Dify is an LLM app development platform for building agents, RAG knowledge pipelines and visual workflows, then shipping them as web apps, APIs or MCP tools. Use the hosted Dify Cloud plans or self-host the Community Edition with Docker under its modified Apache 2.0 license.
Used this tool? Rate it
Used this tool? Rate it
Dify is an LLM app development platform for building agentic workflows, agents, knowledge pipelines and tool integrations on a visual canvas, then publishing the result as a web app, an API or an MCP-compatible tool. It is operated by LangGenius, Inc., a Delaware corporation, and comes in three editions that differ mainly in who runs the infrastructure.
The practical choice is where Dify runs: Cloud plans meter credits, members, apps and storage per workspace, while the free self-hosted edition carries license conditions covered under Limitations.
As an agentic workflow builder, Dify assembles workflows from visual building blocks and AI prompts that define how an app thinks, retrieves data, makes decisions, uses tools, asks for human input and completes tasks.
An agent can be created by chatting with the builder or by configuring it manually. The same agent can then run as a standalone app, be embedded in a website, or be reused as an agent node inside a larger workflow.
Knowledge pipelines take files, websites, online documents and drives, then extract, clean, chunk and index the content and let you test retrieval before the knowledge base is connected to agents or workflows.
Dify is model-agnostic. On Dify Cloud, built-in AI credits cover models from providers such as OpenAI, Anthropic, Gemini, xAI and Tongyi. One AI response is a single model call that counts once no matter how many tokens it uses. Model providers, tools, data sources and MCP integrations can be installed from the Marketplace or defined yourself. Packages can also come from any public GitHub repository by URL and version, or be uploaded locally as a .zip for private or internal integrations.
Publishing creates a web app and an API endpoint from the app's latest configuration. The web app can also be embedded on another site as a chat widget or inline frame. After launch, logs, user feedback, annotations, latency and usage data show how the app behaved.
For self-hosted Dify, the documented minimum for a Docker Compose deployment is a machine with at least 2 CPU cores and 4 GiB of RAM. Dify also requires Docker Compose 2.24.0 or later.
dify/docker directory..env.example to .env; this file is where ports, storage, vector stores and feature options are customized later.docker compose up -d, then run docker compose ps and confirm that every container reports Up or healthy, while an exited init_permissions task is expected./install on localhost or on your server's IP address to create the administrator account, then sign in at the root address.mcpServers in a project's .cursor/mcp.json for Cursor.Dify's documentation and solution pages describe these patterns:
Both figures are customer outcomes published by Dify, not independent measurements.
Dify's plans are each described for a type of team:
Inside a workspace, Dify separates builders from consumers with four roles — Owner, Admin, Editor and Normal — and only the Owner and Admin can manage roles, billing and models.
It fits poorly for reselling Dify as a multi-tenant service or shipping its frontend without the Dify logo, since both need Dify's permission.
Dify pricing on the Cloud edition is per workspace, and prices exclude applicable taxes, which may be added at checkout. Professional costs $59 per workspace per month and Team $159, with annual billing saving $118 and $318 a year respectively. Billed annually, Professional is $590 per workspace per year, a discount the pricing page labels as 17%. Team is $1590 per workspace per year on annual billing.
| Plan | Message credits | Members | Apps | Knowledge documents | Knowledge storage |
|---|---|---|---|---|---|
| Sandbox (Free) | 200 | 1 | 5 | 50 | 50MB |
| Professional | 5,000 / month | 3 | 50 | 500 | 5GB |
| Team | 10,000 / month | 50 | 200 | 1,000 | 20GB |
Sandbox's 200 credits are a total allowance with no monthly reset.
Enterprise is quoted individually through sales and is the only tier that includes commercial license authorization, multiple workspaces and SSO. The self-hosted Community Edition is free.
Billing rules that change the real cost:
One community discussion about free, open-source LLM app builders listed Dify alongside Flowise, Langflow, n8n and Make, noting that n8n and Make are more general while the others are tailored to LLMs.
Against these, Dify stands apart through its built-in knowledge pipeline, bundled model credits on hosted plans, and its license conditions.
Dify is licensed under a modified version of the Apache License 2.0 with additional conditions:
Dify's own pages describe the Community Edition in different terms, and they are listed here as published rather than reconciled. The homepage card calls it "Source-available" under an "Apache-2.0-derivative license". The site navigation labels the same edition "Open-source, self-deploy with Docker". The documentation index calls Dify "an open-source platform for building AI applications." On pricing, the Community tier is framed for non-commercial projects and commercial license authorization is listed as Enterprise-only, whereas the enterprise pricing FAQ says the Community Edition can be self-hosted for free with "no license required."
An independent security-news report dated June 22, 2026 described four vulnerabilities, collectively named DifyTap, of which two were rated critical and three had cross-tenant impact on Dify's multi-tenant cloud service. According to that report, all of them except CVE-2026-41948 were addressed in version 1.14.2, and a fix for the remaining flaw was expected in the next Dify release.
The Terms of Service tell users not to rely on factual assertions in output without independent fact-checking, or on designs, workflows or code without reviewing them. They also prohibit using Dify to make automated decisions that may harm individual rights without appropriate human supervision.
Yes. Your own key and the included AI credits can coexist, and a Usage Priority setting on each provider card decides which one Dify draws from first before falling back to the other. A key added by the owner or an admin works across the whole workspace and bills to your own account with that provider.
Current students, teachers and educational staff can verify their affiliation to use the Yearly Professional plan for free, with re-verification required annually. Applicants must be at least 18 and register with a school-issued email address. Education workspaces keep Professional features and quotas, but AI credits stay at 200 one-time credits with no monthly refill.
Not yet. Pay by Invoice through bank transfer is being explored but is not available; under the proposed arrangement it would apply only to annual Team subscriptions.
Nothing is deleted. If a downgrade or expired subscription leaves your data above the new plan's limit, apps can still retrieve from those knowledge bases, but adding content stays blocked until usage fits the limit or you upgrade again.
Dify Cloud is described as running on hardened infrastructure with encryption in transit and at rest, with data stored in Dify's managed cloud region; the specific region is not named on the pricing page.