🦜🔗 LangChain Python Examples

A structured collection of LangChain examples covering agents, models, messages, tools, short-term memory, event streaming, graph streaming, structured outputs, middleware, guardrails, runtime context, context engineering, Model Context Protocol (MCP), and human-in-the-loop — built with Python and designed for learning the LangChain / LangGraph ecosystem.

---

📁 Project Structure

langchain-python/
├── 1_agents/              # LangChain agent patterns
├── 2_models/              # LLM model initialization & parameters
├── 3_messages/            # Message types, history & streaming
├── 4_tools/               # Tool definitions, schemas & dynamic selection
├── 5_short_term_memory/   # Memory Saver, trimming & summarization
├── 6_event_streaming/     # Event Streaming v3 protocol & projections
├── 7_streaming/           # Graph Streaming modes (updates, values, messages, debug)
├── 8_structured_output/   # Validated schemas, strategies & retry mechanisms
├── 9_middleware/          # Built-in middleware, HITL, custom hooks & guardrails
├── 10_guardrails/         # PIIMiddleware, deterministic & model-based guardrails
├── 11_runtime/            # Runtime context, ToolRuntime, execution_info & server_info
├── 12_context_engineering/ # Model/Tool/Life-cycle context × State/Store/Runtime
├── 13_mcp/                # Model Context Protocol — servers, tools, resources, interceptors
├── 14_human_in_the_loop/  # HITL middleware — approve, edit, reject, respond decisions
├── 15_multi_agent/        # Subagents, Handoffs, Skills, Router patterns
├── 16_retrieval/          # RAG — knowledge base, 2-step, agentic, hybrid
├── 17_longterm_memory/    # Store — semantic/episodic/procedural memory across sessions
└── 18_frontend/           # Generative UI — streaming, tool cards, HITL, time travel

---

📂 Modules

🤖 1_agents — Agent Patterns

File	Description
01_basic_agent.py	Basic agent setup with tool use
02_agent_with_memory.py	Agent with conversation memory
03_structured_output.py	Structured/typed output from agents
04_streaming.py	Token streaming with agents
05_middleware.py	Middleware and error handling
06_context_and_runtime.py	Context injection at runtime
07_full_agent_showcase.py	End-to-end agent showcase
agents_overview.py	Complete agents overview in one file

🧠 2_models — LLM Models

File	Description
01_init_and_invoke.py	Model initialization and invocation
02_parameters.py	Configuring model parameters
03_structured_output.py	Model structured outputs via Pydantic/TypedDict
04_multimodality.py	Multimodal inputs & image analysis
05_tool_calling.py	Tool binding and manual loops
models_overview.py	Complete models overview in one file

💬 3_messages — Messages & History

File	Description
01_message_types.py	HumanMessage, AIMessage, SystemMessage
02_conversation_history.py	Managing conversation history
03_message_content_and_streaming.py	Content blocks and streaming
04_tool_message_loop.py	Tool call / ToolMessage cycles
05_multimodal_messages.py	Images and multimodal inputs
06_full_messages_showcase.py	Full messages showcase
messages_overview.py	Complete messages overview in one file

🔧 4_tools — Tools

File	Description
01_basic_tools.py	Defining and using basic tools
02_advanced_schemas.py	Advanced Pydantic schemas for tools
03_tool_runtime_context.py	Injecting runtime context into tools
04_tool_return_values.py	Handling tool return values
05_dynamic_tool_selection.py	Dynamic tool selection
06_full_tools_showcase.py	Full tools showcase
tools_overview.py	Complete tools overview in one file

🧠 5_short_term_memory — Short-Term Memory

File	Description
01_checkpointer_basics.py	Thread-based conversation state
02_custom_state.py	Persistent custom variables
03_trim_and_delete_messages.py	Trimming historical messages
04_summarization_and_dynamic_prompt.py	Compressing history & dynamic prompts
05_full_memory_showcase.py	Full memory personal assistant
short_term_memory_overview.py	Complete short-term memory overview in one file

🌊 6_event_streaming — Event Streaming (v3 Projections)

File	Description
01_stream_events_basics.py	Modern token projections
02_tool_call_streaming.py	Argument and execution tracking
03_state_and_values_streaming.py	Value snapshot tracking
04_subagents_and_multiple_projections.py	Child agent event routing
05_full_streaming_showcase.py	Production streaming chatbot
event_streaming_overview.py	Complete event streaming overview in one file

📡 7_streaming — Graph Streaming (Step-Level API)

File	Description
01_updates_mode.py	Step-level incremental state updates
02_values_mode.py	Step-level full state snapshots
03_messages_mode.py	Low-level model token streaming
04_debug_mode.py	Detailed internal tracing events
05_full_streaming_showcase.py	Interactive terminal streaming showcase

🧱 8_structured_output — Structured Output

File	Description
01_auto_strategy.py	Auto strategy selection (Pydantic, Dataclasses, TypedDict)
02_provider_strategy.py	Native ProviderStrategy with strict validation & JSON Schema
03_tool_strategy_basics.py	ToolStrategy for tool-calling models
04_custom_tool_message.py	Custom tool message representations in chat logs
05_error_handling_and_retries.py	Automatic Pydantic retries and custom handlers
06_full_structured_output_showcase.py	Support ticket classification and Order lookup showcase
structured_output_overview.py	Complete structured output overview in one file

🛡️ 9_middleware — Middleware

File	Description
01_built_in_middleware.py	SummarizationMiddleware, ModelCallLimitMiddleware, ToolCallLimitMiddleware, ToolRetryMiddleware
02_human_in_the_loop.py	HITL approve / edit / reject workflows with per-tool policies
03_custom_middleware.py	BaseMiddleware hooks: logging, timing, cost estimation, tool-specific middleware
04_pii_detection_and_guardrails.py	PIIDetectionMiddleware, content guardrails, input validation, output sanitization
05_agent_loop_middleware.py	Loop observer, rate limiter, early exit, hook firing order
06_full_middleware_showcase.py	Customer support triage agent with 6 stacked middleware layers
middleware_overview.py	Complete middleware overview in one file

🛑 10_guardrails — Guardrails

File	Description
01_pii_middleware.py	PIIMiddleware with redact / mask / hash / block strategies and custom regex detector
02_deterministic_guardrails.py	before_agent class & decorator hooks, keyword filter, rate limiter, input length validation
03_model_based_guardrails.py	after_agent class & decorator hooks, LLM-as-judge safety, topic relevance, quality gate
04_hitl_as_guardrail.py	HumanInTheLoopMiddleware for financial/database/email ops, full approve/edit/reject lifecycle
05_full_guardrails_showcase.py	Financial advisory agent with 7-layer guardrail stack across 5 real-world scenarios
guardrails_overview.py	Complete guardrails overview in one file

⚡ 11_runtime — Runtime Context

File	Description
01_context_schema.py	context_schema dataclass, injecting context= at invoke time, @dynamic_prompt from context
02_tool_runtime.py	ToolRuntime[Context] in tools, runtime.context, runtime.store, runtime.writer
03_runtime_in_middleware.py	@before_model / @after_model with Runtime[Context], execution_info, server_info, RBAC
04_execution_and_server_info.py	thread_id, run_id, attempt, retry detection, audit trail, production auth gate
05_full_runtime_showcase.py	Multi-tenant CRM agent with context injection, RBAC, dynamic prompts, store, and audit logging
runtime_overview.py	Complete runtime overview in one file

🧠 12_context_engineering — Context Engineering

File	Description
01_model_context_system_prompt.py	@dynamic_prompt from State, Store, and Runtime Context; combined multi-source prompt
02_model_context_messages.py	@wrap_model_call transient injection: file context, writing style, compliance rules
03_model_context_tools_and_model.py	Dynamic tool filtering (RBAC, feature flags, auth) and dynamic model switching (cost tier)
04_model_context_response_format.py	Dynamic Pydantic schema selection by conversation stage, verbosity pref, and role
05_tool_context_reads_writes.py	Tool reads from state/store/context; writes via Command and store.put()
06_lifecycle_context.py	SummarizationMiddleware, persistent before_model state updates, audit logging
07_full_context_engineering_showcase.py	Smart Legal Research Agent — all 3 context types × all 3 data sources × 3 scenarios
context_engineering_overview.py	Complete context engineering overview in one file

🔌 13_mcp — Model Context Protocol

File	Description
servers/math_server.py	FastMCP stdio server — add, subtract, multiply, divide
servers/weather_server.py	FastMCP HTTP server — get_weather, get_forecast, get_air_quality
servers/rich_server.py	FastMCP HTTP server with structured content, resources, prompts, progress
01_mcp_basics.py	MultiServerMCPClient, get_tools(), stateless vs stateful sessions
02_mcp_transports.py	stdio vs HTTP transports, custom headers, multi-server config
03_mcp_tools_resources_prompts.py	Tools, structured content, Resources (Blob), Prompts (messages)
04_mcp_interceptors.py	Logging, runtime context, store, state auth, request.override(), retry, composition
05_mcp_callbacks.py	on_progress, on_logging_message, on_elicitation (accept/decline/cancel)
06_full_mcp_showcase.py	Smart Data Assistant — interceptors + callbacks + middleware + multi-turn memory
mcp_overview.py	Complete MCP overview in one file

🧑‍⚖️ 14_human_in_the_loop — Human-in-the-Loop (HITL)

File	Description
01_hitl_basics.py	HumanInTheLoopMiddleware, interrupt_on, version="v2", approve, reject, auto-approve
02_decision_types.py	All 4 types: approve, edit (args + tool swap), reject, respond
03_multiple_decisions.py	Multiple simultaneous interrupts, mixed decisions, sequential rounds
04_hitl_streaming.py	stream() with stream_mode=["updates","messages"], interrupt detection in stream
05_full_hitl_showcase.py	Secure Financial Operations Agent — risk-tiered policies, all 4 decisions, streaming
hitl_overview.py	Complete HITL overview in one file

🤖 15_multi_agent — Multi-Agent Systems

File	Description
01_subagents.py	Tool-per-agent wrapping, ToolRuntime, Command+InjectedToolCallId, parallel calls
02_subagents_dispatch.py	Single dispatch task tool, enum constraint, tool-based discovery, async background jobs
03_handoffs.py	current_step state, Command transitions, @dynamic_prompt, @wrap_model_call
04_skills.py	On-demand skill loading, stateful reuse across turns, Store cache
05_router.py	LLM structured output routing, keyword routing, fan-out+merge, async parallel, nested
06_full_multi_agent_showcase.py	Enterprise Assistant — Router dispatches to Skills / Subagents / Handoffs
multi_agent_overview.py	Complete multi-agent overview in one file

🔍 16_retrieval — Retrieval-Augmented Generation (RAG)

File	Description
01_knowledge_base.py	Document loaders, text splitters, embeddings, vector stores, retrievers
02_two_step_rag.py	2-Step RAG — fixed retrieve-then-generate, LCEL chains, multi-query
03_agentic_rag.py	Agentic RAG — agent with retrieval tools, fetch_url, multi-source
04_hybrid_rag.py	Hybrid RAG — query enhancement, retrieval validation, answer quality check
05_full_retrieval_showcase.py	Smart Q&A Assistant — all three RAG architectures with routing
retrieval_overview.py	Complete retrieval overview in one file

🧠 17_longterm_memory — Long-Term Memory

File	Description
01_store_basics.py	InMemoryStore, IndexConfig, put/get/delete/search, namespace patterns, StoreValue
02_read_memory_in_tools.py	ToolRuntime[Context], context_schema, runtime.store.get(), multi-namespace reads, semantic search
03_write_memory_from_tools.py	TypedDict schemas, save_user_info, episodic events, procedural rules, delete
04_memory_types.py	Semantic, episodic, procedural memory types, extraction LLM, cross-memory synthesis
05_full_longterm_memory_showcase.py	Personal AI — learn in session 1, recall in session 2, multi-user isolation
longterm_memory_overview.py	Complete longterm memory overview in one file

🌐 18_frontend — Generative UI & Agent Frontends

File	Description
01_agent_server.py	Python backend — create_agent + FastAPI SSE streaming + LangGraph Dev server config
frontend/01_basic_chat.html	Token streaming chat with live tool call cards (demo + real server mode)
frontend/02_hitl_approval.html	HITL — agent pauses, human approves/rejects/edits before tool execution
frontend/03_full_chat_app.html	Full showcase — streaming, tools, HITL, structured output, markdown, time travel
frontend_overview.py	Complete frontend overview in one file

---

⚙️ Setup

Prerequisites

• Python 3.11+
• An API key for your chosen LLM provider (e.g. OpenAI, Anthropic, Google)

Installation

# Clone the repo
git clone https://github.com/vins13pattar/langchain-python.git
cd langchain-python

# Create and activate virtual environment
python -m venv .venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

Environment Variables

A .env.example file is provided with all supported keys. Copy it and fill in your values:

cp .env.example .env

Then edit .env with your actual API keys:

# OpenAI
OPENAI_API_KEY=your_openai_api_key_here

# Anthropic (Claude)
ANTHROPIC_API_KEY=your_anthropic_api_key_here

# Google Gemini
GOOGLE_API_KEY=your_google_api_key_here

# LangSmith (optional — for tracing & observability)
LANGCHAIN_TRACING_V2=true
LANGCHAIN_API_KEY=your_langsmith_api_key_here
LANGCHAIN_PROJECT=langchain-python

Note: .env is gitignored and will never be committed. Only .env.example (with placeholder values) is tracked in version control.

---

🚀 Running Examples

# Run any example directly
python 1_agents/01_basic_agent.py
python 2_models/01_init_and_invoke.py
python 3_messages/01_message_types.py
python 4_tools/01_basic_tools.py
python 9_middleware/01_built_in_middleware.py
python 10_guardrails/01_pii_middleware.py
python 11_runtime/01_context_schema.py
python 12_context_engineering/01_model_context_system_prompt.py
python 13_mcp/01_mcp_basics.py
python 14_human_in_the_loop/01_hitl_basics.py
python 15_multi_agent/01_subagents.py
python 16_retrieval/01_knowledge_base.py

---

🛠️ Tech Stack

Technology	Purpose
LangChain	LLM application framework
LangGraph	Stateful agent orchestration
OpenAI / Anthropic	LLM providers
Pydantic	Data validation & schemas
python-dotenv	Environment variable management

---

📄 License

MIT License — feel free to use and adapt these examples for your own projects.

---

🙋 Author

Vinod — github.com/vins13pattar