Foundation: Introduction to Deep Agents
by LangChain Academy
Build long-running, model-neutral agents with LangChain's open-source Deep Agents harness: sandboxes, context management, subagents and deployment.
Overview
Introduction to Deep Agents is a free foundation course from LangChain Academy. It frames an agent as a model plus a harness: the software layer that gives a model execution environments, context management, delegation and human steering. It teaches that harness through Deep Agents, LangChain's open-source, model-neutral library built on LangGraph (MIT-licensed, about 29.5k GitHub stars). The 43-lesson course starts with setup lessons for Python and TypeScript and then runs five modules. Module 1, Building a Deep Agent, covers running an agent, choosing models, the system prompt, tools, MCP, messages, threads and checkpointers, and human-in-the-loop approval, and ends with a skills test. Module 2, Execution Environment, covers filesystem backends, sandboxes, the local shell and a code interpreter. Module 3, Context Management, covers summarization and offloading, skills, and memory. Module 4, Delegation, builds a subagent team and dynamic subagents. Module 5 brings everything together: local deployment, a sales-assistant project, asynchronous subagents, and an advanced sales assistant that uses a modified agent-chat-ui. The companion repo, langchain-ai/lca-deepagents, uses uv and Python 3.11–3.14. It needs a LangSmith API key plus an Anthropic, OpenAI or Google key. Tavily is optional for the web-search labs in Modules 4 and 5, and OpenRouter is optional for free open-source models. As of September 2026 the repo still marks its TypeScript materials "coming soon", so Python is the complete track.
At a Glance
- Topic
- Agentic
- Level
- Intermediate
- Format
- Course
- Cost
- Free
- Duration
- ~30 minutes of video across 43 lessons, plus self-paced Python labs in the companion repo
- Provider
- LangChain Academy
- Hands-on
- Yes — code/exercises
- Certificate
- None
What You’ll Learn
- ✓Run a Deep Agent and configure its model, system prompt and tools
- ✓Connect MCP servers and persist conversations with messages, threads and checkpointers
- ✓Add human-in-the-loop checks so tool calls can be approved, edited or rejected
- ✓Choose between filesystem backends, sandboxes, a local shell and an interpreter for execution
- ✓Keep long-running agents inside the context window using summarization and context offloading
- ✓Package reusable behaviour as skills and give agents persistent memory across sessions
- ✓Delegate work to a subagent team, including dynamic and asynchronous subagents
- ✓Deploy a sales-assistant agent locally and trace its runs in LangSmith
Highlights
- •Official course from the team that maintains Deep Agents (about 29.5k GitHub stars, MIT license)
- •Organized around the harness capabilities (execution environment, context management, delegation and human steering) rather than around a single API
- •Model-neutral labs run on Anthropic, OpenAI or Google models, with OpenRouter as an optional route to free open models
- •The capstone builds a sales assistant twice: a basic version, then an advanced one with asynchronous subagents and a custom chat UI
- •Light on video (about 30 minutes) and heavy on runnable code in a public companion repository
Who It’s For
Best For
- ✓Python developers moving from single-call LLM apps to long-running research or coding agents
- ✓Teams already using LangChain or LangGraph who are evaluating Deep Agents as their agent harness
- ✓Engineers who want hands-on practice with sandboxes, subagents and context offloading
Prerequisites
- •Working Python and comfort installing dependencies with uv from the command line
- •A LangSmith API key plus an Anthropic, OpenAI or Google API key (the model provider bills your usage)
- •Basic familiarity with LLM tool calling; LangGraph experience helps but is not listed as required
FAQ
What is Foundation: Introduction to Deep Agents?
A free LangChain Academy course, launched in July 2026, on building long-running agents with Deep Agents, LangChain's open-source agent harness. It is for Python developers who already call LLM APIs and want to build research- and coding-style agents with filesystems, sandboxes, skills, memory, subagents and human approval. It ends with a deployed sales-assistant agent.
Is Foundation: Introduction to Deep Agents free?
Foundation: Introduction to Deep Agents is free to access.
What level is Foundation: Introduction to Deep Agents for?
Foundation: Introduction to Deep Agents is aimed at a intermediate audience. Recommended background: Working Python and comfort installing dependencies with uv from the command line, A LangSmith API key plus an Anthropic, OpenAI or Google API key (the model provider bills your usage), Basic familiarity with LLM tool calling; LangGraph experience helps but is not listed as required.
How long does Foundation: Introduction to Deep Agents take?
Expect roughly ~30 minutes of video across 43 lessons, plus self-paced Python labs in the companion repo. Most learners work through it at their own pace.
What will I learn from Foundation: Introduction to Deep Agents?
You'll learn: Run a Deep Agent and configure its model, system prompt and tools; Connect MCP servers and persist conversations with messages, threads and checkpointers; Add human-in-the-loop checks so tool calls can be approved, edited or rejected; Choose between filesystem backends, sandboxes, a local shell and an interpreter for execution; Keep long-running agents inside the context window using summarization and context offloading; Package reusable behaviour as skills and give agents persistent memory across sessions; Delegate work to a subagent team, including dynamic and asynchronous subagents; Deploy a sales-assistant agent locally and trace its runs in LangSmith.
Topics
Sources
This page was written from 5 sources, 4 on domains other than academy.langchain.com.