Skip to content

agentsope/SkillAlchemy / skills/agentsop-dify

agentsop-dify

Verified

SOP for building LLM applications on Dify — visual workflow + chatflow + agent + RAG knowledge base + plugin marketplace + observability, self-hostable. Use when shipping LLM apps fast with a "no-code to pro-code" gradient, especially when non-engineers need to co-author the flow.

references/R5-ecosystem-context.md

View on GitHub
# R5 — Ecosystem Context

## Landscape Map

Dify sits in the **"open-source LLM application platform"** category. Competitors fall into several buckets:

```
                          full-stack platform
                                  ↑
              [Dify] ←─────── [Coze (ByteDance, hosted only)]
                ↑                  ↑
                |                  |
        visual ─┼──────────────────┼─ code
                |                  |
            [Flowise]          [LangFlow]
            [n8n + LLM]            ↑
                                   |
                          [LangChain / LangGraph / LlamaIndex]
                                   ↓
                          code framework, no UI

  Specialized: [RAGFlow] (RAG depth) | [CrewAI] (role-based agents)
```

## Pairwise Comparisons

### Dify vs Flowise
[blog.elest.io/dify-vs-langflow-vs-flowise-which-open-source-llm-app-builder-actually-ships-to-production], [toolhalla.ai/blog/dify-vs-flowise-vs-langflow-2026]

| Dimension | Dify | Flowise |
|---|---|---|
| Min RAM | 4 GB | 1 GB |
| App types | 5 | basically chatbot |
| Multi-tenant workspaces | yes | no |
| Knowledge base | full pipeline | basic |
| Production readiness | high | low-medium |
| Best for | LLM SaaS, internal multi-team product | quick chatbot demo |

Quote: "If your use case is 'chatbot with document retrieval' and you want the fastest, cheapest path to production, Flowise delivers without the complexity overhead."  [blog.elest.io/...]

### Dify vs LangFlow
| Dimension | Dify | LangFlow |
|---|---|---|
| Ecosystem coupling | own framework | LangChain-bound |
| Code export | DSL YAML (configuration) | Python code (executable) |
| Custom Python nodes | yes (Code node, sandboxed) | yes (richer, integrated with LangChain) |
| Multi-tenant | yes | weaker |
| Path forward when ceiling hit | externalize logic | export to Python, evolve |

Quote: "Langflow is for teams that need power now and will need more later. The LangGraph integration and custom Python nodes mean you won't outgrow it as fast."  [blog.elest.io/...]

### Dify vs Coze (扣子)
[jimmysong.io/blog/open-source-ai-agent-workflow-comparison]

| Dimension | Dify | Coze |
|---|---|---|
| License | open source | ByteDance hosted only |
| Self-host | yes | no (Coze global) / limited (CN) |
| Ecosystem integration | model-agnostic, 100+ providers | ByteDance ecosystem (Feishu/Lark, Doubao) |
| Audience | global, dev-leaning | rapid no-code bots, ByteDance-aligned teams |

When Coze wins: you want managed bots in the Feishu/Lark ecosystem with zero ops.
When Dify wins: open source, multi-region, full control.

### Dify vs RAGFlow
[sider.ai/blog/ai-tools/dify-vs-ragflow-which-rag-platform-should-you-build-on-in-2025], [learnwithparam.com/blog/batteries-included-rag-platforms-dify-ragflow-onyx]

| Dimension | Dify | RAGFlow |
|---|---|---|
| Core mission | LLM app platform with RAG | RAG engine with thin UI |
| Document parsing | standard | deep (tables, layout, KG-RAG) |
| App orchestration | visual workflow + agents | minimal — backend engine |
| Frontend / web UI | polished | thin |
| When chosen | rapid app assembly | RAG quality is the bottleneck |

Quote: "Dify is the only tool that gives you data ingestion, RAG, an API, and a polished, shareable web UI in one click… RAGFlow is not a full app builder but a highly specialized, 'data-first' RAG engine obsessed with retrieval."  [learnwithparam.com/blog/...]

### Dify vs n8n (with AI nodes)
[zhuanlan.zhihu.com/p/1898775808660710158], [zedyer.com/iot-knowledge/n8n-vs-dify-ai-workflow]

| Dimension | Dify | n8n |
|---|---|---|
| Core mission | AI-native app platform | general workflow automation |
| Non-AI integrations | sparse (50+ AI tools) | 400+ (Salesforce, GSuite, Stripe…) |
| AI as primary | yes (workflow + chatflow + agents) | yes via AI nodes but secondary |
| RAG built-in | yes (knowledge base + pipeline) | DIY |
| License | Apache-like (some restrictions) | "fair-code" (SaaS restrictions) |

Rule of thumb: AI-first → Dify; automation-first with AI sprinkled → n8n.

### Dify vs LangGraph
| Dimension | Dify | LangGraph |
|---|---|---|
| Layer | platform | code framework |
| Abstraction | visual DAG | stateful graph in Python |
| State machine | limited | rich (interrupts, time-travel, checkpointing) |
| Human-in-loop | not supported | native (interrupt + resume) |
| UI / API / Auth | built-in | DIY |
| Audience | mixed teams | engineering teams |

Common hybrid: **Dify for frontend/RAG/auth + LangGraph for core agent logic** behind HTTP.

### Dify vs LlamaIndex
| Dimension | Dify | LlamaIndex |
|---|---|---|
| Layer | application platform | data/RAG framework |
| RAG depth | good (and abstracted) | extreme (data ingestion, indices, query engines) |
| App scaffolding | full | none |
| Audience | app builders | data engineers, RAG specialists |

Not direct competitors. Pattern: LlamaIndex builds the retrieval, exposed as a service, Dify's External Knowledge points at it.

### Dify vs CrewAI
| Dimension | Dify | CrewAI |
|---|---|---|
| Paradigm | platform with workflow + agent nodes | role-based multi-agent code framework |
| Visual canvas | yes | no |
| Production scaffolding | yes (UI/API/Auth/Monitoring) | no |
| Multi-agent role play | possible via multiple Agent Nodes | native |

Different worlds. Combine: Dify hosts the UI/API; CrewAI runs the multi-agent crew behind HTTP.

## Why Dify Reached 50k+ Stars

1. **It compresses scaffolding** — auth, UI, API, vector store, model providers, logs, versioning, web app — all in one box. Saves weeks per project.  [memo.d.foundation/breakdown/dify]
2. **Cross-role friendly** — PMs/ops draw chatflows; engineers drop into Code nodes; the artifact is shared.  [docs.dify.ai/en/use-dify/getting-started/key-concepts]
3. **i18n strength** — strong Chinese-language community (LangGenius is China-rooted), but full English docs + global user base. Bridges Asia/West better than competitors.  [jimmysong.io/...]
4. **Active production hardening** — v1.0 introduced plugins, v1.9 redesigned the graph engine to queue-based, weekly-ish releases.  [github.com/langgenius/dify/discussions/26138]
5. **Honest about ceilings** — team transparently lists limitations and addresses them in roadmap.  [memo.d.foundation/breakdown/dify]

## When Dify Has Outgrown Its Usefulness

- Single graph > 50 nodes (canvas lag, comprehension collapses)
- Need pause-wait-resume or rich time-travel
- QPS > 10/pod sustained
- Extreme RAG experimentation
- Core value is agent autonomy itself (not the surrounding platform)

In these cases Dify retreats to "frontend + monitoring + simple RAG repository," and core logic moves to LangGraph / Temporal / custom services.

## Strategic Read

Dify's bet: **the bottleneck in LLM app productionization is scaffolding, not the LLM call**. By packaging the scaffolding, it captures the "ship LLM apps fast" segment. The trade-off is expressive ceiling — for teams whose bottleneck is actually expressive depth (complex agents, state machines, sub-second latency), Dify is overhead.

Adoption follows: enterprises with mixed-role teams (PM + ops + eng) and "internal AI hub" needs love Dify; pure ML research teams stick with code frameworks.

## Real-World Adopters (named)

- **Maersk** — document review, knowledge copilots
- **ETS** — customer support automation
- **Samsung Electronics** — vLLM + Dify in air-gapped infrastructure (presented at Dify Studio Korea Meetup)
- Banks deploying internally as governed AI hub
Source: [github.com/langgenius/dify], referenced summaries.
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 471kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 322kUpdated Claude CodeCodexVerifiedInstall