Dify — open-source platform for building LLM apps and agents
langgenius/difyDify is an open-source platform for building LLM applications and agentic workflows. It combines a visual builder, retrieval (RAG) and model management so teams can ship AI features faster.
A strong pick if you want to build and operate LLM apps without wiring everything from scratch. Self-hostable, but you take on running it; the license is non-standard, so check terms.
The problem it solves
Building a production LLM app means stitching together prompts, retrieval, model providers and orchestration. Doing that from scratch for every project is slow and hard to maintain.
What is it?
Dify provides a platform layer over LLMs: a visual workflow/agent builder, built-in retrieval-augmented generation, model provider management and app deployment. You assemble AI features visually and via APIs rather than coding all the plumbing.
Why it's getting attention
With roughly 148k GitHub stars it is one of the most popular open-source LLM-app platforms. Interest tracks the shift from raw model APIs toward tooling that helps teams build and operate agentic apps.
Key features
- ✓Visual workflow / agent builder
- ✓Built-in retrieval-augmented generation (RAG)
- ✓Model provider management across LLMs
- ✓App deployment and APIs
- ✓Self-hostable, open source
Best use cases
- •Build an internal AI assistant over your own documents (RAG)
- •Prototype and ship agentic workflows without heavy coding
- •Standardize LLM app building across a team
- •Self-host an AI app platform for data control
How to install / try
Dify is commonly self-hosted (for example via Docker) or used via its cloud. See the repository for the current self-hosting setup and configuration.
How to use
Create an app, connect a model provider, add knowledge/RAG sources, and design the workflow or agent in the visual builder; then expose it via API. Refer to the docs for current features.
Strengths
- ✓Speeds up building and operating LLM apps
- ✓Visual builder lowers the barrier for teams
- ✓Self-hosting gives data control
- ✓Large, active open-source community
Limitations & risks
- △Self-hosting means you run and maintain the platform
- △A platform adds its own concepts to learn versus a thin library
- △License is non-standard on GitHub — review terms before commercial use
- △As a fast-moving project, features and APIs can change
Alternatives
Who should try it — and who should skip
Try it if you want to build and operate LLM apps/agents with less plumbing and value self-hosting. Skip it if you prefer a thin code library or need a fully managed service with guaranteed SLAs.
Frequently asked questions
Yes. Dify is open source and commonly self-hosted (for example with Docker); see the repository for setup. A cloud option also exists.
Yes. Retrieval-augmented generation over your own knowledge sources is a built-in part of the platform.
Chatbots, AI assistants, and agentic workflows over your own data, using its visual builder and built-in RAG.