I want something simpler than LangGraph for building agents

If you want something simpler than LangGraph, a few good options depend on what you mean by “agent”:

1. Plain Python with a small loop

Best if you want maximum simplicity.

• Use a system prompt
• Call the model
• If it returns a tool call, run the tool
• Feed the result back
• Repeat until done

This is often enough for simple agents and is much easier to reason about than a graph framework.

2. OpenAI Responses API / function calling

Good if you want built-in tool use without extra orchestration.

• Define tools/functions
• Let the model choose when to call them
• You handle the loop in a few lines of code

This is usually the cleanest “simple agent” setup.

3. PydanticAI

A nice lightweight framework if you want structure without complexity.

• Typed outputs
• Tool calling
• Simple agent abstractions
• Easier than LangGraph for many use cases

4. Smolagents

Very lightweight and designed for agent-style workflows.

• Simple to get started
• Tool use and code-style agent patterns
• Good if you want something opinionated but minimal

5. CrewAI

If your main goal is multi-agent workflows, but still simpler than LangGraph.

• Easier conceptual model
• More “roles/tasks” oriented
• Less flexible than LangGraph, but simpler

---

My practical recommendation

• For the simplest thing: use plain Python + tool calling
• For a small framework: use PydanticAI or Smolagents
• For multi-agent coordination: consider CrewAI
• If you already use OpenAI APIs: use Responses API directly

If you want, I can show you:

1. a minimal agent in plain Python, or
2. a PydanticAI example, or
3. a LangGraph-to-simple-python rewrite.