What's new in LangChain v1 ↗

Original Documentation

Documentation Index#
Fetch the complete documentation index at: https://docs.langchain.com/llms.txt Use this file to discover all available pages before exploring further.

LangChain v1 is a focused, production-ready foundation for building agents. We’ve streamlined the framework around three core improvements:

The new standard for building agents in LangChain, replacing langgraph.prebuilt.create_react_agent.

A new content_blocks property that provides unified access to modern LLM features across providers.

The langchain namespace has been streamlined to focus on essential building blocks for agents, with legacy functionality moved to langchain-classic.

To upgrade,

pip install -U langchain

uv add langchain

For a complete list of changes, see the migration guide.

`create_agent`#

create_agent is the standard way to build agents in LangChain 1.0. It provides a simpler interface than langgraph.prebuilt.create_react_agent while offering greater customization potential by using middleware.

from langchain.agents import create_agent

agent = create_agent(
    model="claude-sonnet-4-6",
    tools=[search_web, analyze_data, send_email],
    system_prompt="You are a helpful research assistant."
)

result = agent.invoke({
    "messages": [
        {"role": "user", "content": "Research AI safety trends"}
    ]
})

Under the hood, create_agent is built on the basic agent loop – calling a model, letting it choose tools to execute, and then finishing when it calls no more tools:

For more information, see Agents.

Middleware#

Middleware is the defining feature of create_agent. It offers a highly customizable entry-point, raising the ceiling for what you can build.

Great agents require context engineering: getting the right information to the model at the right time. Middleware helps you control dynamic prompts, conversation summarization, selective tool access, state management, and guardrails through a composable abstraction.

Prebuilt middleware#

LangChain provides a few prebuilt middlewares for common patterns, including:

PIIMiddleware: Redact sensitive information before sending to the model
SummarizationMiddleware: Condense conversation history when it gets too long
HumanInTheLoopMiddleware: Require approval for sensitive tool calls

from langchain.agents import create_agent
from langchain.agents.middleware import (
    PIIMiddleware,
    SummarizationMiddleware,
    HumanInTheLoopMiddleware
)


agent = create_agent(
    model="claude-sonnet-4-6",
    tools=[read_email, send_email],
    middleware=[
        PIIMiddleware("email", strategy="redact", apply_to_input=True),
        PIIMiddleware(
            "phone_number",
            detector=(
                r"(?:\+?\d{1,3}[\s.-]?)?"
                r"(?:\(?\d{2,4}\)?[\s.-]?)?"
                r"\d{3,4}[\s.-]?\d{4}"
			),
			strategy="block"
        ),
        SummarizationMiddleware(
            model="claude-sonnet-4-6",
            trigger={"tokens": 500}
        ),
        HumanInTheLoopMiddleware(
            interrupt_on={
                "send_email": {
                    "allowed_decisions": ["approve", "edit", "reject"]
                }
            }
        ),
    ]
)

Custom middleware#

You can also build custom middleware to fit your needs. Middleware exposes hooks at each step in an agent’s execution:

Build custom middleware by implementing any of these hooks on a subclass of the AgentMiddleware class:

Hook	When it runs	Use cases
`before_agent`	Before calling the agent	Load memory, validate input
`before_model`	Before each LLM call	Update prompts, trim messages
`wrap_model_call`	Around each LLM call	Intercept and modify requests/responses
`wrap_tool_call`	Around each tool call	Intercept and modify tool execution
`after_model`	After each LLM response	Validate output, apply guardrails
`after_agent`	After agent completes	Save results, cleanup

Example custom middleware:

from dataclasses import dataclass
from typing import Callable

from langchain_openai import ChatOpenAI

from langchain.agents.middleware import (
    AgentMiddleware,
    ModelRequest
)
from langchain.agents.middleware.types import ModelResponse

@dataclass
class Context:
    user_expertise: str = "beginner"

class ExpertiseBasedToolMiddleware(AgentMiddleware):
    def wrap_model_call(
        self,
        request: ModelRequest,
        handler: Callable[[ModelRequest], ModelResponse]
    ) -> ModelResponse:
        user_level = request.runtime.context.user_expertise

        if user_level == "expert":
            # More powerful model
            model = ChatOpenAI(model="gpt-5")
            tools = [advanced_search, data_analysis]
        else:
            # Less powerful model
            model = ChatOpenAI(model="gpt-5-nano")
            tools = [simple_search, basic_calculator]

        return handler(request.override(model=model, tools=tools))

agent = create_agent(
    model="claude-sonnet-4-6",
    tools=[
        simple_search,
        advanced_search,
        basic_calculator,
        data_analysis
    ],
    middleware=[ExpertiseBasedToolMiddleware()],
    context_schema=Context
)

For more information, see the complete middleware guide.

Built on LangGraph#

Because create_agent is built on LangGraph, you automatically get built in support for long running and reliable agents via:

<span class=“card-start” data-card-raw=“title=“Persistence” icon=“database”"> Conversations automatically persist across sessions with built-in checkpointing

<span class=“card-start” data-card-raw=“title=“Streaming” icon=“droplet”"> Stream tokens, tool calls, and reasoning traces in real-time

<span class=“card-start” data-card-raw=“title=“Human-in-the-loop” icon=“hand-stop”"> Pause agent execution for human approval before sensitive actions

<span class=“card-start” data-card-raw=“title=“Time travel” icon=“history”"> Rewind conversations to any point and explore alternate paths and prompts

You don’t need to learn LangGraph to use these features—they work out of the box.

Structured output#

create_agent has improved structured output generation:

Main loop integration: Structured output is now generated in the main loop instead of requiring an additional LLM call
Structured output strategy: Models can choose between calling tools or using provider-side structured output generation
Cost reduction: Eliminates extra expense from additional LLM calls

from langchain.agents import create_agent
from langchain.agents.structured_output import ToolStrategy
from pydantic import BaseModel


class Weather(BaseModel):
    temperature: float
    condition: str

def weather_tool(city: str) -> str:
    """Get the weather for a city."""
    return f"it's sunny and 70 degrees in {city}"

agent = create_agent(
    "gpt-4.1-mini",
    tools=[weather_tool],
    response_format=ToolStrategy(Weather)
)

result = agent.invoke({
    "messages": [{"role": "user", "content": "What's the weather in SF?"}]
})

print(repr(result["structured_response"]))
# results in `Weather(temperature=70.0, condition='sunny')`

Error handling: Control error handling via the handle_errors parameter to ToolStrategy:

Parsing errors: Model generates data that doesn’t match desired structure
Multiple tool calls: Model generates 2+ tool calls for structured output schemas

Standard content blocks#

Content block support is currently only available for the following integrations:

Broader support for content blocks will be rolled out gradually across more providers.

The new content_blocks property introduces a standard representation for message content that works across providers:

from langchain_anthropic import ChatAnthropic

model = ChatAnthropic(model="claude-sonnet-4-6")
response = model.invoke("What's the capital of France?")

# Unified access to content blocks
for block in response.content_blocks:
    if block["type"] == "reasoning":
        print(f"Model reasoning: {block['reasoning']}")
    elif block["type"] == "text":
        print(f"Response: {block['text']}")
    elif block["type"] == "tool_call":
        print(f"Tool call: {block['name']}({block['args']})")

Benefits#

Provider agnostic: Access reasoning traces, citations, built-in tools (web search, code interpreters, etc.), and other features using the same API regardless of provider
Type safe: Full type hints for all content block types
Backward compatible: Standard content can be loaded lazily, so there are no associated breaking changes

For more information, see our guide on content blocks.

Simplified package#

LangChain v1 streamlines the langchain package namespace to focus on essential building blocks for agents. The refined namespace exposes the most useful and relevant functionality:

Namespace#

Module	What’s available	Notes
`langchain.agents`	`create_agent`, `AgentState`	Core agent creation functionality
`langchain.messages`	Message types, content blocks, `trim_messages`	Re-exported from `langchain-core`
`langchain.tools`	`@tool`, `BaseTool`, injection helpers	Re-exported from `langchain-core`
`langchain.chat_models`	`init_chat_model`, `BaseChatModel`	Unified model initialization
`langchain.embeddings`	`Embeddings`, `init_embeddings`	Embedding models

Most of these are re-exported from langchain-core for convenience, which gives you a focused API surface for building agents.

# Agent building
from langchain.agents import create_agent

# Messages and content
from langchain.messages import AIMessage, HumanMessage

# Tools
from langchain.tools import tool

# Model initialization
from langchain.chat_models import init_chat_model
from langchain.embeddings import init_embeddings

`langchain-classic`#

Legacy functionality has moved to langchain-classic to keep the core packages lean and focused.

What’s in langchain-classic:

Legacy chains and chain implementations
Retrievers (e.g. MultiQueryRetriever or anything from the previous langchain.retrievers module)
The indexing API
The hub module (for managing prompts programmatically)
langchain-community exports
Other deprecated functionality

If you use any of this functionality, install langchain-classic:

pip install langchain-classic

uv add langchain-classic

Then update your imports:

from langchain import ...  # [!code --]
from langchain_classic import ...  # [!code ++]

from langchain.chains import ...  # [!code --]
from langchain_classic.chains import ...  # [!code ++]

from langchain.retrievers import ...  # [!code --]
from langchain_classic.retrievers import ...  # [!code ++]

from langchain import hub  # [!code --]
from langchain_classic import hub  # [!code ++]

Migration guide#

See our migration guide for help updating your code to LangChain v1.

Reporting issues#

Please report any issues discovered with 1.0 on GitHub using the 'v1' label.

Additional resources#

Read the announcement

Deep dive into middleware

Full agent documentation

New content blocks API

How to migrate to LangChain v1

Report issues or contribute