OpenAI's AgentKit promises big productivity gains. But if you want to use it with Claude or another LLM, you'll probably need an alternative.

AgentKit from OpenAI promises to cut the time required to create AI agents based on its Large Language Model (LLM), ChatGPT. OpenAI launched the toolkit in October 2025 to streamline agent development from prototype to production.

But what are its capabilities? For example, does it support using another LLM from vendors like Anthropic, such as Claude? And what are the benefits – and risks – of building your applications using OpenAI's AgentKit as a foundation?

In this article, we'll explore what AgentKit does, its benefits and limitations, and how to create AI agents quickly with any LLM.

What are AI agents?

LLMs are deep learning models trained on vast amounts of data. They use a neural network architecture called a transformer, acting as a statistical prediction machine that enables them to generate artifacts, such as text and images, that mimic human output.

While LLMs are powerful, they're reactive, responding to direct user queries. They also struggle to break down and respond to complex queries directly.

AI agents are AI-driven systems that autonomously perform tasks. Unlike traditional LLMs that simply respond to prompts, AI agents are proactive. They use LLMs to create either single agents or multi-agent networks that can reason and act, independent of human input.

An AI agent typically consists of three key components:

LLM: provides natural language processing and content generation capabilities.

Tools: External APIs, data sources, memory systems, machine learning models, etc. that the agent can call. The agent can either feed tool responses to the LLM or hand LLM output to a tool (e.g., an ML model that evaluates the LLM's output).

Workflow orchestration: Coordinates performing tasks in a defined sequence. This enables creating complex AI agents composed of multiple LLMs, tools, and even other agents, each with its own specialized responsibilities.

What is AgentKit?

AgentKit is OpenAI's toolkit to support developers building agents based on ChatGPT. It brings together several previously fragmented tools into a single platform. The suite includes Agent Builder, ChatKit, and Evals.

Agent Builder

With Agent Builder, you can build agentic workflows visually. It provides a drag-and-drop canvas where you can compose application logic flow, easily connect tools, and establish guardrails. The visual interface enables engineers and subject matter experts to collaborate in a single UI.

The workflow builder supports everything from simple linear tasks to complex multi-agent systems. Each workflow supports full versioning, so teams can experiment freely and roll back when needed. Agent Builder also includes inline evaluation configuration and preview runs for fast iteration.

ChatKit

ChatKit is a toolkit for embedding customizable chat-based agent experiences into your applications. You can drop OpenAI solutions into an HTML app to add instant chatbot support. You can add ChatKit to your HTML apps easily using JavaScript or React.

You can tailor ChatKit's UI to match your brand with deep customization options. It also provides support for managing threads, streaming responses, and showing model processing logic.

ChatKit handles the complex frontend work that would otherwise take weeks to build from scratch. Combined with OpenAI AgentKit, it creates a powerful one-two punch for building both the backend and frontend components of production-ready AI assistants.

Evals

Evals helps you measure and improve agent performance. The evaluation capabilities include datasets with automated graders and annotations, trace grading that runs end-to-end assessments of workflows, and automated prompt optimization.

Evals can trace the performance of calls to third-party tools and LLMs as well as ChatGPT. This means you can compare how your agents would perform with different models like Claude, Gemini, or GPT, even if Agent Builder itself is locked to OpenAI.

Can I use AgentKit with Claude?

The short answer is no. AgentKit is exclusively locked to OpenAI's ecosystem. The agent component only supports GPT models.

This vendor lock-in extends throughout the entire AgentKit suite. While you can use Evals to test external models, you can't actually run Agent Builder workflows with Claude or any other non-OpenAI LLM. Your entire agentic workflow must execute on GPT models.

AgentKit is a limited agent builder

This isn't the only limitation of Agent Kit, unfortunately. Compared to other options on the market, AgentKit's functionality is overall very limited.

Only a chat trigger is available, so you can only create chat agents. It lacks the ability to add code blocks or switch between code and visual creation modes. The platform also only integrates with a dozen or so tools out of the box, limiting your ability to connect to external data sources or automate complex workflows.

For teams that need flexibility in their agent architecture, these constraints can be deal-breakers. You're forced to work within OpenAI's predetermined patterns rather than adapting the tools to your specific use cases.

Alternatives to AgentKit when using Claude

How can you build Claude agents without AgentKit? Here are a couple of alternatives that give you the flexibility to work with Anthropic's powerful LLM.

Alternative 1: Claude Agent SDK

Claude doesn't support its own visual builder. However, Anthropic does have its own Agents SDK for building agents from code using Claude.

The Claude Agent SDK is the same infrastructure that powers Claude Code, Anthropic's agentic coding solution. The key design principle is to give your agents access to a computer, allowing them to work like humans do. By giving Claude tools to run bash commands, edit files, create files, and search files, you can build general-purpose agents.

The SDK provides primitives for building Claude agents for whatever workflow you're trying to automate. It includes features like automatic context management, a rich tool ecosystem with file operations and code execution, advanced permissions for fine-grained control, and built-in error handling.

Benefits: The Claude Agent SDK enables rapid development of Claude-based agents for developers. It leverages the same proven infrastructure that powers Anthropic's production tools. The SDK supports both TypeScript and Python, making it accessible to a wide range of developers who want to build agents with tool calling capabilities.

Drawbacks: You don't gain the time-saving benefits of using a visual builder. The code-first approach is harder to collaborate on with non-technical stakeholders. Teams without strong development resources may struggle with the complexity of building agents from scratch, especially when debugging complex workflows or orchestrating multiple agent interactions.

Alternative 2: Non-vendor-specific AI agent builders

Arguably, the most critical feature of AgentKit is the agent builder—it significantly cuts development time for building new agents. However, the issue with using either Agent Kit or the Claude Agent SDK is vendor lock-in. You can't easily swap out another LLM if you find it performs better or is cheaper for your specific scenarios.

Using an agent builder tool from a company that's not also an LLM vendor brings a number of advantages:

• You get support for a larger number of LLMs, giving you the flexibility to choose the best model for each task.
• You can integrate with a large number of external data stores and data streaming services through APIs and connectors.
• You gain access to features not supported by either Agent Builder or the Claude SDK.

Vendor-neutral platforms also protect you from being locked into a single provider's pricing model and feature roadmap. As the AI landscape evolves rapidly, this flexibility becomes increasingly valuable.

Use Langflow to build Claude agents effortlessly

Langflow is a more robust alternative to OpenAI AgentKit and the Claude SDK. It's a full visual builder with a rich feature set, designed for enterprise-grade agent development.

The platform integrates with the most popular language models, including Claude, GPT, and many others. It also supports additional models via Bundles. This makes it easy to switch language models if you find another one that best serves your needs—whether you're building customer support agents, AI assistants, or automated workflows.

Langflow supports both visual editing and Python code, as well as effortlessly switching between them. This means you can use the visual builder for rapid prototyping and simple workflows, then drop down to code when you need more control. Hard tasks get the flexibility they need, while easy tasks stay easy.

You can build new agents – simple agents as well as multi-agent networks – by turning existing agents into tools. This composability enables you to create sophisticated systems in which specialized agents collaborate to solve complex problems with orchestrated tool calls.

The platform provides out-of-the-box components for common patterns like retrieval-augmented generation (RAG), message history management, and external system connections. These pre-built templates reduce development time while remaining fully customizable, allowing you to iterate quickly on agent behavior.

Unlike AgentKit's limitations, Langflow supports multiple trigger types beyond just chat. You can build agents that respond to API calls, scheduled events, data stream changes, and more. You can also expose your Langflow workflows to other external systems by turning the workflow into a Model Context Protocol (MCP) server.

Langflow is available as a self-hosted Docker container for teams with security or compliance requirements. It's also part of the DataStax AI Platform, which pairs the visual builder with Astra DB for vector and NoSQL database capabilities, plus Astra Streaming for real-time data pipelines.

The visual workflow builder makes it easy for cross-functional teams to collaborate. Product managers, engineers, and domain experts can all contribute to agent development without requiring deep technical expertise from everyone involved. The modular architecture and extensive docs on GitHub make it easy to understand and extend agent functionality.

Conclusion

Without a doubt, OpenAI's AgentKit makes agent development more accessible. However, its exclusive focus on OpenAI's ecosystem limits its utility for teams that want to work with Claude or other LLMs from Anthropic and other vendors.

The Claude Agent SDK provides a powerful code-first alternative for developers comfortable building from scratch with Python or TypeScript. But for teams seeking the productivity benefits of visual development without vendor lock-in, vendor-neutral platforms offer the best of both worlds.

Langflow combines the rapid prototyping capabilities of visual builders with the flexibility to work with any LLM and switch seamlessly to code when needed. That reduces development time while preserving architectural flexibility—whether you're automating customer support, building AI copilots, or creating multi-agent orchestration systems.

Whether you're building simple chatbots or complex multi-agent systems, choosing the right development platform can make the difference between a successful deployment and a stalled project. Consider your team's skills, your need for model flexibility, and your long-term architectural requirements before locking yourself in.

Ready to build your first Claude agent? Sign up for Langflow today for free and try the tutorial to build your first agent. You'll be up and running in minutes, with the flexibility to grow your agents as your needs evolve.