Dify

With over 60,000 GitHub stars and more than one million deployed applications, Dify has become one of the most widely adopted open-source AI platforms in the world.

At the heart of Dify is a visual canvas where users can compose complex AI workflows by connecting nodes representing LLM calls, retrieval operations, conditional logic, data transformations, tool invocations, and API connections.

This drag-and-drop orchestration environment eliminates the need to write custom LLM orchestration code, while still providing the flexibility to build sophisticated multi-step AI systems that handle real-world production requirements like branching logic, error handling, and output formatting.

Dify supports seamless integration with hundreds of LLM providers, including OpenAI, Anthropic, Google, Mistral, Meta's Llama family, and any service exposing an OpenAI-compatible API.

Users can swap models within a workflow without restructuring their application logic, enabling model evaluation and cost optimization without rebuilding from scratch. For teams running self-hosted models, Dify's architecture supports local inference providers alongside commercial API-based models in the same workflow.

The platform's RAG (Retrieval-Augmented Generation) pipeline covers the full document ingestion and retrieval lifecycleparsing PDFs, PowerPoints, and other formats; chunking and embedding content; indexing in vector databases; and retrieving relevant context at query time.

This out-of-box RAG capability allows teams to build knowledge-grounded AI applications in hours rather than weeks. An Agent capability layer adds ReAct and function-calling-based reasoning with access to over 50 built-in tools including Google Search, DALL-E, Stable Diffusion, and WolframAlpha.