, which are driven by the same computer usage agent (CUA) model that powers Operators.

Built-in computer usage tools capture mouse and keyboard actions generated by the model, allowing developers to automate computer usage tasks by converting these actions directly into executable commands in its running environment.

const  response  = await openai.responses.create ( {    model :  "computer-use-preview" ,    tools : [{        type :  "computer_use_preview" ,        display_width :  1024 ,        display_height :  768 ,        environment :  "browser" ,    }],    truncation :  "auto" ,    input :  "I'm looking for a new camera. Help me find the best one." ,});
console.log (response.output) ;

The API also brings several usability improvements, including a unified project-based design, simplified polymorphism, intuitive streaming events, and SDK helpers like response.output_text for easily getting the text output of a model.

Agents SDK 

In addition to building the core logic of the agent and providing it with access to the tools to make it work, developers also need to orchestrate the agent workflow.

The new open source Agents SDK simplifies the orchestration of multi-agent workflows and offers significant improvements over Swarm⁠—the experimental SDK—which has been widely adopted by the developer community and successfully deployed by multiple customers.

Improvements include:

• Agents: Easily configurable LLMs with clear instructions and built-in tools.

• Handoff: Intelligently transfer control between agents.

• Guardrails: Configurable safety checks for input and output validation.

• Tracing and Observability: Visualize agent execution traces to debug and optimize performance.

from  agents  import  Agent, Runner, WebSearchTool, function_tool, guardrail
@function_tooldef  submit_refund_request ( item_id:  str , reason:  str ):    # Your refund logic goes here    return  "success"
support_agent = Agent(    name = "Support & Returns" ,    instructions= "You are a support agent who can submit refunds [...]" ,    tools=[submit_refund_request],)
shopping_agent = Agent(    name= "Shopping Assistant" ,    instructions= "You are a shopping assistant who can search the web [...]" ,    tools=[WebSearchTool()],)
triage_agent = Agent(    name= "Triage Agent" ,    instructions= "Route the user to the correct agent." ,    handoffs=[shopping_agent, support_agent],)
output = Runner.run_sync(    starting_agent=triage_agent,    input = "What shoes might work best with my outfit so far?" ,)

The Agents SDK is suitable for a variety of real-world use cases, including customer support automation, multi-step research, content generation, code review, and sales lead generation.

The Agents SDK works with the Responses API and the Chat Completions API. It also works with other vendors' models as long as they provide a Chat Completions-style API endpoint.

Developers can integrate it into their Python codebases immediately, with Node.js support coming soon.