OpenAI recently announced a comprehensive upgrade to its Agents SDK, bringing several important feature updates for developers. This upgrade not only expands programming language support but also introduces groundbreaking voice interaction technology.
The Agents SDK now officially supports TypeScript, with functionality identical to the Python version. Developers can utilize fine-grained control mechanisms and human approval options to manage AI agent behavior more precisely. The SDK is designed to simplify multi-agent workflows, offering configurable language models, agent switching capabilities, built-in security controls, and a complete suite of analytics tools.
Breakthrough in Voice Agent Technology
The newly launched RealtimeAgent feature is the highlight of this upgrade. Developers can now create voice agents that run on both the client and server sides, supporting voice interruption functions and real-time tool invocation during conversations.
The accompanying "Traces" dashboard has undergone significant updates, displaying real-time operational data of voice agents, including audio stream status, tool usage, and interruption count statistics, providing developers with comprehensive performance monitoring.
OpenAI has also improved the underlying language models that support these features. The updated "GPT-4O-Realtime-Preview-2025-06-03" version shows significant improvements in instruction execution accuracy, tool invocation reliability, and graceful handling of interruptions. This model is already available to developers through the real-time and chat APIs.
API Strategy Adjustment
In terms of product roadmap, OpenAI plans to gradually phase out the existing Assistants API and fully transition to the new Responses API after achieving feature parity by mid-2026. The Chat Completions API will continue to serve developers who do not require tool integration, but OpenAI recommends transitioning to the Responses API as soon as possible to access stronger expansion capabilities.
The Agents SDK is fully compatible with the OpenAI API and other third-party model providers that adhere to the Chat Completions standard, providing developers with flexible options.