Langflow

Langflow gives you a node canvas in the browser where prompts, model calls, retrievers, vector stores, agents and tools become boxes you wire together, then run step by step in a built-in playground. Every finished flow is reachable over a REST endpoint and can also be published through an MCP server, so other apps and MCP clients can call it as a tool; flows can be saved to JSON and loaded from Python too. Because each component is plain Python source, you can open one, change its behavior, or write your own. Integrations span the major hosted LLM providers, local runtimes such as Ollama, vector stores including Qdrant, Milvus and Weaviate, and tracing backends like LangSmith and Langfuse. Install the langflow package with uv on Python 3.10 through 3.14 and start the server on port 7860, pull the official Docker image, or use the desktop build with its own bundled runtime. The repository is MIT licensed, so company self-hosting is unrestricted; a free hosted cloud account also exists. Best for builders who want to sketch agent and RAG pipelines visually and still ship them as callable services.