LangChain

LangChain is an open-source framework for building Agent and LLM-driven applications, using a Monorepo architecture (core code in libs/ directory, langchain-core base package at version 1.3.1, Python 99.3%). It runs on the LangGraph orchestration framework, with an upper ecosystem covering Deep Agents, LangSmith observability platform, and extensive third-party integrations.

Core Capabilities#

• Model Access & Routing: Standardized model interface with unified abstraction layer for seamless switching between OpenAI, Anthropic Claude, Google Gemini, Ollama, AWS Bedrock, Azure, etc., avoiding vendor lock-in.
• Agent Building: create_agent() API for minimal-code fully customizable AI Agents; supports Multi-agent collaboration and Human-in-the-loop workflows.
• Deep Agents: Out-of-the-box advanced Agents (via deepagents package) with built-in automatic context compression, virtual file system, sub-Agent scheduling, and automatic planning.
• Tool Integration & Protocols: @tool decorator registers Python functions as Agent-callable tools with automatic schema extraction; supports Model Context Protocol (MCP).
• State & Memory Management: Short/long-term conversation memory with production-grade persistent storage; multi-session isolation via thread_id.
• Output Control: Streaming and structured output support.
• Retrieval & Context: Built-in RAG document retrieval and context engineering.
• Middleware System: Built-in and custom middleware.
• Engineering & Observability: LangSmith for call tracing, debugging, and evaluation; frontend integration patterns (Agent Chat UI); LangSmith Deployment for stateful, long-running Agent deployment.

Ecosystem Architecture#

• LangChain: High-level Agent and application framework.
• LangGraph: Low-level Agent orchestration framework (deterministic + intelligent workflows) providing persistent execution and state management.
• Deep Agents: Out-of-the-box advanced Agents built on LangChain Agents.
• LangSmith: Observability, evaluation, debugging, and deployment platform.
• Integrations: Third-party packages for models, tools, vector stores, etc.

Typical Use Cases#

• RAG Q&A systems connecting LLMs to internal/external data sources.
• Multi-model experimentation and A/B testing via unified interface.
• Tool-calling Agents that invoke APIs, file systems, databases, etc.
• Complex task planning with Deep Agents for multi-step reasoning.
• Enterprise LLM application development with built-in monitoring and evaluation.

Installation & Quick Start#

pip install langchain
# or
uv add langchain

Deep Agent extension:

uv add langchain deepagents

Unconfirmed Information#

• Exact repository location of the deepagents package (likely within monorepo libs/).
• LangChain Skills specific scope and repository address.
• LangSmith Deployment detailed features and pricing.
• Exact version number of the main langchain package (README only specifies langchain-core==1.3.1).