Dify Update Gives Your AI Assistants Memory and Skills
Build AI helpers that remember, learn, and ask before acting — no coding needed.
Dify is like a workshop for building AI assistants — you drag and drop pieces instead of writing code. The new version, v1.17.0, makes the assistants you build much more powerful. Most importantly, AI agents can now remember their state between runs, kind of like saving a game before quitting. An agent that prepared files or installed tools will start right where it left off next time, instead of starting from scratch.
Another standout feature is "skills." Think of skills as reusable superpowers you give your AI once, then use in many projects. There's now a proper manager for creating, publishing, and updating those skills. You also get reusable AI model settings — set up one configuration (say, for summarizing) and use it everywhere without redoing it each time.
For anyone worried about handing control to an AI, there's good news: you can now insert human review steps directly inside automated workflows. That means the AI can do repetitive work but pause and ask for your approval before sending something important, like an email or a purchase order. The new command palette (press Ctrl+K) also lets you zip around your projects quickly.
Security also got a boost with CAPTCHA protection and optional encryption via Azure Key Vault, plus better support for images in conversations. But some of these features — like running agents in cloud sandboxes or enabling full-text search — require your system administrator to configure them. The basics, however, are ready to go. If you've ever wanted an AI that remembers and checks with you, this update makes that a practical reality.
- AI agents can now remember their work between runs, so they handle longer, more complex tasks without losing place.
- New 'skills' feature lets you reuse bundled capabilities across projects, saving setup time and keeping things consistent.
- Human approval steps can be added inside automated workflows, so you stay in control before anything final goes out.
Why It Matters
AI you build can now work longer, remember context, and ask for permission — making it practical for real business jobs.