Dify is a platform for building applications on top of large language models. Its visual interface brings AI workflows, retrieval-augmented generation (RAG), agent features, model management and observability together so teams can move from a prototype to production without stitching many tools together.

What you can do

• Design workflows on a visual canvas and test them as you build.
• Connect many models: integrate hundreds of proprietary and open-source LLMs from dozens of providers, including self-hosted ones.
• Add retrieval and agents to ground answers in your own documents.
• Monitor quality through observability integrations such as Langfuse, Opik and Arize Phoenix.

Self-hosting

The quickest route is Docker Compose; the project lists about 2 CPU cores and 4 GiB of RAM as the minimum. A managed cloud edition and an enterprise edition also exist.

License note: Dify uses a modified Apache 2.0 license with extra conditions. Running it as a multi-tenant service needs written permission, and you may not remove the logo or copyright in its frontend. Read the license before using it commercially.