OpenAI Launches AgentKit: a one-stop platform for building and shipping AI agents

OpenAI unveiled the AgentKit DevDay event, a unified toolkit meant to take AI agents from prototype to production on a single platform. Instead of stitching together orchestration code, ad-hoc connectors, one-off eval scripts, and custom front-ends, AgentKit bundles visual design tools, embeddable UIs, governance controls, and evaluation workflows into one cohesive stack. OpenAI positions it as the fastest path to build, deploy, and optimize agentic applications at scale. 

What’s inside AgentKit

Agent Builder. A visual, drag-and-drop canvas for composing multi-agent workflows with proper versioning, preview runs, and guardrails. Teams can wire up tools, add if/else logic, configure evals inline, and iterate quickly, OpenAI highlights customers reporting dramatic reductions in iteration cycles and time-to-launch. Templates help you start from proven patterns rather than a blank page. 

ChatKit. Shipping an agent usually means weeks of front-end work. ChatKit provides embeddable, customizable chat experiences (branding, theming, streaming, threads) you can drop into your product to make agents feel native without reinventing the UX. Early adopters have used it for support assistants, onboarding guides, and research copilots. 

Evals for agents. Production agents rise or fall on measurement. OpenAI is expanding its eval suite with datasets, trace grading (end-to-end, step-level scoring), automated prompt optimization, and third-party model support, so you can benchmark agent behavior and improve it continuously, even if you’re comparing multiple model providers. 

Connector Registry + Guardrails. A central admin panel governs how agents access internal and third-party systems across ChatGPT and the API. It consolidates integrations (e.g., Google Drive, SharePoint, Teams, Dropbox) and supports third-party MCP servers. On the safety side, OpenAI Guardrails libraries (Python/JS) add jailbreak/PII defenses and validation that can be toggled directly in Agent Builder. 

Code-first path with the Agents SDK. If you prefer code over canvases, OpenAI Agents SDK (Python and JS/TS) provides a minimal set of primitives, agents, tools, handoffs, sessions, and guardrails, with built-in tracing and evaluation hooks. It’s a production-ready successor to earlier experiments like Swarm, and integrates with the Responses API. openai.github.io

Why it matters

AgentKit reflects a broader shift from “chatbots” to agentic systems that execute multi-step workflows, call tools, and coordinate across services. By collapsing orchestration, UI, and evaluation into a single platform, OpenAI is addressing the biggest friction points developers cite when operationalizing agents. The launch also signals competitive intent against other providers racing to ship integrated agent stacks for enterprises. TechCrunch

Pricing and availability

Effective today, ChatKit and the new Evals capabilities are generally available. Agent Builder enters beta, and the Connector Registry is rolling out in beta to organizations using OpenAI Global Admin Console (covering API, ChatGPT Enterprise, and Edu). OpenAI says these tools are included under standard API model pricing, with a Workflows API and agent deployment options in ChatGPT on the roadmap.

Not to be confused…

“AgentKit” has been used by others in the ecosystem (for example, Coinbase AgentKit for giving agents on-chain wallets). OpenAI AgentKit is a separate, end-to-end platform focused on agent orchestration, UI, governance, and evaluation, not crypto wallets. If you see the name elsewhere, check the vendor context. GitHub

Getting started

Developers can explore the Agent Platform overview (for the product lineup and deployment model) and the Agents SDK docs (for code-first workflows and tracing). Together with Agent Builder and ChatKit, these give teams parallel paths, visual and programmatic, to ship reliable agentic applications faster, with stronger governance and measurable performance. 

About Author

Netanel Siboni is a technology leader specializing in AI, cloud, and virtualization. As the founder of Voxfor, he has guided hundreds of projects in hosting, SaaS, and e-commerce with proven results. Connect with Netanel Siboni on LinkedIn to learn more or collaborate on future projects.