Middleware in LangChain 1.0 gives you fine-grained control around the core agent loop of models + tools. It includes:

• Human-in-the-loop so you can approve, modify, or reject risky tool calls before execution

• Summarization of older messages as you approach token limits, so you can manage conversation history

Read the blog about middleware and try out middleware in the LangChain 1.0 alpha here.