by LangGenius, Inc.
Open-source LLM app builder with visual workflows, RAG and agents, run on Dify Cloud or your own Docker host. Backed by a $30M Series Pre-A round.
Choose Dify if your team wants to build and own an LLM app, such as a document assistant or an agent workflow, without writing the backend from scratch. It replaces glue code around model APIs and a vector store. Dify's About page claims 280 enterprise customers, and a self-hosted copy needs only 2 CPU cores.
The same workspace runs on Dify Cloud, in a private VPC or self-hosted from the public repository.
Main trade-off
Per-workspace member and app caps push growing teams to a higher plan or to self-hosting.
| Plan | Monthly price | What it includes |
|---|---|---|
| Sandbox | Free | 200 message credits, 1 workspace and 1 member, 5 apps, 50 knowledge documents and 50MB storage, 3,000 trigger events, up to 2 triggers per workflow, 30 days of log history |
| Community Edition | Free | All core features released in the public repository, Single workspace, Runs under the Dify Open Source License |
| Professional | $59/mo | 5,000 message credits a month, 3 team members, 50 apps, 500 knowledge documents and 5GB storage, 20,000 trigger events a month, Unlimited log history |
| Team | $159/mo | 10,000 message credits a month, 50 team members, 200 apps, 1,000 knowledge documents and 20GB storage, Unlimited trigger events, Priority workflow execution |
| Enterprise | Custom | Multiple workspaces and enterprise management, SSO, Commercial licence authorisation, Advanced security and controls, Updates, maintenance and technical support from Dify |
Cloud workspaces: Sandbox costs nothing; Professional runs $59 monthly or $590 billed yearly; Team runs $159 monthly or $1,590 billed yearly. Enterprise (self-hosted or VPC, SSO, commercial licence) is custom-priced through sales, and the Community Edition is free to self-host. Message credits cover trial use of hosted models; after that, model usage is billed by your own provider through your API key.
Dify is an open-source LLM app development platform that combines a visual workflow canvas, RAG knowledge pipelines, agents and model management in one workspace. Its GitHub repository holds more than 161,000 stars. Apps publish as web apps or APIs and run on Dify Cloud, inside a customer VPC, or self-hosted with Docker Compose.
Visual workflow canvas
Chain model, retrieval, code and HTTP steps on one canvas, with up to 2 triggers per workflow on Sandbox and unlimited triggers on the paid plans.
Knowledge pipeline
Turns uploaded documents into searchable knowledge bases; the Professional workspace allows 500 documents and 5GB of storage.
Chat, agent and workflow apps
Each app publishes as a web app or an API, and the paid plans put no Dify API rate limit on it.
Event triggers
Plugin, Schedule or Webhook triggers start workflows automatically, with 20,000 trigger events a month on Professional.
Plugin marketplace
A marketplace at marketplace.dify.ai carries tools, models and integrations that install into a workspace.
Nested workflow tracing
The newest release sends the steps inside called Workflow Tools to external trace tools such as Langfuse, keeping the LLM prompts with the execution tree.
| Tool | Price from | Free plan | Runs on | Rating |
|---|---|---|---|---|
| $59/mo | Yes | Web | 3.8 | |
| -- | Yes | Web | -- | |
| $49.99/mo | Yes | Web · iOS · Android · Mac · Windows | 4.0 | |
| -- | -- | Web | 3.5 |
Modified Apache 2.0 (Dify Open Source License)
Dify Cloud has three plans billed per workspace: Sandbox is free, Professional is $59 a month ($590 billed yearly) and Team is $159 a month ($1,590 billed yearly). Enterprise, which adds SSO, multiple workspaces and a commercial licence, is custom-priced through sales. The self-hosted Community Edition costs nothing, but you pay for your own server and your own model API usage.
Yes, in two ways. The Sandbox plan on Dify Cloud gives 200 message credits, 5 apps and 1 member, with a month of log history. The Community Edition has no plan caps at all, but you host and update it yourself.
The docs ask for at least 2 CPU cores, 4 GiB of RAM, Docker Compose 2.35.0 or later, Git and OpenSSL. OpenSSL generates the token the agent's local sandbox needs. The default stack bundles PostgreSQL and Weaviate, so no outside database is required, and on macOS the Docker Desktop virtual machine should have 8 GiB.
The code is public on GitHub under the Dify Open Source License, a modified Apache 2.0 that GitHub's licence detector lists as Other rather than a standard licence. Commercial use is allowed, including as a backend for your own apps. Contributors also agree that Dify's maker may change the licence terms and use contributed code in its cloud business.
Message credits let a workspace sample hosted models from five providers without any key, and each model type burns them at its own rate. When they are gone, you switch the workspace to your own API key from the model provider. Dify's pricing page says the credits exist to help you try models, so plan on your own key for production traffic.
Dify is an open-source platform for building LLM applications, made by LangGenius, Inc. The company says it founded Dify in March 2023 and has since raised a $30M Series Pre-A round led by HSG, while its GitHub repository has passed 161,000 stars. It sits with the AI agent builders on Hokai.
You work on a visual canvas: chain model calls, knowledge retrieval, code and HTTP steps into a workflow, or configure a chat or agent app, then publish it as a web app or an API. The hosted plans include message credits for trying models from OpenAI, Anthropic, Gemini, xAI and Tongyi before you bring your own key. A knowledge pipeline turns documents into searchable knowledge bases, and a marketplace adds tools, models and integrations as plugins.
It fits a developer who wants a private LLM stack started with one Docker Compose command, an operations team prototyping an internal assistant without writing a backend, and an enterprise platform group that wants one governed place for every team's agents. Version 1.19.0, released on 4 November 2026, added nested workflow tracing, so an external observability tool can follow a run into the workflow tools it calls. On 17 November 2026 Hokai opened the public repository directly: the licence file lists the two extra conditions, the release notes describe that tracing change, and the newest commit was timestamped that same morning.
Skip it when the job is moving data between business apps with an occasional AI step: n8n and Make are the more direct fit there. A team that wants a code-first library instead of a canvas will find LangChain closer to how it already works, and Microsoft shops can compare Copilot Studio. Flowise, which describes itself as an open source agentic systems development platform, is the nearest like-for-like rival and has no Hokai record yet. The agent builders guide lays out the wider field.
Product facts on this page (pricing, plans, integrations, compliance) are read from LangGenius, Inc.'s own pages and documentation. The verdict is Hokai's reading of them, not the vendor's claim.