langchain-use-skill_skill

Overview

This skill is a compact LangChain 1.0 usage guide for building LLM-driven agents and applications. It provides quick references for Agent creation, Tools, Memory, Middleware, Runtime context, and structured output patterns. Use it when you need practical, copy-paste examples and patterns to integrate LangChain into Python projects.

How this skill works

The guide explains core workflows: create_agent -> ReAct loop -> Tool invocation -> result return. It shows how to define @tool functions, access ToolRuntime and context, and wire in short-term and long-term memory via checkpointers and stores. It also covers middleware decorators for intercepting model and tool calls and how to enable structured outputs and streaming.

When to use it

• When you need a minimal working Agent example in Python.
• When integrating model providers (OpenAI, Anthropic) with LangChain.
• When adding Tools that access runtime context or external services.
• When you must persist conversation memory (in-memory or Postgres).
• When implementing middleware to preprocess messages or wrap tool calls.

Best practices

• Prefer Python 3.10+ and install provider integrations (langchain-openai, langchain-anthropic) alongside langchain.
• Define Tools with clear signatures and docstrings; use ToolRuntime for context-aware tools.
• Use checkpointers (InMemorySaver, PostgresSaver) for session persistence and thread_id to isolate conversations.
• Start with a low temperature for deterministic outputs in production; enable streaming for UX where needed.
• Use structured response formats (Pydantic/dataclass) for predictable outputs and downstream parsing.

Example use cases

• A chat assistant that calls get_weather and returns structured responses with confidence scores.
• A user-specific agent that reads user_id from ToolRuntime to personalize replies.
• A production agent persisting session history in PostgresSaver for multi-session analytics.
• Middleware that trims or normalizes messages before model calls to reduce token use.
• A retrieval-augmented agent combining a retriever and tools to answer domain questions.