Dify
Dify.AI is a next-generation platform designed to build and operate generative AI applications. It offers functionalities for creating AI workflows and custom a...
Last verified:
What is Dify?
Dify is an open-source platform for building production-ready AI applications and agentic workflows. It combines Backend-as-a-Service and LLMOps to streamline generative AI development, making it accessible to both developers and non-technical innovators. The platform lets you define processes visually using a drag-and-drop interface, connect existing tools and data sources, and deploy AI applications that solve real problems.
Key features include: visual workflow orchestration for building sophisticated AI apps in minutes; comprehensive RAG (Retrieval-Augmented Generation) pipeline with document ingestion, chunking, and vector storage; agent capabilities supporting LLM function calling and ReAct patterns with 50+ built-in tools like Google Search, DALL·E, and WolframAlpha; support for 100+ LLM providers including open-source models via Ollama; native MCP (Model Context Protocol) integration for bridging systems and publishing universal MCP servers; prompt IDE with version history and A/B testing; built-in observability with execution traces, token usage tracking, and logging; and a plugin marketplace for extensibility without source code modification.
Dify is designed for ambitious teams ranging from startups validating AI ideas to enterprises deploying AI across departments. Developers benefit from the complete self-hosting option with no enterprise feature paywalls, while non-technical users can build AI apps through the intuitive no-code visual interface. The platform serves over 1 million applications across 100+ countries, trusted by industry leaders from biomedicine to automotive.
Dify pricing
Pricing model: Freemium
Dify offers three cloud pricing tiers plus a free self-hosted option. Sandbox (Free): 200 OpenAI message calls, 5 applications, 1 member, 5MB vector storage, 30 days logs, standard support. Professional ($59/month or $49/month annually): 5,000 credits/month, 50 applications, 3 members, 200MB Knowledge Base storage, unlimited log history, priority support, custom branding. Team ($159/month or $132/month annually): 10,000 credits/month, 200 applications, 50 members, 20GB storage, unlimited members, priority support, advanced analytics. Enterprise plan with SOC 2 Type II compliance and dedicated support available on request via [email protected]. The open-source Community Edition (CE) is free under Apache 2.0 license with unlimited apps and all core features, but commercial use requiring logo removal, multiple workspaces, or SaaS model needs commercial licensing.
Dify pros
- Open-source option (Apache 2.0) reduces vendor lock-in compared to closed no-code AI platforms
- Visual drag-and-drop workflow builder accessible to non-developers while maintaining technical depth
- Production-grade RAG pipeline with advanced document processing, chunking, and vector storage
- Support for 100+ LLM providers including open-source, proprietary, and local models via Ollama
- Agent mode with autonomous tool selection supporting LLM function calling and ReAct patterns
- 50+ built-in tools for AI agents including Google Search, DALL·E, Stable Diffusion, WolframAlpha
- Native MCP integration as both consumer and universal MCP server publisher
- Structured Output feature with visual JSON schema editor for accurate formatted LLM outputs
- Complete self-hosting option with no enterprise feature paywalls for core capabilities
- Built-in observability with execution traces, latency tracking, token usage per node, and annotations
- Prompt IDE with version history, model comparison, and A/B testing capabilities
- Backend-as-a-Service handles API complexity, rate limits, and infrastructure automatically
- Plugin marketplace for extending capabilities without touching source code
- Scalable architecture that effortlessly handles increasing traffic and evolving needs
- Enterprise-grade security with SOC 2 Type II compliance available for Enterprise plan
Dify cons
- Steep learning curve for non-technical users without developer assistance
- Docker deployment complexity requires DevOps knowledge and significant resources
- Self-hosted build lacks some cloud-only features available in managed version
- Workflow version control and collaboration features still evolving
- Plugin marketplace is in beta with smaller ecosystem than established automation platforms
- Complex debugging can be manual without advanced automated tools
- Visual workflow builder becomes unwieldy for very complex multi-step processes
- More complex than simple chatbot builders; non-technical users may need developer help for data sources and tools
Frequently asked questions about Dify
What is Dify and what does it do?
Dify is an open-source platform for building agentic workflows and AI applications. The name comes from 'Do It For You.' It combines Backend-as-a-Service and LLMOps to streamline generative AI development, letting you define processes visually through drag-and-drop, connect existing tools and data sources, and deploy AI applications that solve real problems. Key capabilities include AI workflow orchestration, RAG pipelines, agent capabilities, model management, and observability features.
Is Dify free to use?
Yes, Dify offers multiple free options. The Cloud Sandbox plan is free with 200 message credits, 5 applications, 1 member, and 30 days of logs. The Community Edition is completely free open-source software under Apache 2.0 license with unlimited apps and all core features when self-hosted. However, after using up the free OpenAI allowance, you need to obtain LLM API keys from model providers and add them under Settings → Model Providers. Commercial use of self-hosted version requiring logo removal, multiple workspaces, or SaaS model requires contacting [email protected] for licensing.
How do I build an AI agent with Dify?
To build an AI agent in Dify: 1) Go to Studio → Workflow → Create from Blank → Orchestrate → New Chatflow. 2) Add an Agent node using the Add Node button. 3) In the right panel, click Find more in Marketplace to install an Agent Strategy (like Dify Agent Strategy). 4) Add an LLM node after the Agent node to process text, images, and documents. 5) Configure your model provider in System Settings → Model Providers first. 6) Add tools for the agent to use (50+ built-in tools available like Google Search, DALL·E). 7) Connect nodes and test using Preview. The agent supports LLM function calling and ReAct patterns with autonomous tool selection.
What LLM models does Dify support?
Dify supports 100+ LLM providers covering open-source, proprietary, and self-hosted solutions. This includes GPT, Mistral, Llama3, Anthropic, Google, Azure OpenAI, Hugging Face, Replicate, Llama2, and any OpenAI API-compatible models. You can also use local models through Ollama and any OpenAI-like API. The platform allows you to access, switch, and compare performance of different LLMs worldwide with model provider management that lets you swap models in one click.
What is RAG in Dify and how does it work?
RAG (Retrieval-Augmented Generation) in Dify is a pipeline that gets your data LLM-ready. It extracts data from various sources, transforms it, and indexes it into vector databases for optimal LLM use. The pipeline covers everything from document ingestion to retrieval, with out-of-box support for text extraction from PDFs, PPTs, and other common document formats. It includes automatic document chunking, embedding, and vector storage. Each 1MB of vector storage can store about 1.2 million characters of vectorized data (estimated using OpenAI Embeddings). You can create a knowledge base, then configure a knowledge retrieval node in your workflow to query it.
What is Structured Output in Dify?
Structured Output is a native feature in Dify's workflow that makes AI agent chat with APIs easier. It ensures the LLM's output is accurate and formatted without requiring coding, prompting, and validation. The visual JSON schema editor is easy to use and can seamlessly use formats already existing in your database. You toggle Output Variables Structured ON, click Configure → Import from JSON, and paste your JSON format. The structured output object generated by the LLM can then be used in subsequent nodes like HTTP requests to post data into your database.
What is MCP integration in Dify?
MCP (Model Context Protocol) integration in Dify is native support for bridging your systems and platforms. It allows you to access external APIs, databases, and services through standardized MCP protocols, eliminating integration complexity and maintenance overhead. Dify supports HTTP-based MCP services (protocol 2025-03-26) with pre-authorized and auth-free modes. You can also publish your Dify-built workflow or agent as a universal MCP server, making it accessible across unlimited MCP clients. Community developers have contributed multiple MCP plugins to the marketplace for connecting external MCP services like Zapier to Dify Agent applications.
How do I deploy and publish my Dify application?
Dify offers flexible publishing options tailored to diverse needs. You can publish your AI app as a WebApp with a chat widget, expose it via standard API for integration into your own applications, or publish as a universal MCP server for MCP client access. Dify's Backend-as-a-Service handles the complexities of API calls, rate limits, and infrastructure. For cloud deployment, use cloud.dify.ai/apps. For self-hosting, deploy via Docker Compose following the quick-start guide. The platform is production-ready from day one with scalable architecture, stable operation, and enterprise-grade security.
What observability and monitoring features does Dify have?
Dify includes built-in LLMOps capabilities for monitoring and analyzing application performance. Features include execution traces showing inputs/outputs, latency tracking, token usage per node, run logs with duration measurements, and annotation systems for manual editing and customization of high-quality responses. You can continuously improve prompts, datasets, and models based on production data. The platform supports integration with external observability tools including Opik, Langfuse, and Arize Phoenix. Log history is available for 30 days on Sandbox, unlimited on Professional and Team plans.
Can I customize and extend Dify with plugins?
Yes, Dify has a plugin marketplace that allows you to bridge any models and tools without touching the source code. You can discover, extend, or submit plugins that bring more capabilities to the community. The marketplace offers advanced large language models including multimodal abilities, tools for building advanced agentic AI, and connectors for automation. Community developers have contributed multiple MCP plugins. The plugin marketplace is currently in beta. You can also build and publish your own plugins following the development guide in the documentation.