What it does
Dify is a platform for building and running LLM applications. A visual workflow editor lets you lay out how an app retrieves data, makes decisions, calls tools and asks for human input, so the logic stays visible and testable. You can build agents by chatting or by configuring them manually, then use them as standalone apps, embed them in a website, or reuse them as nodes inside workflows. A knowledge pipeline extracts, cleans, chunks and indexes files, websites and documents, and lets you test retrieval before connecting it to an agent. A marketplace supplies model providers, tools, data sources and MCP integrations. Finished apps can be published as hosted web apps, API endpoints, embeds or MCP-compatible tools, with logs, feedback, annotations, latency and usage data for monitoring. There are three ways to run it: a managed cloud, an enterprise edition for private or VPC deployment, and a community edition you self-host with Docker. The site describes the community licence as an Apache-2.0 derivative and source-available.
Capabilities
- Visual workflow studio for agentic workflows
- Agent builder with skills, tools and knowledge
- Knowledge pipeline for RAG with retrieval testing
- Marketplace of model providers, tools and MCP integrations
- Publish as a web app, API endpoint, embed or MCP tool
- Logs, annotations, latency and usage monitoring
When to use it
- You want to prototype and ship an LLM app or agent without hand-wiring each integration
- You need a RAG knowledge base where retrieval can be tested before use
- You want to self-host the platform on your own infrastructure or in a VPC
- You want non-developers on a team to build and reuse approved AI workflows
