Dify
Open-source LLM app platform from LangGenius for agentic workflows and RAG pipelines. Modified Apache 2.0: self-host freely, multi-tenant SaaS needs a commercial licence.
Overview
LangGenius put Dify together as a single workspace where a team can build an LLM application without stitching five libraries together. The canvas handles branching workflows, loops, HTTP calls and code nodes. Underneath sits a full RAG pipeline with ingestion, chunking, hybrid retrieval and reranking, plus a model layer that already knows how to talk to DeepSeek, Qwen, Moonshot, GLM, OpenAI, Anthropic and local endpoints served by Ollama or Xinference. Every app you build gets an OpenAI-compatible REST endpoint, which is what keeps the visual builder from becoming a trap.
Read the licence before you plan a product around it. Dify is not plain Apache 2.0. The Dify Open Source License adds two conditions. You may not use the source to operate a multi-tenant environment, defined as one workspace per organisation, without written authorisation from LangGenius. And you may not remove or modify the logo or copyright information in the Dify console, which applies specifically to the web frontend. Self-hosting for your own company, building commercial applications on top and selling those applications are all permitted. Reselling Dify itself as a hosted service is not.
Practically, the project is in unusually good shape. English documentation is complete, GitHub issues get answered in English, and both the Docker Compose and Helm paths work without local patching. v1.17.0 shipped 25 August 2026 with E2B sandbox support, home snapshots and skill management for the Agent product that entered open beta in July. The repository stands at 153,746 stars with commits landing the same day this was checked. Cloud pricing is per workspace in USD at 59 dollars a month for Professional and 159 for Team, so mainland teams generally self-host and skip the payment friction entirely.
Key Features
- ✓ Visual workflow canvas with branching, loops, HTTP nodes and code nodes, plus the Dify Agent product that entered open beta in v1.16.0 on 17 July 2026
- ✓ Built-in RAG pipeline covering ingestion, chunking, hybrid retrieval and reranking, with no separate retrieval framework required
- ✓ Agent sandbox for code and shell execution, hardened in v1.16.1 behind a dedicated Squid forward proxy with strict ACL rules
- ✓ Model providers preconfigured for DeepSeek, Qwen, Moonshot, GLM, OpenAI, Anthropic and local endpoints served by Ollama or Xinference
- ✓ Every app is exposed as an OpenAI-compatible REST endpoint, so work built in the UI moves to code without a rewrite
- ✓ Self-host through Docker Compose or a Helm chart, with observability and tracing integrations added through the 1.17 line
Where it holds
- • The licence still permits commercial self-hosting and selling applications built on top. The restriction only bites if you resell Dify itself as a hosted multi-tenant service
- • 153,746 stars, 24,300 forks and commits landing daily, so support for new Chinese model releases usually appears within days
- • Complete English documentation and an English-speaking issue tracker, which is uncommon for a project with a Chinese core team
- • OpenAI-compatible per-app endpoints mean the visual builder is a starting point rather than a dead end
Where it breaks
- • Not OSI-approved. The Dify Open Source License forbids using the source to operate a multi-tenant environment, defined as one workspace per organisation, without written authorisation, and forbids removing or modifying the logo and copyright in the web console
- • Postgres, Redis, a vector store and the sandbox container add up to a heavy footprint for a small deployment, and minor-version upgrades have historically needed careful migration
- • The canvas hits a ceiling on genuinely complex agent logic, at which point teams write custom nodes anyway
- • Cloud tiers are priced per workspace in USD, which is awkward to expense on a mainland corporate card and pushes most local teams to self-host
My Take
The licence is the thing to read first. Dify ships under a modified Apache 2.0 that lets you self-host and sell what you build on it, but forbids running Dify's own source as a multi-tenant service without written authorisation, and forbids stripping the logo from the web console. Inside that boundary it is the most complete open LLM app builder available, at 153,746 GitHub stars with v1.17.0 tagged 25 August 2026. Providers for DeepSeek, Qwen, Moonshot and GLM are wired in by default, which saves real configuration time on a mainland deployment.
Quick Info
- Pricing:
- freemium
- Openness:
- Open source
- Licence:
- Dify Open Source License (Apache 2.0 with additional conditions)
- Starting at:
- Self-hosting is free under the Dify Open Source License via Docker Compose or Helm. Dify Cloud is billed per workspace in USD: Sandbox 0 dollars with 200 total message credits, Professional 59 dollars a month (5,000 credits, 3 members, 50 apps, 5 GB knowledge storage), Team 159 dollars a month (10,000 credits per workspace). Annual billing takes roughly 17 percent off. Operating Dify source as a multi-tenant service requires a separate commercial agreement with LangGenius.
- Added:
- Aug 2026
- Updated:
- Aug 2026
Use Cases
Judge it on your own work
The notes above say where Dify holds and where it breaks. The fastest check is your own workload.
Visit website ↗Alternatives to Dify
Coze / 扣子
freeByteDance’s AI bot-building platform that enables anyone to create custom AI assistants with plugins, workflows, and knowledge bases without coding.
LangGraph
open-sourceLow-level orchestration framework for stateful, long-running LLM agents, with checkpointing, human-in-the-loop pauses and a managed deployment tier through LangSmith.
RAGFlow
open-sourceInfiniFlow's Apache 2.0 RAG engine built on deep document parsing. DeepDoc handles layout, tables and OCR before chunking, and every citation points back to a highlighted region in the source PDF.
LlamaIndex
freemiumOpen-source Python and TypeScript framework for RAG and agentic workflows, paired with LlamaCloud for document parsing, extraction and managed indexes.