Dify

Dify is a visual builder for teams creating AI agents and applications that use their own data and tools. Its canvas makes application logic visible so teams can inspect and test how a task proceeds.

• Visual workflows: Connect prompts, information retrieval, decisions, tool calls and human input into an application workflow.
• Agents with tools: Give agents skills, Marketplace tools, MCP connections or custom APIs, then use them as chat apps or steps within a workflow.
• Knowledge pipelines: Prepare files, websites, online documents and drives for retrieval, with support for extracting text from PDFs and presentations and testing search results.
• Model choice: Use GPT, Mistral, Llama3 or OpenAI-compatible models, including models served through self-hosted inference systems.
• Publishing and monitoring: Expose apps as web experiences, APIs, website embeds or MCP-compatible tools, and inspect logs, feedback and performance data.

The source-available Community Edition runs on your own server with Docker and uses a custom Apache-2.0-derivative license. The server requires at least two CPU cores and 4 GiB of RAM. Dify also offers a separately hosted cloud service and deployment in a VPC. Self-hosting the builder doesn't make a cloud model local: requests to an external model provider go to that provider.