langchain-middleware

What problem does it solve?

Middleware provides extensibility to customize LangChain agent behavior by injecting logic at key execution points such as before_model, after_model, wrap_model_call, and wrap_tool_call.

Core Features & Use Cases

• Hook-based customization: run code before and after model/tool calls for validation, logging, or input shaping.
• Wrap methods: intercept and modify requests and responses to route through dynamic logic or guardrails.
• Composability: chain multiple middleware components into a single, flexible workflow.
• Flexible definitions: implement as a class-based middleware or decorator-based hooks, with support for async operation and context.
• Real-world scenarios: enforce guardrails, dynamically switch models, or filter available tools based on user context.

Quick Start

Create a simple LoggingMiddleware class extending AgentMiddleware and attach it to your agent to observe before_model behavior.