LangChain 1.0 is our first major stable release, marking our commitment to no breaking changes until 2.0. After three years of feedback, we've streamlined the framework to focus on what matters most: building agents quickly while having the flexibility to customize when needed.

What we shipped:

• create_agent abstraction: The fastest way to build an agent with any model provider, built on LangGraph runtime for reliability.

• Middleware system: Fine-grained control at every step of the agent loop. Ships with built-in middleware for human-in-the-loop, summarization, and PII redaction. Support for custom middleware lets you hook into any point in the agent execution.

• Improved structured output generation: Integrated directly into the main agent loop, reducing both latency and cost by eliminating extra LLM calls. Fine-grained control over generation strategy via tool calling or provider-native structured output.

• Standard content blocks: Provider-agnostic spec for model outputs that works consistently across OpenAI, Anthropic, and hundreds of other providers. Includes support for reasoning traces, citations, and server-side tool calls with full backward compatibility.

We also reduced the surface area of the package to focus on core abstractions. Legacy functionality has been moved to langchain-classic for backwards compatibility.

Check out our new documentation at docs.langchain.com or read our blog post to learn more.

Release overviews: Python, JavaScript

Migration guides: Python, JavaScript