langflow-ai/langflow

Langflow – Visual Builder for AI Agents & Workflows

What it is Langflow is an open‑source platform that lets you design, test, and deploy AI‑powered agents and data‑flow pipelines through a web‑based visual editor. It ships as a Python package (and a desktop app) and can also be run in Docker. Under the hood it connects to any large language model (LLM), vector store, or tool you plug in, turning the visual flow into a callable API or an "MCP" server.

Key capabilities

Feature	What you get
Visual builder	Drag‑and‑drop nodes that represent LLM calls, retrieval, tool usage, etc.; you can run the flow step‑by‑step in a playground.
Code access	Each node is a Python component you can edit, so custom logic is possible without leaving the UI.
Multi‑agent orchestration	Build pipelines that coordinate several agents, manage conversations, and perform retrieval‑augmented generation.
Deployable API	One click turns the flow into a REST endpoint; you can also export the flow as JSON for direct use in Python code.
MCP server	A specialised server mode that lets other "MCP clients" consume the flow as a tool.
Observability hooks	Built‑in integrations with LangSmith, LangFuse and similar services for tracing and monitoring.
Enterprise‑ready	Supports authentication via environment variables, role‑based super‑user, and can be hardened for production.
Cross‑platform desktop	Pre‑bundled binaries for Windows/macOS that include all dependencies, avoiding manual Python setup.

How you run it

1. Python install – uv pip install langflow -U (requires Python 3.10‑3.14). Set LANGFLOW_SUPERUSER_PASSWORD and start with uv run langflow run. The UI appears at http://127.0.0.1:7860.
2. Docker – docker run -p 7860:7860 -e LANGFLOW_SUPERUSER_PASSWORD … langflowai/langflow:latest.
3. From source – make run_cli after cloning, for contributors.

Typical use‑cases

• Rapid prototyping of RAG pipelines: connect an LLM → vector DB → prompt template → output.
• Building chat‑bots that switch between multiple specialized agents.
• Turning a complex workflow into a micro‑service that other applications can call via HTTP.
• Experimenting with new AI tools (e.g., function‑calling, tool‑use) without writing boilerplate code.

Who it’s for

• Developers who want a UI to iterate on LLM‑centric pipelines before codifying them.
• Product teams needing a low‑code way to expose AI capabilities as APIs.
• Researchers who want to visualise and share reproducible agent setups.

Community & support The project is MIT‑licensed, has a Discord, Twitter, YouTube channel, and a growing contributor base (see the star‑history chart). Documentation covers installation, deployment, and security.

---

All details above are taken directly from the repository’s README.