Dify

What is Dify?

Dify is an open-source platform for building production LLM applications and agents. On a visual canvas you assemble workflows that retrieve data, call tools, make decisions, ask for human input and complete tasks, then ship them as hosted apps, API endpoints, embeds or MCP-compatible tools. It bundles the pieces most teams otherwise stitch together: a RAG knowledge pipeline, an agent runtime, model management across hundreds of LLMs, and observability. Because the core is open-source (an Apache-2.0-derived licence, 149K+ GitHub stars), you can self-host it with Docker or use Dify’s managed cloud. Built by LangGenius, founded in 2023 in the San Francisco Bay Area (with satellite teams in Tokyo, Shanghai and Suzhou) by CEO Luyu Zhang and co-founder/Chief Engineer John Wang, the company raised a $30M Series Pre-A round led by HSG in March 2026, with GL Ventures, Alt-Alpha Capital, 5Y Capital, Mizuho Leaguer Investment and NYX Ventures also participating.

What does Dify do?

Dify gives you a visual Workflow Studio where you wire nodes, model calls, tool calls, knowledge retrieval, conditionals, code and human-input steps, into an app that behaves like an agent. Its RAG pipeline handles the whole knowledge path: ingest files, websites, docs and drives, then extract, clean, chunk and index them for retrieval. Agent nodes use function-calling or ReAct-style reasoning and can draw on 50+ built-in tools plus anything you add from the plugin marketplace. It’s model-agnostic, integrating hundreds of LLMs from dozens of providers (OpenAI, Anthropic, Google, plus open and local models), so you can swap or mix models per node. Once built, an app publishes as a hosted web experience, a REST API (Backend-as-a-Service), an embeddable widget, or an MCP tool other agents can call. An observability layer logs runs, captures feedback and annotations, tracks usage, and integrates with LangFuse and Arize Phoenix for tracing. You can run all of this three ways: the free open-source Community Edition via Docker on your own infrastructure, Dify’s managed cloud (a free Sandbox plus paid plans), or a self-hosted Enterprise edition with SSO, RBAC and audit logs. The open core plus self-host option is the main reason teams pick Dify over closed SaaS builders.

Key features

• Visual Workflow Studio: Wire model calls, tools, retrieval, logic and human-input nodes into an agentic app on a canvas.
• Built-in RAG pipeline: Ingest files, sites, docs and drives, then extract, clean, chunk and index them for retrieval.
• Model-agnostic: Integrates hundreds of LLMs from dozens of providers, including open and local models, swappable per node.
• Publish anywhere: Ship apps as a hosted web app, REST API, embed, or MCP-compatible tool.
• Observability: Logs, feedback, annotations and usage analytics, with LangFuse and Arize Phoenix integrations.

What are Dify's use cases?

• Internal RAG assistant: Build a knowledge-grounded support or research agent over your own documents and self-host it for data control.
• Ship an LLM feature fast: Prototype an agentic workflow visually, then expose it as an API endpoint your product calls.

What does Dify integrate with?

• Model providers — OpenAI, Anthropic, Google + open/local models
• LangFuse & Arize Phoenix (observability)
• Plugin/tool marketplace + MCP
• 50+ built-in tools
• REST API / Backend-as-a-Service

Why use Dify?

• Open-source core (149K+ GitHub stars) you can self-host with Docker, no vendor lock-in.
• One platform folds together workflow, RAG, agents, model management and observability.
• Model-agnostic, so you can mix frontier, open and local models per step.

Pros & cons

Dify pricing

• Community EditionFree
• Sandbox (Cloud)Free
• Professional (Cloud)$590 / /year
• Team (Cloud)$1,590 / /year
• EnterpriseCustom

See current pricing on dify.ai ↗Compare Dify alternatives →

Dify review

Dify has become the default open-source answer to "how do we build and ship LLM apps without gluing five tools together." Its visual Workflow Studio, built-in RAG pipeline, model-agnostic runtime and observability cover the whole path from prototype to production, and with a large open-source community and a Docker self-host option you get real data control and no lock-in. The nuances matter: the licence is Apache-2.0-derived with extra conditions rather than pure open source, self-hosting means you own the ops burden, and the managed cloud meters usage in message credits that heavier apps will outgrow. For deeply bespoke agent logic a code-first framework can still be more flexible. Pick Dify if you want a self-hostable, all-in-one platform for RAG and agent apps and have the technical chops to run it; skip it if you want a hands-off managed "AI employee" with no infrastructure to think about.

The open-source, self-hostable all-in-one for LLM apps, the go-to when data control and no lock-in matter more than a managed experience.

Frequently asked questions