The Complete LangChain, LangGraph, & LangSmith Course (2026)

Explore all LangChain components, including chat models, message types, prompt templates, document loaders, embeddings, vector stores, retrievers, tools, memory, and agents, with a focus on their retrieval augmented generation workflow.

Create your first LangChain project by setting up a Python virtual environment, installing LangChain and OpenAI packages, and securely configuring your OpenAI API key.

Explore message types in LangChain, including system messages, user messages, ai messages, and tool messages, and learn to construct prompts using message classes for chat models.

Explains prompt templates as blueprints for prompts, including base and chat templates, their structure and parameters, and demonstrates invoking them with variables in the LangChain workflow.

Learn to extract structured outputs from chat models using JSON schema or a Pydantic class, defining schemas to pull name, country, and optional age from responses.

Learn to stream responses from chat models, emitting tokens in real time and handling synchronous, asynchronous, and event-based streaming, with practical demos using LangChain tools.

Understand how tokens measure load and pricing for input tokens and output tokens, whether using a service API or a local LM, and track usage with LongChain.

Cache model responses to avoid repeated prompts and save token costs by using memory or sqlite caches with the long chain lm cache, wrappers, and globals.

Learn how prompt messages form the building blocks of LangChain prompts, detailing system, user, assistant, and tool messages, and how to create dynamic, parameterized prompts.

Learn to design chat prompt templates in LangChain by using a structured array of messages: system, user, and example pairs, with dynamic variables and a final prompt.

Explore properties and methods that manage prompt templates, including template strings, template format (f strings), input variables, optional variables, and validation options, plus chat templates with messages.

Discover template methods in LangChain, including invoke, format, format prompt, batch, prettyprint, and save, to produce, display, batch process, and persist prompt templates.

Explore few-shot prompting in long chain prompts with LangChain’s FewShotPromptTemplate, building an example formatter, concise examples, a suffix-driven question, and strategies to reduce non-determinism.

Discover how to use few-shot chat prompts in LangChain, LangGraph, and LangSmith workflows by building a few-shot chat message prompt template, an example formatter, and a final embedded prompt.

Learn to pass prompts to language models using few-shot prompts, invoked prompts, and prompt templates or formatted prompts, and build chains with the long chain expression language (lcl) and orchestration.

Explore how to use built-in runnables, including a chat model, to invoke, batch, and stream outputs in LangChain with hands-on Python demos.

Explore composing rendezvous in parallel using the runnable parallel class, feeding a single input to multiple tasks, with dictionary or keyword-argument styles, LCL usage, and practical caveats.

Visualize runnable chains in a LangChain workflow using Gandalf to render an ASCII graph. See parallel sentence and keyword chains and a runnable branch for conditional execution.

Extract and visualize every prompt in a runnable chain by calling the get prompts function, looping over results, enumerating prompts, and pretty printing each one.

Explore lifecycle events in long chain to monitor, log, and react to start, error, and end events using weight listeners with run metadata.

Learn to attach an onerror handler to a runnable chain, log and act on errors, and simulate value, network, and JSON errors to understand error details and retry strategies.

Explore configuring runnable and chain executions with runnable config, including run name, tags, metadata, max concurrency, recursion limit, and runtime values; implement per invocation or persistent with config and callbacks.

Learn to configure chain models at runtime by defining configurable fields, assigning IDs, and passing runtime configs to adjust max tokens and temperature on the fly.

Explore the react pattern for LangChain agents, detailing the thought, action, observation loop from task understanding to completion, using tools like calculators to solve math problems.

Explore the create_agent function in LangChain to build a single-purpose agent with LangGraph, tools, and structured outputs. Learn about model, system prompts, context, state, checkpoint, and middleware.

Learn to inject runtime user context into LangChain tools using a user context schema, enabling personalized balance checks, transactions, and greetings for Alice and Bob.

Explore LangChain, LangGraph, and LangSmith by using middlewares to gracefully catch and undo tool errors, demonstrated with a transfer money tool and an error-handling middleware.

Learn to produce structured outputs from agents by defining a Pydantic-based schema and using LangChain strategies to generate a structured financial response with summary, details, actions, warnings, and confidence.

Learn how the graph API models ai agent workflows as a stateful graph with state, nodes, and edges, including start and end nodes, compile and run to update state.

Define and manage a graph agent's state in Landgraaf by detailing state schema and reducers, update state through nodes, and compare dictionary, TypedDict, and Pydantic approaches.

Learn how state reducers in Landgraaf control updates to state variables, compare default vs custom reducers, and implement counting and list updates via Python or JavaScript.

Explore the line graph add_messages reducer to manage lm conversations by packaging ordered human, ai, and system messages into a history-enabled state, enabling conversational agents and ai workflows.

Learn to manage LLM messages in graphs with a messages state class, covering human, AI, tool, and system messages, plus accessing and updating the history.

Explore how line graph nodes serve as action takers, with start and end nodes, performing processing in runnable lambdas, and how edges, state updates, and runtime context shape workflows.

Master graph edges in the LangChain graph API. Use normal edges to move between nodes and conditional edges to route by state, from start to end, with an execution path.

Explore how the Send() API enables map-reduce in graph workflows by spinning up parallel node instances. Pass private state for each data item.

Explore augmented LMs gaining real-world capabilities by pairing with tools and a runtime to invoke external functions and services. See how structured outputs use schemas to deliver machine-readable data.

Augment an llm with two tools—the weather checker and a tip calculator—demonstrating tool calling and tool binding in a LangChain OpenAI workflow for the LangChain, LangGraph, and LangSmith course.

Learn to augment a language model to produce structured outputs by binding a schema, using JSON, for downstream services with e-commerce reviews.

Explore a prompt chaining workflow that links drafts, reviews, and sending emails into a sequential pipeline of LLM prompts and tools, with modular, verifiable steps.

Demonstrates a blog post content generation pipeline with four steps in a prompt chain: generate draft, fact check, improve with feedback, and format for publication, showcasing quality control.

Discover the parallelization workflow that runs independent LM tasks concurrently to speed up content generation and analysis, comparing outputs across different llms and criteria using a central aggregator.

Shows parallelization with a LangChain and LangGraph workflow that generates Instagram, Twitter, and LinkedIn posts in parallel, then aggregates them into a final social media content package.

Explore routing patterns that direct tasks to specialized nodes via a router. Apply routing to customer support, document processing, and medical triage with a router that directs to specialized nodes.

Explore the orchestrator/worker workflow, where the orchestrator splits a book into chapters, assigns them to summarization workers, and aggregates final results via parallel MapReduce-style processing.

Demonstrates an orchestrator–worker workflow that converts a topic into research sources, assigns workers to explore each source, and synthesizes findings into a final research report.

Master the evaluator/optimizer workflow where a generator drafts and an evaluator checks against criteria, iterating with measurable scores until criteria are met.

Apply an evaluator/optimizer workflow to create compelling product details for an e-commerce site, using OpenAI for generation and Gemini for unbiased evaluation to refine marketing content.

Explore how short term memory works in LangGraph, using threads, checkpoints, and snapshots to persist session data across runs, and compare memory saver, SQLite saver, and Postgres options.

Create a memory-enabled chat app by integrating checkpoints and a memory saver to persist session state across turns, using a thread id, graph-based flow, and runnable config.

Learn to retrieve state from any checkpoint in a LangChain graph using snapshots, including final state and full history, via a document analysis workflow.

Learn to replay a LangGraph workflow from a checkpoint by extracting the checkpoint ID from history, then run forward from analyze step using a replay config with the thread ID.

Learn to manually update the final graph state with a config and update state method, enabling manual overrides, human-in-the-loop reviews, and adjustments to quality score, approval, and issues found.

Switch memory storage from a memory saver to a production Postgres database, implement short-term memory checkpoints, run one-time migrations, and validate checkpoints in a production setup.

LangSmith provides end-to-end observability for LLM-powered apps built with LangChain and LangGraph, tracing calls, tokens, and latency. It also enables monitoring, evaluation, and prompt management to improve performance and debugging.