Dify

Open-source LLM platform: visual workflows, RAG pipelines and agent orchestration you can self-host.

Dify is the open-source LLM application platform: a visual canvas for chaining models, retrieval, tools and logic into apps, prompt management with versions, a RAG pipeline with document ingestion, agent workflows, and an API that turns any canvas into a product backend.

Why founders use it: the distance from prompt to product is where LLM prototypes die, and Dify industrializes the middle, prompt iteration without deploys, retrieval wired to your documents, tool calls and branching as configuration, and usage logs per app. Self-hosting keeps customer data and model traffic inside your perimeter, which buys the compliance paragraph enterprise deals require.

A realistic setup: a customer-support assistant as a canvas, knowledge base ingestion of the help docs, retrieval top-k tuned, an LLM with a strict output schema, human-review logging on, and the whole thing exposed to the product through its API.

Trade-offs: visual canvases resist version control and review, complex flows outgrow the canvas and want code, and the platform's pace of change is fast. Compare with Langfuse for observability-first, Flowise for the graph metaphor, and plain frameworks when the team prefers code.