langgraph-editor
langgraph-editor copied to clipboard
This is a visual editor for langgraph workflow. It helps to quickly design and debug the workflow from scratch.
langgraph editor
This tool is designed to speed up your AI Agent
development. It help to visually edit langgraph workflow. You can open the notebook in google's colab.
Below is the interface of the tool. It is based on litegraph.js as its workflow engine (the same as ComfyUI). You can add, remove, and layout different nodes as you wish, get the workflow as you wish. As it is visualized and editable, it is easier for understanding and modifying.
Installation
Local machine
pip3 install -q tornado
git clone https://github.com/Erickrus/langgraph-editor
Google colab
ngrok
or other NAT Traversal tool is required (e.g. cpolar
) is used to display the iframe.
When using ngrok
, you need input NGROK_AUTH_KEY
and NGROK_BEARER_API_KEY
.
Run
Local machine
open the url http://localhost:8082/static/index.html
in your browser.
Google colab
In google colab, run through the cells, and you will see the interface.

Nodes
3 types of nodes are added, including:
-
LangGraphNode
: normal workflow node -
LangGraphConditionalNode
: work as the conditional edge -
LangGraphEndNode
:END
Compile the workflow to python code
When finish editing the workflow, click Compile
menuitem to compile the workflow as a python script. The file will be saved as /content/output.py
by default. Following is the output python script
from langgraph.graph import END, StateGraph
workflow = StateGraph(GraphState)
# Define the nodes
workflow.add_node("query", query) # query
workflow.add_node("web_search", web_search) # web_search
# Build graph
workflow.add_conditional_edge(
"query",
decide,
{
"end": END,
"web_search": "web_search",
})
workflow.set_entry_point("query")
workflow.add_edge("web_search", END)
# Compile
app = workflow.compile()
Load and Save workflow
The workflow can be loaded or saved locally from or to /content/workflow.json
.