Dify

The Dify team will be at Ai4 - Artificial Intelligence Conferences in Las Vegas from August 4-6. Come visit us at booth 1567 to see how companies are using Dify to build, govern, and run AI agents in production. Check out live demos, chat with our team about how you're using AI today, and stop by to say hello. We'll also be presenting a session with Maersk, "From Pilot to Production: A Case Study with Maersk," on August 5, 4:05–4:25 PM PDT. We look forward to seeing you there.

When we shipped difyctl a few days ago, we said it was Dify’s answer to a world where terminals, scripts, and coding agents are becoming the primary interface for getting work done. This follow-up shows what that looks like in practice. AI coding agents can write code, but real business processes also need structured workflows, organizational knowledge, human approval, and reliable execution. That’s where Dify comes in. With difyctl, compatible AI agents can invoke existing Dify workflows directly from the terminal. The agent decides when to trigger a workflow, Dify executes the business logic, and Human Input keeps people involved whenever approval is required. Instead of jumping between tools, a complex process becomes one instruction and one review. Full walkthrough 👇 https://lnkd.in/gDcbvr9X #Dify #CLI #Codex #ClaudeCode #Antigravity

Cutting AI costs is easy. Making sure every dollar of AI spend is actually doing something worth doing is much harder. This is the real challenge behind enterprise AI adoption. Any team can slash their bill by downgrading models or throttling usage. But if outputs quietly degrade, so does everything the AI was supposed to deliver. Real cost control isn't about spending less. It's about spending with intention. Our teammate Anne Zhu wrote a great breakdown on how Dify builders think about this. One line from the piece captures it well: "𝗔𝗜 𝗰𝗼𝘀𝘁𝘀 𝗱𝗼𝗻'𝘁 𝘀𝗰𝗮𝗹𝗲 𝘄𝗶𝘁𝗵 𝘂𝘀𝗮𝗴𝗲. 𝗧𝗵𝗲𝘆 𝘀𝗰𝗮𝗹𝗲 𝘄𝗶𝘁𝗵 𝘂𝗻𝗼𝗽𝘁𝗶𝗺𝗶𝘇𝗲𝗱 𝗮𝗿𝗰𝗵𝗶𝘁𝗲𝗰𝘁𝘂𝗿𝗲." The full piece walks through four practical patterns: ▪️ Route each task to the right-sized model, saving on routine work without giving up quality where it matters ▪️ Use Annotation Reply to cache high-quality answers, so repeat questions don't get billed as new ones ▪️ Self-host on your own infrastructure to trade variable per-token pricing for predictable capacity ▪️ Track cost, latency, and quality per node, so every optimization is grounded in real data instead of gut feel The teams that get the most out of AI aren't the ones spending the least. They're the ones designing every workflow to be worth what it costs. Read the full piece 👇 https://lnkd.in/g4JYFEjm #Dify #LLM #LLMOps #EnterpriseAI

Building AI applications on Dify often means switching between models: DeepSeek for reasoning, Kimi for long context, Qwen, MiniMax, or GLM for everything else. That usually means juggling multiple API keys and reconfiguring your workflows every time. The new Qubrid AI plugin on Dify Marketplace makes that much simpler: 🔹 One plugin, one API key 🔹 Access DeepSeek, Kimi, Qwen, MiniMax, GLM, and more from a single provider 🔹 Switch models within your workflows without re-integration Try it now: https://lnkd.in/gMKxaHuB Learn more: https://lnkd.in/gdCSsT4m

𝗔𝗪𝗦 𝗦𝘂𝗺𝗺𝗶𝘁 𝗧𝗮𝗶𝗽𝗲𝗶 𝟮𝟬𝟮𝟲: 𝗟𝗶𝘃𝗲 𝗮𝘁 𝗧𝗜𝗖𝗖 📍 Yesterday marked a great first day for Dify (powered by LangGenius K.K.) at AWS Summit Taipei 2026. Thank you to everyone who stopped by our booth (P25) to talk about bringing generative AI into real production workflows and audit systems. We had great conversations with teams across finance, manufacturing, retail, and software. Today, July 16, is the final day of the event, and our team is still on-site at TICC, ready to discuss knowledge-base applications, cross-department AI governance, and the feasibility of private deployment. It’s the final day, so come find us at the Dify booth (P25)! #𝗔𝗪𝗦𝗦𝘂𝗺𝗺𝗶𝘁𝗧𝗮𝗶𝗽𝗲𝗶𝟮𝟬𝟮𝟲 #𝗗𝗶𝗳𝘆 #𝗘𝗻𝘁𝗲𝗿𝗽𝗿𝗶𝘀𝗲𝗔𝗜 #𝗟𝗟𝗠𝗢𝗽𝘀 #𝗚𝗲𝗻𝗲𝗿𝗮𝘁𝗶𝘃𝗲𝗔𝗜

𝗗𝗶𝗳𝘆 𝗪𝗼𝗿𝗸𝘀𝗵𝗼𝗽 | 𝗦𝗲𝗼𝘂𝗹 | 𝗝𝘂𝗹𝘆 𝟮𝟯 Spending an afternoon in Seoul on July 23? Join us for a hands-on Dify workshop at COZY Convention Center. In one afternoon, you'll go from zero to a working, production-ready AI app, without writing a single line of code. What we'll cover: 🔹 Dify's core concepts and key features, walked through with real-world examples 🔹 Live build: your first custom AI app, step by step 🔹 Q&A with the Dify team on the ground 📍 COZY Convention Center, Seoul 🎟️ Free Reserve your spot 👇 https://luma.com/307r8yon *Please read the full event details before signing up. #Dify #Seoul #AIWorkshop #NoCode #GenerativeAI

Dify CLI is live, bringing Dify apps directly into your agent workflows. difyctl lets you call Dify apps directly from your terminal. No browser, no clicking through a UI. Agents, local scripts, and CI pipelines can all trigger a workflow with one command. Pair it with a SKILL.md file, and an agent picks up the calling convention in one shot, instead of being re-taught it in every conversation. A few other things shipped alongside it: ▪️ Chain-of-thought, visible everywhere Chat, workflow debug, and now the CLI all surface the model's reasoning separately from its final answer. That matters for debugging, and for teams that need to audit how an AI decision was reached. ▪️ Better Human Input forms, plus Excel image extraction Human Input forms now support dropdown fields and file upload, making approval flows more structured than typed text alone. Knowledge base imports can also extract embedded images from Excel files, so charts and screenshots are preserved for retrieval. ▪️ From the community /create generates a working workflow from a single line typed into search. Credential authorization for API keys and OAuth tokens is now more clearly scoped, built with team and enterprise use in mind. Deep dive: https://lnkd.in/gEJRMiSM Release notes: https://lnkd.in/gsxr4UeR Docs: https://lnkd.in/gyWMwWkT

See you at the AI DevMaster Hackathon with AMD, where the Agentic AI track can be built with Dify. Let's dive in. 🚀

The AMD AI DevMaster Hackathon is here. Build your Agentic AI project with Dify. 🚀 Building an AI agent from scratch usually means fighting infrastructure before you ever get to the interesting part: environment setup, tool integrations, RAG pipelines, local inference tuning. We're partnering with AMD on the AI DevMaster Hackathon to change that. 🔹 Skip the infrastructure boilerplate. Use Dify's visual orchestration to handle reasoning, planning, tool use, and memory.  🔹 Connect external capabilities through Dify's plugin nodes.  🔹 Build local knowledge retrieval with Knowledge Pipeline. 🔹 Let ROCm handle local inference deployment and performance tuning on AMD Radeon™ GPUs. Register here: https://lnkd.in/gvxigysd #Dify #AMD #Hackathon #AgenticAI #LLMOps

📣 OpenAI GPT-5.6 Support is Now Live on Dify You can now build more stable and manageable AI solutions on Dify. By supporting OpenAI GPT-5.6, you can better handle multimodal inputs, precise structured outputs, and real-time streaming responses, while reducing errors during complex executions. To unlock these capabilities, the Dify OpenAI plugin has undergone a major version rewrite, transitioning its default protocol from Chat Completions to OpenAI's new Responses API. This foundational upgrade unifies messages, reasoning, tool calls, and execution results into a single, streamlined flow, ensuring your complex, production-grade Agent workflows run more stably and reliably than ever. Download the update: https://lnkd.in/gfNFpfw6