Latest Picks

Handpicked repositories selected for utility and relevance.

spec-kit interface preview

GitHub Spec Kit: Fixing AI Code That Looks Right But Breaks

Vibe-coding with AI agents hits a wall when prompts produce code that compiles sometimes, misses intent often, and breaks in production. GitHub's Spec Kit tackles this by turning specifications into executable artifacts that structure how AI agents build software—trading the chaos of unstructured prompts for predictable outcomes, with honest trade-offs around context overhead.

githubGI

spec-kit

github

puter interface preview

Puter: The Open-Source OS Built to Challenge Cloud Giants

Puter is a full browser-based operating system—not just file storage, but a complete desktop environment with apps. Built with vanilla JavaScript by a small team, it offers a privacy-first, self-hostable alternative to Dropbox and Google Drive for developers who value open-source infrastructure.

HeyPuterHE

puter

HeyPuter

NeoPass interface preview

NeoPass: When Exam Software Locks Down Too Hard

Exam proctoring software like NeoExamShield locks down browsers so tightly that students can't access basic functionality. NeoPass, a Chrome extension with 25,000+ GitHub stars, bypasses these restrictions by masquerading as the proctoring software itself. The tool's popularity reveals a tension: when academic integrity tools restrict too much, students engineer workarounds.

Max-EeeMA

NeoPass

Max-Eee

oh-my-opencode interface preview

Solo Dev Builds AI Tool With IDE Powers Claude Code Lacks

A solo developer identified what billion-dollar AI coding tools were missing—proper IDE capabilities like LSP and AstGrep—and built an open-source alternative that quickly gained 26,500 GitHub stars. The result sparked both adoption and controversy around API terms of service and resource consumption.

code-yeongyuCO

oh-my-opencode

code-yeongyu

DeepSeek-V3 interface preview

DeepSeek-V3: Open-Source Model Matches GPT-4 at 1/53rd Cost

A Chinese AI lab released an open-source model that matches frontier AI performance at a fraction of the cost. DeepSeek-V3 uses Multi-head Latent Attention and auxiliary-loss-free load balancing to achieve benchmark parity with GPT-4 and Claude, while the entire training run cost under $6 million. The model has real limitations—64k context windows, chatbot censorship—but 101k GitHub stars in days signals what engineers want: powerful, accessible AI they can run themselves.

deepseek-aiDE

DeepSeek-V3

deepseek-ai

ladybird interface preview

Ladybird Browser: Building a New Engine from Scratch

Ladybird is attempting one of computing's hardest challenges: building a modern browser engine from zero. The project has backing from Cloudflare, Shopify, and GitHub's co-founder—serious technologists who believe browser monoculture threatens the open web. It's pre-alpha, buggy, and slow, but that's expected when you're competing with decades-old codebases from Google, Apple, and Mozilla.

LadybirdBrowserLA

ladybird

LadybirdBrowser