Dify

Open-source LLM app platform

agentslow codeFree tier

Overview

Dify is an open-source platform for building LLM applications. Features visual workflow builder, RAG pipeline, agent tools, and multi-model support. Self-host for free or use cloud. 70K+ GitHub stars. One of the fastest-growing open-source AI projects.

Best For

+LLM app development
+RAG systems
+Custom chatbots
+Enterprise AI

News & Updates

Open-Weight AI vs. Open-Source AI: The Licensing Distinction That Matters

A new guide clarifies the difference between open-weight AI, which releases trained model parameters under a license, and open-source AI, which implies broader freedoms across software, data, documentation, and the development process. The distinction matters because most models labeled 'open' are only open-weight, not fully open-source. This affects how developers can legally use, modify, and redistribute models like Llama or Mistral. The guide explains the mechanism, trade-offs, evaluation, and controls that matter in practice.

breakthroughs10/15/2026

Robots Achieve Physical Recursive Self-Improvement in New Study

Researchers Hai Zhou, Fan Yang, Zeyi Chen and colleagues published a study on Physical Recursive Self-Improvement (Physical RSI), where robots use physical experience to autonomously update both their capabilities and their improvement procedures. Unlike traditional methods that rely on human engineers to tweak algorithms or hardware, this approach allows robots to learn from their own physical interactions and modify how they learn. The work demonstrates that robots can achieve lasting self-improvement without human intervention, marking a step toward fully autonomous robotic systems. This could lead to robots that adapt to new tasks and environments on their own, reducing the need for costly reprogramming.

robots10/14/2026

New Guide Breaks Down L1, L2, Dropout, and Early Stopping to Curb Overfitting

A new technical guide explains regularization in machine learning, covering L1, L2, dropout, and early stopping. It details how these methods modify the training objective, architecture, or stopping process to discourage brittle complexity and improve generalization. The guide also discusses trade-offs, evaluation, and practical controls for implementing regularization effectively. This resource helps practitioners choose the right technique to prevent overfitting and build more robust models.

breakthroughs10/13/2026

High Schoolers Launch Powermove, AI Video Editor That Rewrites Its Own Features

Two high school students launched Powermove, a video-editing tool that connects to an AI agent so users can modify the app's featureset and generate custom effects and extensions. The founders argue this 'malleable software' approach is the next step for applications, pointing to Anthropic's recent addition of similar self-modifying capabilities in Claude Code as validation. Powermove is currently available via a Show HN post, targeting creators who want to extend their editing tools without waiting for official plugin releases.

new_tools10/13/2026