Dify review
★ 4.6/5Open-source LLM app & agent builder
Dify is an open-source platform for building LLM apps, agentic workflows and RAG pipelines through a visual canvas, with a hosted cloud tier and a free self-hostable Community Edition. It blends a no-code/low-code builder with backend-as-a-service APIs, native MCP (client and server), and integration with hundreds of model providers, so teams can ship chatbots, agents and workflows without wiring infrastructure from scratch.
Affiliate link · how we make money
Best for LLM app building. Dify starts Free, then ~$59/mo on a Flat monthly plan (per-workspace) with included monthly message credits; free self-hosted Community Edition model. The Professional plan is $59/mo and includes 5,000 monthly message credits. Because credits are consumed unevenly by model (Dify lists ~200 credits as ~200 GPT-3.5 messages or only ~20 GPT-4-class messages), a team running mostly GPT-4-class chats burns roughly 10 credits per exchange, so 5,000 credits covers only ~500 such messages/month before you must upgrade to Team ($159/mo, 10,000 credits) or switch to your own API key. Self-hosting the Community Edition is $0 in license fees (you pay only your model API and infra costs).
Spec
The gotcha
"Message credits" are not 1 credit per message: premium models like GPT-4 consume roughly 10x the credits of GPT-3.5, so the headline 5,000/10,000 credit allowances run out far faster than they look. Credits only apply to Dify-hosted models; bring your own provider API key and you bypass them entirely (paying the model vendor directly instead).
Decode the real cost →Best for
- You want to visually build and ship LLM agents, RAG apps and workflows fast, with the option to self-host the open-source core for free or use the managed cloud with hosted model credits.
source: dify.ai/pricing →
Dify pricing tiers
Prices move. Cells flagged verify link to the live vendor page.
| tier | monthly | annual | included | unit |
|---|---|---|---|---|
| Sandbox (Free) | $0verify | n/a | 200 message credits (one-time trial), 1 team member, 5 apps, 50 knowledge docs, 50MB storage, 5,000 API calls/mo | per workspace |
| Professional | $59verify | $49 | 5,000 message credits/mo, 3 team members, 50 apps, 500 knowledge docs, 5GB storage, 20,000 trigger events/mo, no Dify API rate limit | per workspace/mo |
| Team | $159verify | $132 | 10,000 message credits/mo, 50 team members, 200 apps, 1,000 knowledge docs, 20GB storage, unlimited triggers, advanced analytics | per workspace/mo |
| Enterprise | Custom (contact sales)verify | n/a | SSO, SOC 2 Type II, dedicated support, custom limits; AWS Marketplace listing starts ~$150,000/yr | custom |
| Community Edition (self-hosted) | $0verify | n/a | Open-source core (modified Apache 2.0); free to self-host, you pay your own model/infra costs | self-hosted |
// pools and per-unit rates are volatile · cells flagged verify link to the live vendor page
Compare Dify with…
Questions
How much does Dify cost?
Dify starts Free, then ~$59/mo. The Professional plan is $59/mo and includes 5,000 monthly message credits. Because credits are consumed unevenly by model (Dify lists ~200 credits as ~200 GPT-3.5 messages or only ~20 GPT-4-class messages), a team running mostly GPT-4-class chats burns roughly 10 credits per exchange, so 5,000 credits covers only ~500 such messages/month before you must upgrade to Team ($159/mo, 10,000 credits) or switch to your own API key. Self-hosting the Community Edition is $0 in license fees (you pay only your model API and infra costs).
Can you self-host Dify?
Yes, Dify can be self-hosted.
Does Dify support MCP?
Dify supports MCP as both client and server, so it works with assistants like Claude and ChatGPT.
What's the catch with Dify's pricing?
"Message credits" are not 1 credit per message: premium models like GPT-4 consume roughly 10x the credits of GPT-3.5, so the headline 5,000/10,000 credit allowances run out far faster than they look. Credits only apply to Dify-hosted models; bring your own provider API key and you bypass them entirely (paying the model vendor directly instead).