# Dify

> The visual platform for LLM apps and agentic workflows — canvas-built chatflows, RAG pipeline, agent nodes with 50+ tools, and LLMOps, self-hosted via Docker.

Dify (~145k stars) is the visual answer to LLM app building: a workflow canvas for chatflows and agents, a built-in RAG pipeline from ingestion to retrieval, agent nodes with 50+ tools, hundreds of models via any provider, a prompt IDE, and LLMOps — self-hosted with one docker compose or cloud freemium. License caveat: it's a modified Apache-2.0 with conditions.

Website: https://dify.ai

Dify is the heavyweight of visual LLM-app platforms — ~145k GitHub stars and a self-description that tracks the era: "production-ready platform for agentic workflow development." Its proposition: everything between a model API and a shipped AI product — RAG, agents, prompts, ops — on one canvas, deployable with one `docker compose up`.

## Highlights

- **Visual workflow canvas** — chatflows and agentic workflows assembled from nodes, debugged visually, versioned.
- **RAG pipeline included** — ingestion through retrieval (PDF/PPT and common formats out of the box) without assembling a vector stack by hand.
- **Agent nodes** — Function-calling or ReAct agents with 50+ built-in tools (search, image generation, WolframAlpha…) plus a plugin marketplace of strategies.
- **Every model, one panel** — hundreds of proprietary and open models across dozens of providers, including any OpenAI-compatible endpoint (so [local models](/tools/ollama) plug in).
- **Prompt IDE + LLMOps** — compare models on prompts, then log, annotate, and improve from production traffic.
- **Backend-as-a-Service** — your Dify apps expose APIs, so the visual build embeds into real products.

## In an AI-assisted workflow

```bash
git clone https://github.com/langgenius/dify.git && cd dify/docker
cp .env.example .env && docker compose up -d    # → http://localhost/install
```

The fit: teams that want the AI app *built* more than they want to own its plumbing — prototypes that become products, internal tools, and the "let domain experts iterate on prompts" workflow code-first frameworks can't offer.

> [!WARNING]
> Read the license before building a business on it: Dify's modified Apache-2.0 forbids **multi-tenant** operation without a commercial license and requires keeping frontend branding. Internal single-tenant self-hosting is the cleanly-free case; SaaS-on-Dify is a conversation with LangGenius.

## Good to know

Cloud is freemium (a sandbox tier with message credits, then per-workspace plans); self-hosting is a real multi-service stack (API, worker, Postgres, Redis, vector DB) — budget ops accordingly. v1.0 landed February 2025 with the plugin architecture; the 1.x line iterates steadily. Versus the automation-first alternative: [n8n vs Dify](/guides/comparisons/n8n-vs-dify); versus code-first frameworks: [Agent Frameworks in 2026](/guides/concepts/agent-frameworks-2026).

---

_Source: https://agentscamp.com/tools/dify — Tool on AgentsCamp._