Dify is a visual platform for building AI agent applications. Instead of writing code for every step, the team assembles the flow in an editor, connects models and tools and publishes the result as an app or API.
What it is for
- Creating agents and assistants with your own knowledge base (RAG)
- Assembling agentic flows with decision, code, HTTP and tool nodes
- Publishing the result as a web app, API or tool for another system
- Standardising agent building across a team, with prompt versioning
How it works
- The editor mixes deterministic nodes and agent nodes, with variables and conditions
- The knowledge base is fed with documents and retrieved when answering
- Plugins and the MCP protocol connect external services and custom tools
- It includes run observability, prompt testing and per-app logs
- The same core runs on managed cloud or in the customer’s infrastructure
Availability
- Public code under a modified Apache 2.0 licence, with restrictions on offering it as a multi-tenant service and on removing the brand; the enterprise edition is commercial
- Self-hosted install via Docker Compose or Kubernetes
Points to note
- The licence is not OSI-approved: review the terms before reselling the service
- Very large flows get hard to maintain without naming and versioning discipline
- Compared with code frameworks, it gives less fine control over the agent loop