Principle:Langchain ai Langchain Tool Execution

Knowledge Sources	LangChain ReAct Pattern
Domains	Agentic_AI, Tool_Use
Last Updated	2026-02-11 00:00 GMT

Overview

The execution of a tool call requested by an LLM, producing a ToolMessage that links the result back to the original call via a unique identifier.

Description

After the model emits tool calls, the application must execute each tool and package the results as ToolMessage objects. Each ToolMessage carries:

• content: The tool's output (string or structured data)
• tool_call_id: The ID linking this result to the specific tool call in the AIMessage
• name: The tool's name (optional but recommended)

This linking mechanism allows the model to match results to their corresponding requests when multiple tools are called in parallel.

Usage

Execute tool calls in the application code between receiving the model's tool call response and re-invoking the model with the results. This is the "action" step in the ReAct (Reason + Act) pattern.

Theoretical Basis

Tool execution follows the ReAct loop pattern:

# Abstract algorithm (not real code)
for tool_call in ai_message.tool_calls:
    tool = tool_registry[tool_call.name]
    result = tool.invoke(tool_call)
    tool_messages.append(ToolMessage(
        content=result,
        tool_call_id=tool_call.id,
    ))

Related Pages

Implemented By

• Implementation:Langchain_ai_Langchain_BaseTool_Invoke