Dify

Dify is a platform for building LLM applications. A visual canvas lets you compose workflows from model calls, retrieval steps, tools, and logic, and the same platform covers prompt authoring, retrieval-augmented generation (RAG) pipelines, agents, model management, and monitoring. Every capability is also available through an API, so a Dify app can serve as a backend for your own software.

Model support spans hundreds of proprietary and open-source LLMs from many inference providers and self-hosted setups, including GPT, Mistral, Llama 3, and any OpenAI-API-compatible model. Observability integrates with Opik, Langfuse, and Arize Phoenix. Dify runs self-hosted from a Docker Compose stack (minimum 2 CPU cores and 4 GiB RAM), as the hosted Dify Cloud, or as an enterprise edition.

Features

• Workflow canvas: build and test AI workflows visually, combining models, retrieval, tools, and control flow
• Model support: hundreds of LLMs from dozens of providers plus self-hosted and OpenAI-compatible endpoints
• Prompt IDE: author prompts, compare model outputs, and add features such as text-to-speech to chat apps
• RAG pipeline: document ingestion through retrieval, with text extraction from PDFs, PPTs, and other formats
• Agents: function-calling or ReAct agents with more than 50 built-in tools, including Google Search, DALL-E, Stable Diffusion, and WolframAlpha, plus custom tools
• LLMOps: log and performance monitoring over time, with annotation to improve prompts, datasets, and models
• Backend-as-a-service: APIs for every offering so apps integrate into your own business logic
• Observability integrations: Opik, Langfuse, and Arize Phoenix