Agent Inbox LangGraph Example

The repository contains a bare minimum code example to get started with the Agent Inbox with LangGraph.

Tip

Looking for the TypeScript version of this example repository? Click here

Getting Started

To get started, clone the repository:

git clone https://github.com/langchain-ai/agent-inbox-langgraph-example.git

cd agent-inbox-langgraph-example

Then, install the dependencies:

poetry install

Next, install the LangGraph CLI if not already installed. We're installing the in-memory version so we can run the LangGraph server without Docker.

pip install -U "langgraph-cli[inmem]"

After this, we can start the LangGraph server:

langgraph dev

This may take a second to start. Once the server is running, it should open a new browser tab to the LangGraph Studio through LangSmith. If this does not happen automatically, visit this URL: https://smith.langchain.com/studio/thread?baseUrl=http%3A%2F%2F127.0.0.1%3A2024

Now that our LangGraph server is running, we can start a new run in the Studio. To do this, simply enter any string into the Interrupt Response field, then click the Submit button. This will execute the graph, and interrupt on the human_node. Once the graph has interrupted, we can visit the Agent Inbox site to add your graph, and manage the interrupted thread.

Agent Inbox Setup

Visit dev.agentinbox.ai. If it's your first time visiting the site, you'll be prompted to add a new graph.

Enter the following fields into the form:

• Graph/Assistant ID: agent - this corresponds to the ID of the graph defined in the langgraph.json file, or the ID of an assistant tied to your graph.
• Deployment URL: http://127.0.0.1:2024 - this is the URL of the LangGraph server running locally.
• Name: My Agent - this is the optional name of the graph. You can set this to any value you want, or leave it empty and it'll be auto-assigned.

Click Submit and watch your graph appear in the sidebar. This should automatically fetch the interrupted threads, but if it does not, click on the sidebar item & refresh the page. Once you've done this, you should see a single interrupted item in the table:

Click on this row, and you'll be taken to the interrupted item page. From here, you can respond in any way you'd like:

• Accept
• Edit
• Respond
• Ignore

Once you take an action, the graph will resume the execution and end. The final state returned from the graph will be a string containing the type of action which was taken, and the value of the action args (unless ignore was chosen).

To view the result of the graph, go back to the LangGraph Studio and inspect the most recent thread results.

Go Deeper

If you'd like to go deeper, you can:

• Checkout the Agent Inbox docs, and codebase here: github.com/langchain-ai/agent-inbox
• See an open source Executive AI Assistant here: github.com/langchain-ai/executive-ai-assistant
• See an open source Social Media Agent here: github.com/langchain-ai/social-media-agent