Dify is an open-source platform for building LLM applications on a visual canvas. Workflows, RAG pipelines, agents, model management, and observability sit in one interface, and every app it builds is also exposed through an API, so Dify can act as the backend for your own business logic.
Agents run in a sandbox of their own where they run commands, install software, and handle files to complete open-ended tasks. You describe the agent you want and it builds itself, then you give it skills, connect tools from the Dify Marketplace, MCP servers, or your own APIs, and deploy it as a chat app or as a step in a workflow.
Dify runs as a hosted service (Dify Cloud), as a self-hosted Community Edition with Docker Compose, or as a private Enterprise deployment with SSO and role-based access control. Deployment targets include a cloud, a VPC, or your own servers.
Features
- Workflow canvas: build and test AI workflows visually, combining models, knowledge retrieval, agents, and tools
- Model support: hundreds of proprietary and open-source LLMs from dozens of inference providers and self-hosted solutions, including GPT, Mistral, Llama 3, and any OpenAI API-compatible model
- Prompt IDE: write prompts, compare model performance, and add features such as text-to-speech to chat apps
- RAG pipeline: covers document ingestion through retrieval, with text extraction from PDFs, PPTs, and other common formats
- Sandboxed agents: agents run commands, install software, and handle files, with skills and tools from the Marketplace, MCP servers, or custom APIs
- LLMOps: monitor application logs and performance, and improve prompts, datasets, and models from production data and annotations
- Observability integrations: Opik, Langfuse, and Arize Phoenix
- Backend-as-a-Service: every feature has a matching API for embedding Dify in other applications
