Langflow: Visual Flow Building With a Python Escape Hatch
Langflow builds language model pipelines as graphs, with the distinguishing property that the nodes are Python components you can open and modify.
The escape hatch matters
Every visual builder eventually hits something its node library does not do. Most leave you either abandoning the tool or writing an awkward workaround.
Langflow lets you write a custom component in Python and drop it into the same canvas. The visual layer is a convenience over real code rather than a wall around it, which considerably raises the ceiling.
What is in the box
Components for model providers, vector stores, document loaders, text splitters, memory, agents, and tools. Local models work through Ollama or any OpenAI-compatible endpoint.
Flows export as JSON, which makes them versionable and shareable in a way that a database-stored graph is not.
Deployment
Flows are exposed as API endpoints. There is a playground for testing interactively before wiring anything up.
Against the alternatives
Flowise is lighter and faster to a first result.
Langflow is the better choice if you expect to write custom components, since that path is designed rather than bolted on.
Dify is more of a complete application platform with user management and prompt versioning.
The three overlap substantially. Trying one for an afternoon tells you more than comparing feature lists.
Practical notes
It is a Python application with a substantial dependency tree, so the container image is the sensible path.
As with any of these, put authentication in front of it. A tool that executes arbitrary Python components and holds your API keys is not something to leave open.
License
Langflow is released under the MIT License.