OpenAI Operator: API for own AI agents published
Create and integrate AI agents yourself: OpenAI makes Responses API, tools and SDK available to developers.
(Image: Shutterstock/ioda)
Developers should be able to create their own AI agents with the help of OpenAI, similar to the Operator and Deep Research –, which are the company's general AI agent and an agent specializing in research. OpenAI is releasing a new Responses API, integrated tools and a new software development kit (ADK) for this purpose.
According to the blog post, this is just the first set of building blocks that OpenAI is making available to developers. The Responses API combines the existing Chat Completions API and the Assistant API, whereby the former is responsible for the communicative part and the latter contains the tool use functions, meaning that the agent can actually act and not just communicate. The Responses API contains integrated tools such as Web Search, i.e., the ability to access the Internet, File Search for large documents and Computer Use to use applications on the computer. However, you have to use gpt-4o or gpt-4o mini, for example, for the web search to actually work.
Responses API replaces Assistants API
The new SDK is used to orchestrate single and multi-agent workflows, “including integrated observability tools”. These are intended to help overgrow agent workflows and make them easier to understand.
Videos by heise
The API is now included for users with a developer account, at no additional cost. The Chat Completion API is still available separately. However, OpenAI recommends using the Responses API for new integrations, even if agent capabilities are not yet required. The end is already in sight for the Assistants API – By mid-2026, OpenAI expects it to be completely shut down and replaced by the Responses API.
(emw)
