The difference between create_tool_calling_agent and create_react_agent

[theprobugmaker]

What is the difference between create_tool_calling_agent and create_react_agent? And when to use one or the other?

[dosubot[bot]]

Hey @theprobugmaker! I'm here to help you with any bugs, questions, or contributions you have. Just tag me in your issues or pull requests, and I'll be right with you. If you need assistance, feel free to ask!

The create_tool_calling_agent and create_react_agent serve different purposes within the LangChain framework:

• create_tool_calling_agent: This is used in the traditional LangChain framework to create an agent that can call tools based on a defined prompt and model. It involves manually managing the agent's state and execution flow, with a max_iterations parameter to control the number of iterations.

• create_react_agent: Part of the LangGraph framework, this manages the agent's state as a list of messages and processes them until there are no more tool calls. It automatically handles conversation history and state management, making it suitable for complex interactions. It uses a recursion_limit to control execution flow.

When to use each:

• Use create_tool_calling_agent for more control over execution and state management within the traditional LangChain framework.
• Use create_react_agent for complex interactions requiring automatic state management and when leveraging LangGraph's flexibility is beneficial ^[1].

To continue talking to Dosu, mention @dosu.

---

^{Help Dosu learn! Give it feedback: Great Response | Irrelevant Answer | Incorrect Sources | Too Verbose | Hallucination | Bug Report | Other}

[theprobugmaker]

@dosu ReAct is a general paradigm, right? What I'm confused is to know if create_tool_calling_agent is using it behind the scenes or it's not.

[dosubot[bot]]

The create_tool_calling_agent function does align with the ReAct paradigm through its use of intermediate steps and tool outputs, which are formatted and processed as part of the agent's reasoning and action-taking process. It incorporates elements like a message_formatter and an agent_scratchpad to manage these steps, allowing the agent to reflect on past actions and decisions, which is characteristic of the ReAct paradigm ^[1]. Additionally, comments in the documentation reference the creation of a ReAct-style agent using this function ^[2].

To continue talking to Dosu, mention @dosu.

---

^{Help Dosu learn! Give it feedback: Great Response | Irrelevant Answer | Incorrect Sources | Too Verbose | Hallucination | Bug Report | Other}