Deep Agents Workshop

A hands-on workshop covering Deep Agents, Deep Agents Deploy, and LangSmith evaluations.

Part	Topic	Duration
1	Deep Agents: Harness, Filesystem, Tools, Subagents, Memory, Middleware, HITL, Skills	~45 min
2	Deep Agents Deploy: Ship to LangSmith	~10 min
3	LangSmith: Tracing, Datasets & Evaluations	~20 min

Prerequisites

• Python 3.11+
• uv

Setup

1. Install dependencies:

uv sync

2. Copy .env.example to .env and fill in your API keys:

cp .env.example .env

You'll need:

Key	Where to get it
OPENAI_API_KEY	platform.openai.com
LANGSMITH_API_KEY	smith.langchain.com
TAVILY_API_KEY	tavily.com

3. Start the notebook:

uv run jupyter notebook notebooks/workshop.ipynb

Switching Models

The notebook uses init_chat_model() which supports any provider. Change the model in the setup cell:

# OpenAI (default)
model = init_chat_model("openai:gpt-4.1-mini")

# Anthropic
model = init_chat_model("anthropic:claude-sonnet-4-20250514")

# Azure OpenAI
model = init_chat_model("azure_openai:gpt-4.1-mini", azure_deployment="your-deployment")

# AWS Bedrock
model = init_chat_model("bedrock:anthropic.claude-sonnet-4-20250514-v1:0")

Some providers need an extra package:

uv add langchain-anthropic   # Anthropic
uv add langchain-aws          # AWS Bedrock

Deploy (Part 2)

Part 2 deploys your agent to LangSmith. This requires the langgraph CLI:

uv tool install deepagents-cli
uv tool install 'langgraph-cli[inmem]'

Your LANGSMITH_API_KEY must have deployment permissions (lsv2_sk_... service key, not a personal token).

Project Structure

workshop/
├── notebooks/
│   └── workshop.ipynb          # Main workshop notebook
├── agents/
│   └── deep_agent/             # Deployable agent (used in Part 2)
│       ├── agent.py
│       ├── AGENTS.md
│       ├── deepagents.toml
│       └── skills/
│           ├── linkedin-post/
│           └── twitter-post/
├── utils/
│   └── models.py               # Centralized model config
├── pyproject.toml
└── .env.example

Memory-backed deployment script

This repo also includes a one-file deployment script that creates a deep agent with a CompositeBackend and Context Hub-backed /memories/, then wires a LangSmith issues board.

uv run python agents/memory_backed_agent/deploy_memory_backed_agent.py \
  --agent-name my-agent

See agents/memory_backed_agent/README.md for details.

Deploy fails with 403 / permission denied Your LangSmith API key needs deployment permissions. Use a service key (lsv2_sk_...), not a personal access token (lsv2_pt_...).