Dify

Dify is an open-source platform for building applications on top of large language models. Developed by LangGenius and hosted on GitHub as langgenius/dify, it lets teams build chatbots, AI agents and agentic workflows on a visual canvas, connect them to their own documents, and publish the result as a web app or an API 1 2. The project describes itself as a place to "build agentic workflows, RAG pipelines, with rich AI model and tool support on one collaborative workspace" 1.

What you can build with Dify

Dify organises work into apps. The main types are chat assistants, agents that choose and call tools on their own, and workflows, where you connect nodes (model calls, knowledge retrieval, conditions, code, HTTP requests, tools) into a fixed pipeline 1 2. A chat-style workflow, called a chatflow, adds conversation memory to the same canvas. Every app can be published as a hosted web app or called from your own code through its REST API, which Dify calls backend-as-a-service 1 2.

Core features

• Visual workflow builder: design, run and debug multi-step AI workflows on a canvas 1
• Model support: the repository lists integration with hundreds of proprietary and open-source LLMs from dozens of inference providers, so one app can switch models without a rewrite 1
• Prompt IDE: write prompts and compare how different models respond 1
• Knowledge and RAG: upload PDFs, slides and other documents, which Dify chunks, indexes and retrieves at answer time (see retrieval-augmented generation) 1 3
• Agents and tools: agents can call built-in tools, plugins from the Dify Marketplace and external APIs 1 2
• LLMOps: logs and monitoring to track how an app performs over time 1

How the knowledge base works

A Dify knowledge base can use one of two index methods. High Quality turns each chunk into an embedding vector and supports vector search, full-text search or hybrid search that combines both, with an optional rerank model. Economical uses keyword-based inverted indexing, costs no tokens and is less accurate 3. Retrieval settings such as Top K and a score threshold decide how many chunks reach the model. A knowledge base created with High Quality indexing cannot be switched to Economical later 3.

Dify and MCP

Dify supports the Model Context Protocol in both directions. You can add an external MCP server by giving its URL, a name and an identifier; Dify connects, handles OAuth if the server needs it, and imports the server's tools. Those tools can then be used inside agent apps or as tool nodes in workflows and chatflows 4. Only MCP servers that use HTTP transport are supported 4. The same documentation points to a separate guide for publishing your own Dify app as an MCP server, so other clients can call it 4.

Cloud or self-hosted

Dify is available as a managed cloud service or as a self-hosted Community Edition 2. The repository's quick start runs it with Docker Compose and lists a minimum of 2 CPU cores and 4 GiB of RAM 1.

The code is released under the Dify Open Source License, which is based on Apache 2.0 with two extra conditions: you may not use the source code to run a multi-tenant service without written permission, and you may not remove or change the logo or copyright information in the Dify console or apps 5. Because of those conditions it is not a plain Apache 2.0 license, which matters if you plan to resell Dify as a hosted product.

Dify compared with similar tools

• LangChain is a code library; Dify is an application with a visual builder, hosting and an API layer on top.
• Langflow is also a visual LLM app builder; Dify puts more into built-in knowledge base management and app publishing.
• n8n and Zapier are general automation tools with AI steps; Dify is built around LLM apps first.

Using Dify for social media workflows

A common pattern is a Dify workflow that drafts posts from a brief or a knowledge base of brand guidelines, followed by a review step and a publishing step. For the last step, a Dify agent can use a social publishing MCP server as a tool. Publora, a social media scheduler and publishing API, runs a Streamable HTTP MCP server with 18 tools, which fits Dify's HTTP-only MCP requirement 4 6. Keep a human approval step before anything goes live.