Dify

Dify is an open-source platform for building applications on top of large language models (LLMs). It describes itself as an open-source LLM app development platform, and its interface combines four things that developers would otherwise assemble from separate tools: an AI workflow editor, a retrieval-augmented generation (RAG) pipeline, agent capabilities and model management.^[1]

The four parts solve different problems. A workflow editor lets a developer lay out the steps of an application — call a model, branch on the result, call another service — as a visible graph instead of code. A RAG pipeline retrieves relevant passages from a collection of documents and puts them into the prompt, so the model answers from that material rather than from what it learned in training. Agent capabilities leave the choice of which tools to call, and in what order, to the model itself instead of a fixed sequence. Model management keeps the configuration for different model providers in one place, so an application can be pointed at another model without being rebuilt. Having all four behind one interface is the core of Dify's pitch: a team can prototype in the same environment it will later use to ship.^[1]

Where it runs

Dify can be used in three ways: on the managed Dify Cloud, inside a customer's own virtual private cloud (VPC), or fully self-hosted.^[1] The company presents managed hosting as a way to build and launch AI apps without managing infrastructure.^[2] The self-hosted and VPC options matter to organisations that need to keep data and model calls inside their own network.

Observability

The repository lists integrations with three observability tools: Opik, Langfuse and Arize Phoenix.^[1] LLM applications are hard to debug from their output alone: a single request may pass through retrieval, several model calls and tool calls, and the final answer says little about which step went wrong. Observability tools record the inputs, outputs and timing of each step so that the chain can be inspected.

Positioning for production

Dify's site positions the product for production-ready agentic workflows and says teams can go from prototype to production without rebuilding the stack.^[2] The site also cites Volvo Cars on using Dify to design and deploy natural language processing pipelines.^[2]

Attention

At the time of writing, the GitHub repository listed 160.6k stars and 25.4k forks.^[1] Star counts measure attention rather than production use, but numbers of that size make the project one of the more visible open-source entries in its category.

Where it sits

An open-source core with a managed cloud on top is a common shape for infrastructure software: the code is public and anyone can run it, while teams that would rather not operate it can use a hosted version. Dify offers both routes,^[1][2] which places it between stitching together separate tools and buying a closed platform.