A toolkit announced by OpenAI at DevDay in October 2025 that bundles building, deploying, and evaluating AI agents. You design multi-agent workflows on the drag-and-drop visual canvas “Agent Builder,” embed them into your own service with the chat UI “ChatKit,” and measure and improve agent quality with the evaluation feature “Evals.” A distinctive point is that you can prototype workflows without writing code and, when needed, export them as Agents SDK code (TypeScript / Python). Note that the discontinuation of Agent Builder and Evals (on November 30, 2026) was announced in June 2026, so checking this migration schedule is essential before adopting them.
Key Features
- Agent Builder (visual canvas): Design multi-step agent workflows by connecting nodes via drag and drop. Configure typed inputs and outputs, test runs with real data, and then deploy
- ChatKit (embeddable chat UI): Embed the agents you build into your own web services and apps as a chat interface. Streaming output and display of reasoning steps come ready out of the box
- Evals (evaluation features): Measure agent response quality using datasets and trace grading, and feed the results back into prompt improvements
- Connector Registry / Guardrails: A Connector Registry lets administrators centrally manage connections to external data sources, and Guardrails provide safety measures such as masking personal information
- Code export to the Agents SDK: Export workflows built in Agent Builder as TypeScript / Python code and move to code-based development (full conversion of the workflow graph is not guaranteed; some parts require manual recreation)
Pricing
| Item | Price | Notes |
|---|---|---|
| AgentKit itself | No additional charge | Included in standard API model pricing |
| Agent runs | Usage-based | API charges according to the token rates of the models you use |
| ChatKit storage | Charged beyond the free tier | Billing began on November 1, 2025 |
Pricing is as of August 2026. Please check the official site for the latest pricing.
Pros & Cons
✅ Pros
- Visually design multi-agent workflows without writing code, enabling fast prototyping
- ChatKit dramatically reduces the effort of implementing a chat UI
- The toolkit itself has no additional charge; you only pay for API usage at run time
- An export path to the Agents SDK allows a gradual transition from no-code to code development
- As an official OpenAI product, integration with the latest models and the Responses API is smooth
⚠️ Cons
- Agent Builder and Evals are announced to end on November 30, 2026, making them hard to recommend for new full-scale adoption (OpenAI directs code development to the Agents SDK and natural-language use cases to Workspace Agents in ChatGPT)
- SDK export does not guarantee full conversion of workflows; branching logic and similar parts must be rebuilt manually
- Tied to OpenAI’s models and ecosystem; combining with other vendors’ models is not the intended use
- As shown by major components being discontinued shortly after launch, there is platform policy-change risk
Comparison with Similar Services
| Comparison | AgentKit | Dify | n8n | LangGraph |
|---|---|---|---|---|
| Provider | OpenAI | LangGenius | n8n GmbH | LangChain |
| Build method | Visual canvas + SDK | Visual workflows | Node-based automation | Code (Python / JS) |
| Supported models | Mainly OpenAI models | Multi-vendor | Multi-vendor | Multi-vendor |
| UI embedding | ChatKit | Embeddable web apps | None (backend-focused) | None (framework) |
| Self-hosting | Not available | Available | Available | Available |
Who Is It For
- Developers already using the OpenAI API who want to prototype agent workflows quickly
- Teams that want to embed a chat-style agent into their own service while keeping UI implementation effort low (ChatKit continues to be offered)
- Projects that plan to validate with no-code first and then move to code development with the Agents SDK
- People who want to learn a workflow for quantitatively evaluating agent response quality (mind the Evals end-of-service schedule)
Summary
AgentKit is an ambitious toolkit that bundles agent design, UI, and evaluation under OpenAI’s official umbrella. At the same time, its core components Agent Builder and Evals are set to end on November 30, 2026, so today it is realistic to see it as two things: a learning and validation environment for visual prototyping, and ChatKit, which continues to be offered. If you are about to develop agents in earnest, plan around the migration destinations — the Agents SDK and Workspace Agents in ChatGPT — while making use of the continuing components such as ChatKit and Guardrails.