LangGraph
LangGraph is a framework for building controllable agentic workflows. You may self-host or deploy to the cloud.
LangGraph Studio is an agent IDE that enables visualization, interaction, and debugging of complex agentic applications.
Install
Install LangGraph with Poetry.
poetry add langgraph
Install LangGraph Studio.
poetry add --group dev "langgraph-cli[inmem]"
Run
poetry run langgraph dev
Concepts
- Control flow
- Fixed - User-defined chain
- Agent - LLM-defined chain
- Runnable protocol
- invoke
- State is a dictionary.
- Nodes are functions.
LangGraph Studio via Docker
LangGraph Studio now runs on the web instead of Docker. It creates a tunnel to your locally hosted graph.
Docker up
A LangGraph project can be started from the command line.
langgraph up
This spins up a Docker container. You can configure custom dependencies and commands. For example:
{
"dockerfile_lines": [
"RUN mkdir -p /input/upload",
"COPY ./input/upload/ /input/upload/"
],
"dependencies": ["./src/sales_ai_graph_pipeline"],
"graphs": {
"pipeline_agent": "./src/sales_ai_graph_pipeline/agents.py:graph"
},
"env": ".env",
"python_version": "3.12"
}
Read more:
Dockerfile
You can also generate the Dockerfile locally.
langgraph dockerfile output/Dockerfile
Where output/Dockerfile is the file path to save the Dockerfile.
Parsing errors
Error: ValueError('An output parsing error occurred. In order to pass this error back to the agent and have it try again, pass handle_parsing_errors=True to the AgentExecutor. This is the error: Could not parse tool input: {'arguments': 'query: "len(df)"', 'name': 'python_repl_ast'} because the arguments is not valid JSON.\nFor troubleshooting, visit: https://python.langchain.com/docs/troubleshooting/errors/OUTPUT_PARSING_FAILURE')
If you see an error similar to the one above, try passing handle_parsing_errors=True to the AgentExecutor.
Resources
- Intro to LangGraph
- LangGraph explained in simple terms | Medium, e.g., agents and their "disposable chains"
- LangGraph Cloud Quick Start | GitHub
- Speed up my agent | LangChain