This guide will walk you through how to use Studio to visualize, interact, and debug your agent locally.
Studio is our free-to-use, powerful agent IDE that integrates with LangSmith to enable tracing, evaluation, and prompt engineering. See exactly how your agent thinks, trace every decision, and ship smarter, more reliable agents.
Prerequisites
Before you begin, ensure you have the following:
Setup local Agent server
1. Install the LangGraph CLI
# Python >= 3.11 is required.
pip install --upgrade "langgraph-cli[inmem]"
2. Prepare your agent
We’ll use the following simple agent as an example:
from langchain.agents import create_agent
def send_email(to: str, subject: str, body: str):
"""Send an email"""
email = {
"to": to,
"subject": subject,
"body": body
}
# ... email sending logic
return f"Email sent to {to}"
agent = create_agent(
"gpt-4o",
tools=[send_email],
system_prompt="You are an email assistant. Always use the send_email tool.",
)
3. Environment variables
Create a .env file in the root of your project and fill in the necessary API keys. We’ll need to set the LANGSMITH_API_KEY environment variable to the API key you get from LangSmith.
Be sure not to commit your .env to version control systems such as Git!
LANGSMITH_API_KEY=lsv2...
4. Create a LangGraph config file
Inside your app’s directory, create a configuration file langgraph.json:
{
"dependencies": ["."],
"graphs": {
"agent": "./src/agent.py:agent"
},
"env": ".env"
}
create_agent automatically returns a compiled LangGraph graph that we can pass to the graphs key in our configuration file.
So far, our project structure looks like this:
my-app/
├── src
│ └── agent.py
├── .env
└── langgraph.json
5. Install dependencies
In the root of your new LangGraph app, install the dependencies:
6. View your agent in Studio
Start your Agent server:
Safari blocks localhost connections to Studio. To work around this, run the above command with --tunnel to access Studio via a secure tunnel.
http://127.0.0.1:2024) and the Studio UI https://smith.langchain.com/studio/?baseUrl=http://127.0.0.1:2024:
Studio makes each step of your agent easily observable. Replay any input and inspect the exact prompt, tool arguments, return values, and token/latency metrics. If a tool throws an exception, Studio records it with surrounding state so you can spend less time debugging.
Keep your dev server running, edit prompts or tool signatures, and watch Studio hot-reload. Re-run the conversation thread from any step to verify behavior changes. See Manage threads for more details.
As your agent grows, the same view scales from a single-tool demo to multi-node graphs, keeping decisions legible and reproducible.
