Dify
Visual AI app builder — RAG, agents, workflows, and MCP in one platform
📋 Overview
Dify is a production-ready visual platform for building AI applications — combining RAG pipelines, agentic workflows, model orchestration, and observability in a single interface. Unlike LangChain (code-only) or Flowise (basic visual builder), Dify gives you a full IDE for AI apps: visual prompt chaining, document ingestion with OCR/embedding, tool-use agents with MCP support, conversation flow design, and built-in analytics. The platform supports 100+ LLM providers (OpenAI, Anthropic, Gemini, Llama, Qwen, etc.) with a unified API layer, and the RAG engine handles chunking, reranking, and hybrid search out of the box. Apache 2.0 licensed with a generous free tier.
✨ Key Features
- •149K+ GitHub stars — largest visual AI builder
- •Full RAG pipeline: ingestion, chunking, embedding, hybrid search, reranking
- •100+ LLM providers via unified API layer
- •Visual agent canvas with MCP tool-use support
- •Built-in observability: conversation logging, annotation, analytics
- •Multi-model orchestration: route prompts by cost, latency, or capability
- •Apache 2.0 license — no vendor lock-in, no usage caps on self-hosted
🎯 The Problem It Solves
Building production AI apps requires stitching together 5+ tools: a vector DB, an embedding service, a prompt manager, an observability stack, and a UI. Dify collapses all of these into one platform — letting teams go from prototype to production in days instead of months.
🔧 How It Works
Apps are built visually on a canvas: drag nodes for LLM calls, knowledge retrieval, conditionals, tool use, and code execution. The RAG engine ingests documents (PDF, HTML, Notion, etc.), chunks them, embeds them, and serves them via hybrid search (dense + sparse). Agents can call external tools via MCP or custom API definitions. All conversations are logged and annotated for continuous improvement. Deploy via Docker or Dify Cloud.
🚀 Installation & Quick Start
Installation
git clone https://github.com/langgenius/dify && cd docker && docker compose up -dQuick Start
- git clone && docker compose up
- Open localhost:install
- Setup admin
- Create app in visual builder
✅ Pros
- •Most complete visual AI builder — RAG + agents + workflows in one
- •100+ LLM providers with unified API — switch models without code changes
- •Production-grade RAG with OCR, hybrid search, and reranking
- •MCP support for agent tool-use is native, not bolted-on
- •Built-in analytics and conversation annotation for continuous improvement
- •Self-hosted is completely free with no message caps
- •Active community with templates, plugins, and enterprise support
❌ Cons
- •Visual canvas becomes limiting for complex multi-step logic
- •Free cloud tier message caps (50K/mo) are tight for production
- •Self-hosting requires PostgreSQL, Redis, and Weaviate — not lightweight
- •Custom tool integrations sometimes require code despite visual promise
- •Enterprise pricing requires sales conversation — no self-serve
💬 Practitioner Verdict
“Dify is the most complete visual AI app builder available in 2026 — the RAG engine alone justifies adoption, and the MCP/agent support keeps it current. The trade-off: the visual canvas can feel limiting for complex multi-step logic, and the free cloud tier has tight message caps. For teams that want to ship AI apps without building infrastructure, Dify is the default.”
Self-Hosted (Free)
Open source, MIT/Apache licensed. Run it yourself.
⭐ Star & Clone on GitHubFree forever. Your infrastructure, your data.
Dify Cloud Free
50K messages/mo, 5 apps
- 50K messages
- 5 apps
- Community support
Dify Cloud Pro
1M messages, unlimited apps
- 1M messages
- Unlimited apps
- Priority support
Dify Cloud Team
SSO, team collaboration, audit
- SSO
- Team workspaces
- Audit logs
Enterprise
VPC, HIPAA, dedicated support
- VPC/HIPAA
- Dedicated support
- Custom SLA
Deployment Options
Ways to run Dify in production
📊 Specifications
- Language
- TypeScript + Python
- License
- Apache-2.0
- Platform
- Linux, macOS, Windows
- Supported Models
- REST API, CLI
💰 Pricing Reality
Self-hosted is 100% free Apache 2.0. Cloud: Free tier (50K messages/mo), Pro 9/mo (1M messages), Team 59/mo (unlimited apps, SSO), Enterprise custom (VPC, HIPAA, dedicated support). LLM API usage billed separately.