DifyDify
Dify 是国产出海的 agentic 工作流平台:把提示词逻辑做成可见、可测、可交接的 Workflow/Chatflow 执行路径, 知识库管道与 Marketplace 集成是一等公民。许可是「Apache 2.0 + 两条附加条件」,多租户限制最该先读。
Dify is the China-origin, globally-facing agentic workflow platform: prompt logic becomes visible, testable, handoff-ready Workflow/Chatflow execution paths, with knowledge pipelines and Marketplace integrations as first-class citizens. The license is “Apache 2.0 + two additional conditions” — the multi-tenant restriction is the first thing to read.
适合与不适合When to use it, when not to
适合:要自托管、要「产品/法务/工程在同一画布上对齐」的团队;知识库+agent 一体交付。 不适合:拿源码做多租户 SaaS 售卖(商业授权线);纯代码编排(去 LangGraph 等 framework 大类)。
A fit for: teams wanting “product/legal/engineering aligned on one canvas”, self-hostable China-origin stacks, knowledge-base + agent delivered together. Not a fit for running multi-tenant SaaS from the source (commercial license line); pure code orchestration (go to LangGraph and the framework category).
中文生态China availability 国产 / 国内可直接用 / 备案状态 · 快照domestic / direct access / CAC filing · snapshot 2026-10-21
- 出品属地Origin
- 国产domestic LangGenius(国产团队出海;GitHub 主体 langgenius)。LangGenius (Chinese-origin team, global-facing; GitHub org langgenius).
- 国内使用Direct access
- 大陆可直连direct OK dify.ai / docs.dify.ai 直连实测 200(2026-10-21)。dify.ai / docs.dify.ai measured 200 direct (2026-10-21).
- 备案状态CAC filing
- 备案未核验filing unknown:未核验到 Dify 在网信办公示名单的对应备案主体(2026-10-21 检索);其境内是否以备案形态服务未核验,标 unknown。no matching filing entity found in the CAC registry (searched 2026-10-21); whether it serves domestically under a filing is unverified — marked unknown.
备案信息以网信办《生成式人工智能服务已备案信息》公示名单为准(按批次滚动更新,本站只抄编号不解读);「大陆可直连」是本站从大陆网络直连实测的结果,不是厂商承诺。GitHub 类站点当日实测有波动,口径写在依据里。Filing status is quoted from the CAC public registry (updated batch by batch; this site copies the number and does not interpret it). "Direct access" is this site's own reachability measurement from a mainland network, not a vendor promise. GitHub-hosted sites fluctuated on the measurement day — the note says so.
固定坐标系Fixed coordinate system
8 个维度dimensions, 与同分区其他对象逐项可比each comparable item by item against others in the same section.
- 模型与开放条件Model & openness
平台经「模型供应商」插件接各家 LLM(OpenAI/Anthropic/Gemini/本地 Ollama 等, Marketplace 统一装);官方称模型可自由切换。具体供应商全清单未逐条核验。
The platform reaches each LLM via "model provider" plugins (OpenAI/Anthropic/Gemini/local Ollama etc., installed uniformly through the Marketplace); officially models can be switched freely. The full provider list has not been verified item by item.
- 运行位置Where it runs
社区版源码开放、Docker 自托管(官方首页口径:单条命令部署、完整 Agent runtime 与 Workflow 编辑器);另有 Dify 官方云服务。自托管数据在你自己的环境。
Community edition source is open with Docker self-hosting (official homepage framing: one-command deployment, full agent runtime and Workflow editor); there is also a Dify official cloud service. Self-hosted data stays in your own environment.
- 本地文件Local files
知识库线官方描述为「为文件、网站、在线文档与云盘建数据管道」(抽取/清洗/分块/索引/检索测试); 自托管时文件与索引落在自有实例。
The knowledge-base line is officially described as "data pipelines for files, websites, online docs and cloud drives" (extraction/cleaning/chunking/indexing/retrieval testing); when self-hosted, files and indexes live on your own instance.
- 关机后的任务Tasks after shutdown
应用发布为托管应用/API/嵌入组件/MCP 兼容工具后由平台运行时承载; 自托管实例常开即可持续服务。定时/后台任务形态未逐条核验。
Apps published as hosted apps/APIs/embeddable widgets/MCP-compatible tools are carried by the platform runtime; a self-hosted instance running continuously can serve persistently. Scheduled/background task forms not verified item by item.
- 工具与扩展Tools & extensions
Marketplace 装模型供应商、工具、数据源与 MCP 集成(官方首页口径); 应用可发布为 MCP 兼容工具——既消费 MCP 也产出 MCP。
Marketplace installs model providers, tools, data sources and MCP integrations (official homepage framing); apps can be published as MCP-compatible tools — it both consumes and produces MCP.
- 上下文与记忆Context & memory
Agent 官方描述含「记住上下文」;知识库检索是官方主打的上下文供给线。 跨会话记忆的具体机制未逐条核验。
The official Agent description includes "remembering context"; knowledge-base retrieval is the flagship official context-supply line. The concrete cross-session memory mechanism has not been verified item by item.
- 权限与限制Permissions & limits
企业线官方列 SSO/SAML、RBAC、审计日志、SOC 2 Type II + ISO 27001(首页口径); 社区版权限模型未逐条核验。许可边界(多租户限制)见 license 字段。
The enterprise line officially lists SSO/SAML, RBAC, audit logs, SOC 2 Type II + ISO 27001 (homepage framing); the community edition's permission model has not been verified item by item. License boundary (multi-tenant restriction) see the license field.
- 适合什么任务Best for
适合:要「提示词逻辑变成可见执行路径」的团队、要自托管的国产出海栈、知识库+agent 一体。 不适合:要源码做多租户 SaaS(需商业授权);要纯代码框架(归 framework)。
Fits: teams that want "prompt logic turned into a visible execution path", self-hosted Chinese-origin global-facing stacks, knowledge-base + agent delivered together. Does not fit: running multi-tenant SaaS from the source (needs commercial licensing); pure-code frameworks (belongs to framework).
头号误解Top misconception
- 把「Apache 2.0 本质」读成无限制——多租户与 LOGO 两条附加条件是硬线Reading “Apache 2.0 at its core” as unlimited — multi-tenant and LOGO conditions are hard lines
- 把云服务与自托管社区版的能力混同——企业项(SSO/RBAC/审计)在商业线Conflating the cloud service with the self-hosted Community edition — enterprise items (SSO/RBAC/audit) sit on the commercial line
- 只把它当聊天机器人搭建器——官方定位是 agentic 工作流平台(Workflow/Chatflow 双形态)Treating it as only a chatbot builder — the official positioning is an agentic workflow platform (Workflow/Chatflow)
未知项清单Known unknowns
- 模型供应商插件的全清单(Marketplace 未逐条核验)
- 社区版权限模型细节(RBAC 是否部分下沉社区版未核验)
- 定时/后台任务形态的官方边界
- 境内备案形态(标 unknown 的原因见 cn 块)
- The full model-provider plugin catalog (Marketplace not verified item by item)
- Community-edition permission-model details (whether RBAC partially lands in the community edition is unverified)
- The official boundary of scheduled/background task forms
- Domestic filing form (reason for marking unknown: see the cn block)
证据来源Evidence sources
判断可回到以下一手源复核。本站核验日Every judgement can be rechecked against the primary sources below. Verified on 2026-10-30,内容更新日content updated on 2026-10-21。
| 类型Type | 名称Name | 链接Link |
|---|---|---|
| docs | langgenius/dify · LICENSE(官方仓,附加条件全文,2026-10-21 直抓核验)langgenius/dify · LICENSE (official repo, full additional conditions, fetched & verified directly 2026-10-21) | https://github.com/langgenius/dify/blob/main/LICENSE |
| docs | Dify · 官方首页(Workflow/Agent/知识流水线/Marketplace/发布线;社区版 148K+ Star 口径)Dify · official homepage (Workflow/Agent/knowledge pipelines/Marketplace/publishing lines; community edition 148K+ Star framing) | https://dify.ai/zh |
| repo | langgenius/dify · 仓库(159,137★,最近推送 2026-10-21,GitHub API 核验)langgenius/dify · repository (159,137★, last push 2026-10-21, verified via GitHub API) | https://github.com/langgenius/dify |
| docs | Dify · 官方文档站Dify · official docs site | https://docs.dify.ai |
| changelog | langgenius/dify · Releases(版本迭代记录)langgenius/dify · Releases (version iteration records) | https://github.com/langgenius/dify/releases |
实测记录Measured runs
本档未跑工作流搭建实测;结论只覆盖平台能力与权限边界。
This site has not completed measured runs. The closest test protocol is the code-side protocols/framework/_protocol.md (the tool-calling and credential-boundary sections are reusable; no workflow-platform-specific protocol exists on this site yet).
许可的两条附加条件The license's two additional conditions
官方 LICENSE 原文要点(转述):未经 Dify 书面许可,不得用源码运营多租户环境(一个 workspace 即一个租户); 使用其前端不得移除或修改控制台/应用的 LOGO 与版权信息(不用前端则不受此限)。 其余按 Apache 2.0。交互设计另受外观专利保护。
Key points of the official LICENSE text (paraphrased): without written permission from Dify, you may not operate a multi-tenant environment from the source code (one workspace = one tenant); when using its frontend you may not remove or modify the console/app LOGO and copyright information (frontend-free use is exempt). Everything else follows Apache 2.0. The interaction design is additionally covered by a design patent.
相关条目Related entries
- LangGraphLangGraph画布 vs 代码对照:同一件事(可控的 agent 执行路径),一边拖节点一边写图。Canvas vs code contrast: the same thing (a controllable agent execution path) — one side drags nodes, the other writes a graph.
本页由This page is generated from the ai-handbook 数据层生成(CC BY 4.0)。data layer (CC BY 4.0).
方法论与坐标系定义见仓库内For the methodology and coordinate definitions, see METHODOLOGY.md.