Dify

Dify is an open-source platform for building applications powered by large language models. It combines a visual workflow builder, a retrieval-augmented generation (RAG) pipeline, and agent tooling in one interface, so teams can go from a prototype to a production LLM app without stitching together separate frameworks. The project supports hundreds of proprietary and open-source models, including GPT, Mistral, Llama 3, and any OpenAI API-compatible endpoint.

Key features

• Visual workflow canvas: Build and test AI workflows using a drag-and-drop canvas that chains prompts, tools, and logic together.
• Broad model support: Connect to hundreds of proprietary and open-source LLMs from dozens of providers, or point Dify at any OpenAI API-compatible endpoint.
• Prompt IDE: Craft and compare prompts side by side, and add extras like text-to-speech to a chat-based app.
• RAG pipeline: Ingest and retrieve documents out of the box, including text extraction from PDFs, PowerPoint files, and other common formats.
• Agent capabilities: Define agents based on LLM function calling or ReAct, and equip them with 50+ built-in tools such as Google Search, DALL-E, Stable Diffusion, and WolframAlpha.
• LLMOps: Monitor application logs and performance over time, then refine prompts, datasets, and models using production data and annotations.
• Backend-as-a-service: Every feature ships with a matching API, so Dify functionality can be called directly from your own backend.

Self-hosting and deployment

Self-hosting Dify requires at least 2 CPU cores and 4 GiB of RAM. The fastest path is Docker Compose: clone the repository, copy .env.example to .env, and run docker compose up -d from the docker directory. Once the containers are running, the setup wizard is available at http://localhost/install. Dify also offers a hosted Dify Cloud option with a free sandbox plan that includes 200 GPT-4 calls, for teams that prefer not to manage infrastructure, alongside a self-hosted Community Edition for teams that want full control over their data and deployment.

Ideal use cases

Dify fits teams building internal copilots, customer-facing chatbots, document Q&A tools, or multi-step agent workflows that call external APIs. Because the RAG pipeline and agent tools are built in, it removes the need to separately wire up a vector store, a retrieval layer, and a tool-calling loop for common LLM app patterns.

Observability and operations

Dify integrates with third-party observability tools, including Opik, Langfuse, and Arize Phoenix, so teams can trace requests and debug LLM behavior alongside the built-in LLMOps dashboards. Application logs and performance metrics are tracked over time, which lets teams feed real production data and annotations back into prompt and dataset improvements rather than tuning models in the dark. The project maintains translated documentation and README files in more than a dozen languages, including Chinese, Japanese, Korean, Arabic, and Turkish, reflecting a large non-English user base alongside its English-language core docs.