This Open-Source WAF from China Has 21.7K GitHub Stars
SafeLine is an open-source WAF from Chinese security company Chaitin with 21.7K GitHub stars. It covers CC protection, bot mitigation, dynamic obfuscation, and identity auth — in a deployment even indie developers can handle.
27B on a Phone: The Density Numbers Behind PrismML’s Bonsai Breakthrough
PrismML Bonsai 27B uses ternary quantization to compress a 27B model to 5.9GB, enabling phone deployment. 10x intelligence density versus full-precision. Benchmark gaps analyzed.
Tencent Hunyuan Hy3 Quantized Models: 295B Parameters, Single-GPU Deployment
Tencent Hunyuan Hy3 in quantized formats: 295B parameters compressed to 85.5GiB (1-bit) and 169.9GiB (4-bit). MTP speculative decoding adds 50-60% throughput. Available on HuggingFace.
Alibaba AutoNavi ABot-WorldStudio: Unified World Model Platform for Interactive Video and 3D Scene Generation
AutoNavi releases ABot-WorldStudio, a general-purpose world model workshop integrating interactive video and 3DGS scene generation with a spatiotemporal portal feature. Single RTX 5090 deployment, open-source dual models.
WorkBuddy Skills Explained: From Setup to Self-Evolution
A complete breakdown of WorkBuddy Skills, how to install, create, and leverage self-evolving mechanisms that keep your AI assistant getting better over time.
The 5 Tools That Actually Make Hermes Worth Using
5 tools that make Hermes worth using. RTK saves tokens, Hindsight adds memory, Tokscale shows spend, Ghostty upgrades terminal, Camoufox gives agents browser.
Moving a Production AI Agent from Claude Opus 4.8 to GPT-5.6: What Actually Changed
Real production data: switching from Claude Opus 4.8 to GPT-5.6 Sol cut build time in half, cost 27% less, and output fewer tokens. But it came with surprises.
The Animation Audit Skill That Changes How You Ship UI
Linear's new /improve-animations skill uses a smart split: expensive models judge, cheap models execute. Here is how the four-phase workflow actually works.
Tutti Review: The Multi-Agent Workspace That Actually Solves the Context Switching Problem
Stop Re-Explaining Your Project to Every AI Agent I ran into a familiar wall last week. I was halfway through building a project with Claude Code when the credits ran out. I switched to Codex to keep going. And then came the painful part: spending…
How to Share Skills Across Multiple AI Agents
Three practical principles for sharing and maintaining Skills across multiple AI agents, without duplicated prompts and outdated versions.
Meituan Open-Sourced a 1.6T Parameter Model. Here Is What Makes LongCat-2.0 Worth Watching.
Meituan open-sourced LongCat-2.0: a 1.6T sparse MoE model with 48B active params and million-token context. Here is what makes it worth testing.
The 18 Hottest Audio & Video Production Skills on GitHub
18 hottest GitHub projects for AI audio and video production, ranked by stars. Covers MoneyPrinterTurbo, ChatTTS, Remotion, voicebox, and more.