An open LLMOps platform
In May 2023, LangGenius, a company founded by Luyu Zhang with offices in Suzhou and San Francisco, released Dify, an open source LLMOps platform designed for building and operating applications based on language models. Dify covers the entire lifecycle: from application development to production rollout, monitoring and iteration.
Visual studio and four app types
The core of Dify is a visual studio that allows building four types of applications: conversational chatbots, node-based workflows for deterministic orchestration, agents with access to external tools, and RAG (Retrieval-Augmented Generation) applications with integrated vector datasets.
The technology stack is based on Python for the backend and Next.js for the interface. The project integrates several models natively (OpenAI, Anthropic, Azure OpenAI, Hugging Face, local models via Ollama) and several vector databases (Weaviate, Qdrant, Milvus, PGVector).
Licence and restrictions
Dify adopts a variant of Apache 2.0 with an additional clause explicitly forbidding rebranding and offering it as a multi-tenant competing service. The licence is therefore not fully OSI-compliant: a point to carefully consider in commercial adoption decisions, especially for service providers reselling SaaS platforms. For internal corporate use and self-hosting on own infrastructure, the terms remain very permissive.
Adoption and deployment
By 2024, Dify surpassed 50,000 GitHub stars, becoming one of the most watched open LLMOps projects. It is available as a managed cloud SaaS by LangGenius or as a self-hosted deployment (Docker containers, Kubernetes Helm chart) for organisations that require full control over data and models used. The combination of a low-code visual studio and open architecture positions it as a concrete alternative to proprietary platforms in the segment.
Link: dify.ai