Baseline version: GitHub langgenius/dify v1.17.1
Estimated half-life: Quarterly (AI app orchestration and protocol evolution moves fast โ€” keep an eye on plugin and MCP developments)
Nature: Personal research notes, sharing what I found
Motivation: I hadnโ€™t heard much about Dify recently and wondered โ€” did it fall behind, or is it quietly growing somewhere else? So I spent some time digging through public sources and wrote down what I found.


1. A Feeling: Why Does It Seem Like โ€œNo Oneโ€™s Talking About Itโ€?

At least in my own circles, between late 2024 and 2026, I rarely saw anyone discussing Dify. My first reaction was: is it dead?

But after going through the materials, I think this โ€œdisappearanceโ€ likely comes more from our own shifting attention and Difyโ€™s own strategic pivot:

2023 ~ Mid-2024 (Geek Carnival)                    2025 ~ 2026 (Divergence)
[Individual devs / Prompt engineers]             [Code-first camp (geeks/backend)]
      โ”‚                                                โ”‚
      โ–ผ                                                โ–ผ
Loved GUI drag-and-drop bots / simple KBs    Shifted to Coding Agents (Cursor/Claude Code)
Dify, FastGPT dominating social media         Shifted to code-native state machines (LangGraph/PydanticAI)
                                               
                                               [Dify's actual pivot]
                                                       โ”‚
                                                       โ–ผ
                                           International expansion (Japan / North America / Europe)
                                           Deep enterprise (compliance / private deployment / global giants)

1.1 The Geek Worldโ€™s Attention Shifted

  • In 2023 to early 2024, the LLM ecosystem was still figuring out โ€œhow to connect models to knowledge bases and prompts.โ€ Visual DAG drag-and-drop was the lowest-barrier, most demo-friendly approach;
  • Later, the tech communityโ€™s attention clearly shifted toward Coding Agents (Cursor, Claude Code, etc.) and code-first state machine frameworks (LangGraph, PydanticAI, etc.). With code, Git management and unit testing feel more native. People on social media stopped posting โ€œwiring diagrams.โ€

1.2 Difyโ€™s Own้‡ๅฟƒ Shifted

  • From public information, the LangGenius team deliberately reduced their domestic geek marketing volume and poured energy into overseas markets and large enterprise customers;
  • The fastest-growing market is overseas (especially Japan and North America). While domestic social media discussion cooled, Dify was holding hundreds-person summits in Japan, joining AWS official channels, and completing private deployments inside multinational enterprise networks.

2. Publicly Available Business Data

Here are the numbers I found from public sources, for reference:

2.1 Funding

  • $30M Series Pre-A (March 2026):

    • Lead: HSG (HongShan);
    • Followers: GL Ventures, 5Y Capital (returning investor), Mizuho Leaguer Investment, Alt-Alpha Capital, and NYX Ventures;
    • Valuation approximately $180M.

2.2 Open Source & Community

  • GitHub Stars: ~158K (global Top 60), 1000+ contributors, 861 recent commits, 168 releases;
  • Global self-hosted installations: Over 1.4 million machines;
  • Enterprise customers: Publicly disclosed 280+, including Maersk, Volvo, Novartis, ThermoFisher;
  • Global presence: Offices in San Francisco Bay Area, Tokyo, Shanghai, and Suzhou;
  • Japan market: Established the โ€œJapan Association for Difyโ€ (ไธ€่ˆฌ็คพๅ›ขๆณ•ไบบ Dify ๅไผš, founded September 2025 with NTT Data and others), and became an AWS Global Partner (awarded 2025 AWS Social Impact Partner of the Year).

3. Product & Technical Evolution (v0.x to v1.17+)

From the version history, Difyโ€™s product form has clearly moved well beyond the early โ€œvisual prompt tunerโ€ stage. The underlying tech stack has gone through multiple deep refactors:

                    โ”Œโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”
                    โ”‚       Dify Marketplace          โ”‚ (Plugin ecosystem / Community assets)
                    โ””โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”ฌโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”˜
                                     โ”‚
โ”Œโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”    โ”‚    โ”Œโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”
โ”‚        New Agent App          โ”‚    โ”‚    โ”‚       Workflow Studio         โ”‚
โ”‚ (Autonomous reasoning /      โ”‚โ—„โ”€โ”€โ”€โ”ดโ”€โ”€โ”€โ–บโ”‚ (Visual orchestration /       โ”‚
โ”‚  Skills / Code sandbox)      โ”‚         โ”‚  Branching / Dynamic routing) โ”‚
โ””โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”ฌโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”˜         โ””โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”ฌโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”˜
                โ”‚                                         โ”‚
                โ””โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”ฌโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”˜
                                    โ–ผ
       โ”Œโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”
       โ”‚   MCP Gateway (Workflow-as-MCP-Server / ClientPassthru)โ”‚
       โ””โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”ฌโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”˜
                                    โ–ผ
       โ”Œโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”
       โ”‚     Core Engine (Async non-blocking / Sandbox / RBAC)   โ”‚
       โ””โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”˜

3.1 Full MCP (Model Context Protocol) Support

  • Workflow-as-MCP-Server: Complex business workflows orchestrated inside Dify can be exposed as standard MCP Servers with one click. Whether itโ€™s Claude Desktop, an IDE, or an external Coding Agent, they can all call Dify workflows as standard MCP tools;
  • Runtime dynamic header pass-through: Supports injecting placeholder authentication at request time (e.g., {{request.headers.X-Custom-Auth}}), solving the security isolation and identity authentication challenge for multi-tenant enterprises calling third-party tools.

3.2 Plugin System & Marketplace โ€” Fully Decoupled

  • No more forking source code: In the early v0.x days, community developers wanting to integrate a new model or private internal system often had to fork the repo, modify core code, or even maintain their own version. v1.0 (early 2025) restructured the plugin architecture โ€” tools and models are fully decoupled as independent plugins;
  • Marketplace & plugin security governance: Introduced plugin security levels, signature verification, and sandbox isolation. Enterprise users can deploy private plugin sources on internal networks, balancing extensibility with compliance.

3.3 Agent Architecture Rebuild: New Agent App & Sandbox Execution

  • Chat-driven Agent building: Supports generating and debugging agents through natural language conversation, automatically extracting and distilling structured, reusable Skills;
  • Built-in secure sandbox (DifySandbox): Provides isolated Python/Shell execution sandboxes for Code nodes and Tool nodes, supporting secure data transformation, logic decisions, and local script execution;
  • Bidirectional workflow nesting: You can embed Agent nodes inside Workflows (letting the agent make local autonomous decisions), or have Agents trigger specific Workflows.

3.4 Enterprise Core & Compliance

  • Async non-blocking execution engine: Database writes and long-chain calls are async-ified. Multi-branch parallel execution time is optimized from โ€œsimple sum of all branchesโ€ to โ€œdepends on the longest branch,โ€ significantly reducing timeout interruption rates;
  • Enterprise governance certifications: Two consecutive years of SOC 2 Type II, ISO 27001:2022, and GDPR compliance certifications, with full audit log tracking and fine-grained RBAC permission controls.

4. My Attempt to Understand Its Positioning

Hereโ€™s my personal understanding from research โ€” may not be fully accurate, just for reference:

DimensionDify (v1.17+)Code-First Frameworks (LangGraph etc.)Pure SaaS Competitors (Coze)
Audience I observeCross-functional teams (Business + IT)Senior backend, AI full-stack engineersIndividual content creators, non-technical ops
Deployment flexibilityOpen-source community / Container private / Multi-cloudPure code library, deploy as you wishDeeply tied to platform ecosystem
Debugging experienceGood node-level historical I/O inspection UIRelies on code breakpoints, LangSmith, etc.Relies on platform logs
Extension approachPlugin marketplace + MCP + Custom code nodesPure code, unlimited extensionLimited to platformโ€™s integration library
Engineering collaborationWorkflow DSL import/export, weaker Git diffingNative Git + unit testsIn-platform version management

My Personal Scenario Understanding

                    Where does your scenario lean?
                              โ”‚
         โ”Œโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”ดโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”
         โ–ผ                                         โ–ผ
ใ€Business-IT collaboration neededใ€‘        ใ€Hardcore geek / Algorithm experimentsใ€‘
โ€ข Need to visually present flows to biz    โ€ข Logic heavily depends on complex recursion, dynamic backtracking
โ€ข Need out-of-box RAG, permissions & audit โ€ข Heavily dependent on Git branching & automated testing
โ€ข Need multi-model aggregation & MCP GW    โ€ข More comfortable with pure code, averse to UI wiring
         โ”‚                                         โ”‚
         โ–ผ                                         โ–ผ
   Dify might be a good fit             LangGraph / custom code might be handier

This is just my rough personal understanding. Real-world selection is far more complex. Many teams probably use a mix.


5. Monetization & Team Status

This is what I was curious about: Is it making money? Is the team still alive?

5.1 Pricing (Publicly Available)

PlanPriceNotes
SandboxFree200 message credits/month, 1 person, 5 Apps
Professional$59/mo ($590/yr)5,000 credits/month, 3 people, 50 Apps
Team$159/mo ($1,590/yr)10,000 credits/month, 50 people, 200 Apps
EnterpriseCustom quoteSSO/SCIM, private deployment, SLA
AWS Marketplace EnterpriseFrom $150,000/yrListed on AWS Marketplace at this price
Self-hosted communityFree (open-source)Users pay for infra + LLM API costs

5.2 Revenue: Not Disclosed, but Profitable

Dify is a pre-Series-A company and hasnโ€™t publicly disclosed specific revenue figures. But from a Forbes interview (2026-02-24), one key data point is confirmed:

โ€œThe company built around it employs 100 people, is profitable and serves more than 280 enterprise customers.โ€

โ€” Profitable, 100-person team. Specific revenue unknown.

Rough estimate (purely personal speculation, no public basis):

  • Of 280+ enterprise customers, assuming some go through Enterprise (AWS lists at $150K+/yr), plus Cloud subscription revenue
  • 100-person team + four offices, and already profitable โ€” means revenue covers costs
  • Notably, Difyโ€™s official careers page states โ€œOver 60% of our codebase is built by LLMsโ€ โ€” R&D efficiency is likely far higher than traditional teams

5.3 Layoffs: None Found

In Layoffs.fyi and various tech layoff trackers, no Dify/LangGenius layoff records were found. Against the backdrop of massive tech industry layoffs in 2025-2026, this is a positive signal.

5.4 Hiring: Still Recruiting

Difyโ€™s official careers page (join.dify.ai) as of 2026-09-18 still had 4 open positions:

  • LTS Backend Software Engineer (China)
  • Backend Development Intern (China)
  • UI/UX Designer (China)
  • Developer Relations Intern (Community Operations)

Not a large hiring scale, but shows the team is actively running, not frozen.

5.5 Development Activity: Very Active

From GitHub data:

  • Recent versions: v1.17.1 (2026-09-10), v1.15.0 (2026-06-25), v1.14.2 (2026-05-19) โ€” maintaining at least one release per month
  • Weekly activity: ~615 new Stars, 132 Pushes
  • Issue activity: New Issues still being filed as recently as 2026-10-13, with active Maintainer responses
  • Contributors: 1000+, 23.9K Forks

6. Personal Takeaways

  1. Dify didnโ€™t disappear โ€” it just went somewhere I canโ€™t easily see: From the domestic geek communityโ€™s perspective, itโ€™s certainly gone โ€œquiet.โ€ But from the overseas enterprise market, Japan ecosystem, and AWS channels, it seems to be running fast on a different track;
  2. Low-code orchestration probably still has its place in enterprise scenarios: While Code-First is the geek aesthetic mainstream, many real business scenarios need cross-department collaboration, visual auditing, and permission controls โ€” these may not be what pure code frameworks do best;
  3. 100 people, profitable, 60%+ code built by LLMs โ€” this combination is interesting in itself. It might be an early example of AI-native development: using AI to build AI tools, a small team leveraging massive scale;
  4. Missing specific revenue numbers is the biggest gap in this research. Understandable for a pre-Series-A company. But from โ€œprofitableโ€ and the $150K/yr AWS Enterprise price tag, the business model seems to be working at least.

All of the above is my personal understanding from research โ€” inevitably with some bias. If you see different information, Iโ€™d welcome the correction.