langchain-agents

name	langchain-agents
description	> Use when this capability is needed.

When to Use

• Building new AI agents with LangChain v1+
• Migrating from create_react_agent to create_agent
• Wiring tools directly into an agent
• Choosing between string model IDs, init_chat_model(), and provider-specific clients
• Configuring system prompts and structured output

Critical Patterns

Pattern 1: Use create_agent

• Prefer create_agent for all new agent work.
• Treat create_react_agent as legacy.

Pattern 2: Pass tools directly

• Supply tools as a plain list.
• Do not wrap them in ToolNode.
• Keep tool definitions small and explicit.

Pattern 3: Choose model configuration intentionally

• Use a string for the common case.
• Use init_chat_model() when you need runtime controls like temperature or timeout.
• Use ChatOpenAI when you need provider-specific parameters.

Pattern 4: Use system_prompt for instructions

• Use a string for the common case.
• Use SystemMessage only when you need advanced provider features such as prompt caching.
• Do not use the old prompt parameter.

Pattern 5: Use response_format for structured output

• Pass a Pydantic model directly for the common case.
• Use ToolStrategy(Schema) when tool-calling behavior is required.
• Use ProviderStrategy(Schema) when the provider supports native structured output.
• Read the validated object from result["structured_response"].

Pattern 6: Langchain libraries

• Always prefer langchain features inside langchain module over langchain-community.
• Use langchain-community only when no other option for the use case is available.
• langchain-x are integrations to specific techs and providers, only use when the additional features over the main langchain library are required.

Code Examples

Example 1: Simple agent with a tool

from langchain.agents import create_agent
from langchain.tools import tool


@tool
def search(query: str) -> str:
    """Search the web for information."""
    return f"Results for: {query}"


agent = create_agent(
    model="gpt-4.1-mini",
    tools=[search],
    system_prompt="You are a helpful assistant."
)

Example 2: Agent with a custom model

from langchain.chat_models import init_chat_model
from langchain.agents import create_agent

model = init_chat_model(
    model="gpt-4.1",
    temperature=0.1,
    max_tokens=1000
)

agent = create_agent(
    model=model,
    tools=[],
    system_prompt="You are a helpful assistant."
)

Example 3: Agent with structured output

from langchain.agents import create_agent
from pydantic import BaseModel

class ContactInfo(BaseModel):
    name: str
    email: str

agent = create_agent(
    model="gpt-4.1-mini",
    tools=[],
    response_format=ContactInfo,
    system_prompt="Extract contact information."
)

result = agent.invoke({"messages": [{"role": "user", "content": "..."}]})
result["structured_response"]  # ContactInfo instance

Commands

uv add langchain

Resources

• Skills: See langchain-tests for testing agents and tools.

---

Source: ColRuDev/job-candidate-matcher — distributed by TomeVault.