OpenAI unveils Responses API, open source Agents SDK, letting developers build their own Deep Research and Operator

Openai is launching New API and tool suite Designed to help developers and businesses build AI-powered agents more effectively. These are delivered on some very same technologies that power their first-party AI agents In-depth research (This independently searches the Internet to develop richly researched, well-organized and cited reports) and Operator (Its tool is used to automatically control the web browser cursor based on the user’s text description and perform operations such as finding a sports ticket or booking).

Now, by accessing the building blocks behind these powerful first-party agents, developers can build their own third-party competitors or more domain products and services targeting their use cases and audiences.

OpenAI’s latest advancements in inference, multi-modal processing and security mechanisms have laid the foundation for these capabilities, especially its “O” family of inference models (O1 and O3).

“It’s hard to overestimate how the key reasoning model of AI agents enables AI agents,” Olivier Godement, product leader at the Openai platform, said in a video call with VentureBeat. “One of the biggest limitations before was dealing with long-distance tasks such as planning.”

But the company said that until now, developers did not have the necessary tools to easily build them into production-ready applications for businesses and their customers.

Execute the workflow that users want and be independent

To address these hurdles, OpenAI has launched several new products: Response API, built-in tools and file search for web searches, computer usage tools and open source proxy SDK. The response API allows developers to set up agents on their technology, but the SDK can help them link agents to other web tools and processes to perform the “workflow” required to autonomy for the “workflow” that users or enterprises want.

These tools are designed to simplify the development of AI agents by reducing the need for a wide range of timely engineering and custom orchestration logic. They should also make it an attractive proposition that can continue to build an attractive proposition on Openai’s platform even in the face of competition from Chinese players Manus,,,,, Alibaba’s Qwen,,,,, DeepSeek and local competitors, e.g. Human and Google.

While other players do provide developers’ tools and products, the continuous evolution of the OpenAI developer platform makes it difficult for people to be a “one-stop shop” that wants to leverage the latest AI advancements in a clean, easy to use, fast and affordable way.

In a move certain to send the AI ​​blogosphere and social media space aflutter, OpenAI is getting back into open source in a big way with the release of its Agents SDK, a toolkit designed to help developers manage, coordinate and optimize agent workflows — even building agents powered by other, non-OpenAI models such as those by competitors Anthropic and Google, or open-source models from DeepSeek, Qwen, Mistral and Meta’s Llama Family.

“SDK proxy is open source, allowing enterprises to mix different models,” Godement said. “We don’t want to force anyone to just use the Openai model.”

The SDK provides key features such as:

• Configurable proxy: An AI model with predefined instructions and tool access.
• Intelligent handover: The mechanism for transmitting tasks between agents between contexts.
• Built-in guardrail: Security measures for input verification and content audit.
• Tracking and observability: Tools for debugging and optimizing proxy performance.

“With the agent SDK, developers can track exactly what the agent is doing – what tasks it generates, the data it collects and how it generates responses,” Nikunj Handa, PM of the OpenAI API team, told VentureBeat.

New responses provided by the API

The centerpiece of this update is the Response API, which combines the functionality of the OpenAI Chat Completion API with the tool usage capabilities of the Assistant API, which will be deprecated in mid-2026. According to the company.

This integration enables developers to take advantage of multiple built-in tools in a single API call, making it easier to build applications that require complex multi-step interactions.

The response API initially supports three built-in tools:

• Online search: Provide real-time referenced answers by obtaining information from the network.
• File Search: Use metadata filtering and optimized query processing to retrieve relevant information from large document repositories.
• Computer usage tools: Enables the AI ​​agent to perform operations on the computer, such as browsing, data entry and navigation software interfaces.

“By responding to APIs, developers can have a better understanding of what the model is doing – what tools it calls, why it calls them, and the decisions made before and after these calls,” Handa said.

With these capabilities, OpenAI envisions the response API as the basis for proxy applications, eliminating the need for multiple external integrations. Now all developers can use the API and use it at OpenAI’s standard token and tool rates.

Additionally, OpenAI notes that while the chat completion API will continue to receive updates, the response API is considered its superset. Developers who require built-in tools or multi-step model interactions should use the Response API for new integrations.

OpenAI also gets its web search, file search and computer usage tools directly through the response API. These tools enable AI agents to access real-world information, retrieve context from documents, and interact with digital environments more efficiently.

Web search provides developers with real-time information with references

The new web search tool allows developers to integrate real-time search capabilities into their applications, making it useful for research assistants, shopping guides, and content aggregation tools. It provides a source of responses to ensure that users can verify the accuracy of information.

“The first thing we started was built-in tools, such as web search, which allow models to access real-time information,” Handa said. “This is the tool that powers Chatgpt search, and now we’re bringing it to the API.”

OpenAI also confirms that web search results in the API will include clear references, allowing users to click on the original source. Developers can use web search as part of a wider search system that includes proprietary data sources.

File Search: Smart Document Retrieval on Private Cloud

Using file search tools, AI agents can quickly retrieve relevant information from large collections of documents. The tool supports a variety of file formats and includes features such as query optimization, metadata filtering, and custom ranking for more accurate results.

“The third tool we launched is file search, which allows developers to easily get all the data, store it in our system, and extract the correct information with high accuracy,” Handa explained.

The file search tool costs $2.50 per thousand queries and the storage fee is $0.10 per GB per day (the first GB is free).

Developers can now also access computers using technology powered by OpenAI operators

Computers use tools extend the proxy functionality beyond simple text-based tasks by allowing AI to interact with computer interfaces.

Powered by OpenAI’s Computer Usage Agent (CUA) model, the tool converts AI-generated operations into executable commands, enabling tasks such as data entry and web navigation.

“We also launched a computer usage tool that allows interaction with the graphical user interface without an existing task API,” Handa noted.

Currently, computer usage tools can be used as a preview of research for selected developers in the 3-5 usage layers. Pricing is set at $12 per million in investment tokens and $12 per million in output tokens.

What does this mean for business leaders

For IT team leaders, CTOs and intermediate managers want to optimize workflows, OpenAI’s new tools provide clear pathways to automate and expand AI-driven processes without extensive custom development.

Built-in web search and file search capabilities enable businesses to quickly integrate AI-powered information retrieval into their existing systems, while computer-using tools can automatically interact with legacy applications that lack API access.

The open source proxy SDK further enables organizations to coordinate AI-driven workflows across teams, making it easier to deploy agents that improve efficiency in areas such as customer support, document processing, and market research.

With the built-in enterprise security and observability of these tools, decision makers can adopt AI solutions with greater transparency and control, ensuring compliance and performance monitoring at scale.

What’s next?

Openai believes that these new versions are the first step in establishing a comprehensive platform for AI agents. The company plans to launch additional tools and integrations in the coming months to help developers deploy, evaluate and scale proxy applications more effectively.

“We think the coming months will be crucial for deploying more and more agents,” Godement said. “We have already done this with first-party agents like delving into the deepest, but Openai won’t build every agent, which is why we have a developer platform.”

OpenAI also said it will continue to improve security features of proxy applications, including safeguards for rapid injection and unauthorized data access.

Developers interested in building with new tools can explore OpenAI’s documentation and API playgrounds to get started today.