Dify AI is an open-source platform for building AI apps, chatbots, agents, and automated workflows using large language models (LLMs). It offers visual tools for prompt design, RAG, automation, publishing, and monitoring through Dify Cloud or self-hosting.
What Is Dify AI?
Dify is an LLM app development platform, sometimes called an LLMOps platform. Instead of writing code to connect a model, a vector database, an API layer and a chat interface, you assemble those pieces on a visual canvas and publish the result in one place.
Founder Luyu Zhang, who previously led Tencent Cloud’s CODING DevOps platform, launched Dify in March 2023. The code went open source on 15 May 2023.
Growth since then has been fast:
| Fact | Detail |
| Developer | LangGenius, Inc. |
| Launched | 2023 (open-sourced May 2023) |
| GitHub | Passed 100,000 stars in June 2025; now among the most-starred projects of all time |
| Footprint | 1.4M+ machines in 175+ countries (March 2026) |
| Funding | $30M Series Pre-A at a $180M valuation, led by HSG (March 2026) |
| Known users | Volvo Cars, Maersk, Ricoh, ETS |
| Current version | 1.17.x line (September 2026) |
What You Can Build With Dify
Every plan lets you create five app types:
- Chatbot: a conversational assistant, such as a support bot.
- Text generator: single-shot output, such as product descriptions or email drafts.
- Agent: an LLM that reasons and calls tools (search, APIs, calculators) to finish a task.
- Chatflow: a chatbot whose replies follow a designed, multi-step logic.
- Workflow: an automated pipeline that runs without chat, such as “summarise every new contract and flag risky clauses.”
You can publish any of these as a web app or an API, so the same bot can sit on a website and inside your own software.
Dify’s Core Features at a Glance
| Feature | What it means in practice |
| Visual workflow builder | Drag nodes (LLM, knowledge retrieval, code, if/else, HTTP) onto a canvas instead of writing glue code |
| RAG knowledge base | Upload PDFs, docs or websites; Dify chunks, indexes and retrieves them for grounded answers |
| Model-agnostic | Works with OpenAI, Anthropic, Llama, Azure OpenAI, Hugging Face and local models via Ollama |
| Agents and tools | Agent nodes can call plugins and external tools, and agents can call Dify workflows directly |
| Plugins and marketplace | Extend Dify with model providers, tools and connectors |
| Observability | Logs, annotations and app monitoring to see what users ask and how the model answers |
The Three Ways to Run Dify
This is the decision most comparison articles skip, and it shapes your cost, workload and risk.
| Dify Cloud | Self-hosted Community | Dify Enterprise | |
| Who runs it | Dify | You (Docker or Kubernetes) | You or Dify, with vendor support |
| Software cost | Free to $159/month per workspace | Free | Custom quote |
| Your costs | Subscription plus LLM usage beyond credits | Servers, maintenance, LLM API fees | License plus infrastructure |
| Updates and patches | Automatic | Your job | Supported |
| Best for | Fast start, small teams | Data control, tech-savvy teams | Large companies with SSO, governance and compliance needs |
Rule of thumb: start on Cloud to prove the idea. Move to self-hosting only once you have a reason, such as keeping data in-house, controlling costs at scale, or running local models.
Dify Pricing in 2026
Dify Cloud prices are set per workspace and exclude taxes.
Current figures from Dify’s official pricing page:
| Plan | Monthly | Annual | Message credits | Members | Apps | Knowledge storage |
| Sandbox | Free | Free | 200 (one-time) | 1 | 5 | 50 MB |
| Professional | $59 | $590/yr | 5,000/month | 3 | 50 | 5 GB |
| Team | $159 | $1,590/yr | 10,000/month | 50 | 200 | 20 GB |
| Enterprise | Custom | Custom | Custom | Custom | Custom | Custom |
Two things buyers often miss:
- The Sandbox credits are a trial, not a monthly allowance. But once they run out, you can plug in your own model API key and keep building.
- Message credits are consumed per model. Bigger models use credits faster, so budget for your actual model choice, not the headline number.
The License Clause Agencies Need to Read
Dify calls itself open source, but its license is a modified version of Apache 2.0 with two extra conditions:
- No multi-tenant service without permission. Unless Dify authorises it in writing, you can’t use the source code to run a multi-tenant environment.
- Keep the branding. You can’t remove or modify the Dify logo or copyright information in the Dify console or applications.
Why this matters: if you’re a web or AI agency planning to host Dify and give each client their own workspace, that setup may need a commercial agreement with LangGenius. Using Dify for your own company’s internal apps, or building one app for one client, is a different situation. Members of the open-source community have also questioned whether these extra terms fit the official open-source definition, which is why some reviewers describe Dify as “source-available.”
This is general information, not legal advice. Check the current LICENSE file or contact Dify before building a resale model.
Security Lessons From DifyTap (2026)
In mid-2026, Zafran Labs disclosed four vulnerabilities in Dify, together called DifyTap. They included authorisation bypasses, a path traversal flaw, and bugs that could expose files across tenants. In Dify’s multi-tenant cloud, this meant one user could potentially read another customer’s AI conversations and uploaded documents. Dify fixed the issues through responsible disclosure, and Zafran says all four CVEs are patched in version 1.15.0.
There’s no evidence in the public reporting that the flaws were exploited.
But they offer a useful checklist for anyone self-hosting:
- Stay current. Run version 1.15.0 or later, and subscribe to Dify’s GitHub releases.
- Don’t expose the admin console to the internet. Zafran found tens of thousands of internet-facing Dify instances.
- Review trace-provider settings for destinations you don’t recognise.
- Scan the application itself, not just the container. Zafran noted that standard container scanners missed Dify’s own code.
- Treat uploaded files as untrusted, especially PDFs going into your knowledge base.
Dify vs n8n vs Langflow vs Flowise
| Dify | n8n | Langflow | Flowise | |
| Main job | Complete LLM apps with knowledge base and publishing | Business process automation with AI steps | Python-friendly visual LLM builder | Visual LLM flow builder |
| License | Modified Apache 2.0 | Sustainable Use (source-available) | MIT | Apache 2.0 |
| Status (2026) | Actively developed, VC-funded | Actively developed | Now under IBM via DataStax | Archived August 2026 |
| Pick it if… | You want a production chatbot or RAG app fast | You need to connect many SaaS tools | Your team lives in Python | Only for existing installs; plan a migration |
If you prefer writing code to visual building, frameworks such as LangGraph and CrewAI give you more control, at the cost of building the interface and hosting yourself.
A 30-Minute First Project: An Internal Policy Bot
Here’s a practical way to test Dify before you commit:
- Sign up for the free Sandbox and add your own model API key so the trial credits last.
- Create a Knowledge base and upload three to five real documents, such as an HR handbook and leave policy.
- Create a Chatflow app with a Knowledge Retrieval node feeding an LLM node. In the prompt, tell the model to answer only from the retrieved documents and say “I don’t know” otherwise.
- Test with ten real questions, including two the documents can’t answer. Check whether the bot correctly refuses those.
- Review the logs to see which document chunks were retrieved for each answer.
If the bot handles the trick questions well, Dify is probably a good fit. If retrieval keeps missing, fix the chunking and document quality before blaming the model.
Who Should (and Shouldn’t) Use Dify
Dify is a strong fit if you:
- Want a working AI chatbot, RAG app or agent in days, not months
- Have mixed teams where non-developers design and developers integrate through the API
- Need to switch between models (GPT, Claude, Llama) without rewriting apps
Look elsewhere if you:
- Need a strictly OSI-approved license for redistribution
- Mainly need to automate SaaS-to-SaaS workflows (n8n fits better)
- Plan to resell Dify itself as a multi-tenant service without a commercial deal
Conclusion
Dify AI has earned its popularity. It removes most of the plumbing between “we have an LLM” and “we have a working AI app,” and it lets developers and non-developers build together. Treat it like any serious piece of infrastructure, though. Start on Cloud, prove one real use case such as the policy bot above, then decide on self-hosting with clear eyes. Read the license if you serve multiple clients, and keep patching. Get those basics right, and Dify is one of the fastest routes from AI idea to production in 2026.
Want to compare Dify with other AI builders? Explore agentic AI tools and the full Rentaai AI tools library to find the right platform for your next project.
Frequently Asked Questions
Is Dify AI free?
Yes, in two ways. The self-hosted Community edition is free to run on your own servers, and Dify Cloud has a free Sandbox plan. Paid Cloud plans start at $59 per workspace per month.
Is Dify AI really open source?
The code is public on GitHub, but the license is a modified Apache 2.0 that restricts multi-tenant use and logo removal. Some community members therefore call it source-available.
Do I need to know how to code to use Dify?
No. Most apps can be built visually. Coding helps for custom code nodes, API integration and self-hosting.
Which AI models does Dify support?
Dify works with major providers such as OpenAI, Anthropic, Azure OpenAI and Hugging Face, plus local models through Ollama.
Is Dify safe for business data?
Dify has had serious vulnerabilities, but they were patched after responsible disclosure. For sensitive data, keep it updated, restrict console access, or consider self-hosting or the Enterprise edition.


