#Introduction to LangChain

#What is LangChain?

LangChain is an open-source framework for building LLM applications, maintained by LangChain Inc.. It packages model calls + tools + state + orchestration into a composable agent harness, so you can ship chatbots, RAG systems, and automation agents in Python (and JavaScript).

LangChain 1.0 (2025) is a major consolidation: the core focuses on the agent loop, runs on the LangGraph runtime, and adds Middleware for context engineering and control.

#The ecosystem

┌─────────────────────────────────────────────────────────────┐
│                    LangChain ecosystem                       │
└─────────────────────────────────────────────────────────────┘
   ┌──────────────┐   ┌──────────────┐   ┌──────────────┐
   │  LangChain   │   │  LangGraph   │   │  LangSmith   │
   │  high-level  │   │  graph runtime│  │  trace/eval  │
   │  create_agent│   │  persistence │   │  observability│
   └──────────────┘   └──────────────┘   └──────────────┘
         │                    ▲
         └──── 1.0 built on ──┘

Key point: LangChain 1.0 agents run on LangGraph. Drop down to LangGraph when you need fine control; compose both (e.g. embed create_agent subgraphs in custom graphs).

#LangChain 1.0 at a glance

Official framing:

Agent = Model + Harness
Harness = model + prompt + tools + middleware (right context at the right time)

#Agent loop (mental model)

flowchart LR
  U[User input] --> M[Model]
  M --> T{Tools?}
  T -->|Yes| E[Run tools]
  E --> M
  T -->|No| R[Reply]
  MW[Middleware] -.-> M
  MW -.-> E

The model loops until the task is done. Middleware hooks (before_model, after_model, wrap_model_call, etc.) power summarization, memory, human-in-the-loop, and subagent delegation.

#Comparison

When to choose LangChain: agents, workflows, observability in one stack. For doc-heavy RAG-only apps, LlamaIndex or plain API may suffice.

#Good fit / caveats

Good fit: tool-using agents, RAG, persistent sessions, HITL, LangSmith tracing.

Watch out: one-shot prompts (use API directly), ultra-low latency (framework overhead), regional provider/compliance constraints.

#Packages

GitHub: langchain-ai/langchain, langchain-ai/langgraph

#Next steps

• Installation
• Quick Start
• Architecture & Core Concepts