OpenAI on Monday (Developers' Day) announced AgentKit, a new toolkit to help developers and companies build, manage, and optimise AI agents.
CEO Sam Altman announced the game-changing toolkit at the third edition of the DevDay 2025 conference.
An AI agent is any AI capable of independently performing a set of tasks to get a particular output.
Apart from announcing AgentKit and its tools, which includes Agent Builder, ChatKit, Connector Registry, and Evals, the AI giant also announced that various third-party apps would now be integrated into the company's flagship ChatGPT.
OpenAI has also made Codex—an AI tool used in software engineering—available to everyone, and announced two new, cheaper real-time voice models.
Why it matters
Before the advent of toolkits like AgentKit, creating apps was a complicated technical process. Developers had to use a set of tools, write code, and spend weeks on designing and debugging the interface.
However, AgentKit streamlines the process from start to finish, making the process more accessible to the average person.
At the heart of AgentKit is Agent Builder, which is a canvas on which the AI agent can be created using visual workflows.
"Agent Builder provides a visual canvas for composing logic with drag-and-drop nodes, connecting tools, and configuring custom guardrails. It supports preview runs, inline eval configuration, and full versioning—ideal for fast iteration," OpenAI said in a press release.
Other tools include Connector Registry (which helps in managing data sharing across platforms), Guardrails (which prevents AI agents from doing harmful things or releasing sensitive information), ChatKit (which lets developers embed chatbots on their websites/apps), and Evals (which offers useful statistics and insights on workflows).
"We plan to add a standalone Workflows API and agent deployment options to ChatGPT soon," the OpenAI press release added.
