OpenAI AgentKit brings chat-ready AI agents to developers
OpenAI rolls out AgentKit for chat‑embedded AI agents
OpenAI has unveiled AgentKit, a toolkit designed to help developers and business teams build, deploy, and manage AI agents directly within chat interfaces. Launched on Oct. 6, the suite aims to simplify multi‑agent workflows and tighten integration across OpenAI’s ecosystem.
What’s in the kit
AgentKit bundles three primary components:
- Agent Builder: a visual workspace for creating and versioning multi‑agent workflows.
- Connector Registry: an admin hub to control how tools and data link across OpenAI products.
- ChatKit: a framework for embedding customizable, chat‑based agent experiences into apps and services.
Evaluation and training updates
Alongside AgentKit, OpenAI introduced new evaluation (evals) features to help test and tune model performance:
- Datasets for agent evals that can expand over time using human annotations and automated graders.
- Trace grading to automatically surface and address workflow gaps.
- Automated prompt optimization that generates improved prompts from annotations and grader outputs.
OpenAI also expanded reinforcement fine‑tuning (RFT) for customizing reasoning models. RFT is generally available on the o4‑mini model and in private beta for GPT‑5, according to OpenAI. New beta capabilities include custom tool calls that train models to select the right tool at the right moment, which the company says improves reasoning.
Strategic positioning
Industry observers see AgentKit as a bid to make OpenAI a platform for third‑party software development, potentially evolving into an “agent store.” By enabling applications to use ChatGPT as the primary interface instead of a browser, the toolkit could lower integration costs, improve security, and increase developer velocity—while strengthening user stickiness around OpenAI’s ecosystem.
Competitive context
Analysts note AgentKit is among the first Western offerings to tightly fuse a consumer‑friendly chat interface with agentic AI. Chinese tech giant ByteDance offers similar agentic features within Duobao, but its capabilities are largely confined to the ByteDance ecosystem and don’t support third‑party software. In contrast, OpenAI’s approach emphasizes seamless chat integration across products.
Rivals in the U.S. are moving in the same direction. Google introduced a Gemini business calling feature in July, and Anthropic’s Sonnet 4.5 model lays the groundwork for agentic behaviours. OpenAI’s advantage, analysts say, is the ability to interact across its broader stack—models, developer tools such as Codex, APIs, and ChatGPT—through a single, familiar interface.
Apps inside ChatGPT
OpenAI also previewed a new generation of apps that users can chat with inside ChatGPT. Developers can begin building these experiences with a new Apps SDK, currently in preview.
Why it matters
For enterprises, AgentKit consolidates the tooling needed to design, test, and ship chat‑embedded agents. The unified interface could shorten time to value for agentic projects while raising the bar for rival platforms. With evals, RFT, and app‑building tools now aligned, OpenAI is signalling a push to make chat the primary front door for agentic AI.
What’s next
As AgentKit gains adoption, watch for an ecosystem of third‑party agents and tools built around ChatGPT, along with competitive responses from major model providers. The extent to which OpenAI can balance openness for developers with ecosystem control will shape how quickly agentic AI moves from pilots to production at scale.
