📦 fractalfir/crustc

crustc

The entirety of the Rust compiler has been translated directly into C. This mind-boggling project, called crustc, features a complete directory tree of translated dependencies, driver logic, and core compiler components compiled down to plain C files. Instead of rewriting everything by hand, this project systematically maps the massive, complex rustc architecture—including complex crates like gimli, hashbrown, and aho-corasick—into functional C source code. It is an incredibly fascinating proof of concept for anyone interested in bootstrapping, language compilation, or simply pushing the limits of code translation. If you want to see what Rust's compiler looks like when forced into C, you have to check this out.

📰 https://news.ycombinator.com/item?id=48768464

🆔 @hackernewsgithubprojects

📦 basecompute/basert

basert

basert is the local model engine that finally gives Apple Silicon chips the speed they deserve. It runs large language models entirely on your Mac, using hand-written Metal kernels to squeeze every bit of performance out of M-series processors. Instead of dealing with slow, complex setups, you get a single tool to pull models from HuggingFace, convert them into a highly optimized format, and chat with them or spin up an OpenAI-compatible server instantly. It even comes with language bindings for Python, Node, Rust, and Swift, so you can build your own local AI apps. Try basert to make your Mac a local AI powerhouse.

🆔 @hackernewsgithubprojects

📦 bytedance-seed/edgebench

Edgebench

AI models actually follow a highly predictable scaling law when learning from real-world environments over time. Edgebench is an evaluation benchmark that moves away from simple one-shot tests, placing autonomous agents inside real, executable task environments for twelve hours or more. By analyzing thousands of hours of agent interactions across over a hundred complex tasks like software engineering and scientific computing, it tracks the actual journey of how agents learn and improve. It turns out that AI progress over time fits a beautiful, mathematically predictable curve. It is a brilliant way to see which models actually learn to solve hard problems when given the time to think.

🆔 @hackernewsgithubprojects

📦 snumprlab/dart

Domain Arithmetic

Dart is the artificial intelligence tool that helps robotic models quickly adapt to new environments using just a single image. Usually, when a robot trained in a bright lab is moved to a dark room or a new table, its brain gets confused and it fails at simple tasks. This project solves that by using a clever mathematical shortcut to adjust the robot's vision model on the fly. By comparing just one picture of the new setup to its original training, it instantly corrects the robot's understanding. It makes physical robots much more flexible without needing hours of retraining.

🆔 @hackernewsgithubprojects

📦 btseytlin/hr-breaker

HR-Breaker

Stop sending the exact same resume to every job posting. Give yourself a massive edge by transforming your existing resume into a perfectly tailored, job-specific PDF that actually gets past automated filters. Just upload your experience in any format and paste the target job description. The tool uses local intelligence to restyle and refine your bullet points to match the role perfectly, and it runs built-in checks to ensure zero hallucination, proper formatting, and optimal keyword density. It even simulates ATS filters in real time to make sure your application stands out. Run it from your terminal or a clean local web browser interface, download your clean single-page PDF, and...

🆔 @hackernewsgithubprojects

📦 shannhk/hermes-agent-control-room

Hermes Agent Control Room

Hermes-agent-control-room is the side control plane that finally lets you manage all your autonomous helpers from one organized dashboard instead of wrestling with a chaotic pile of disconnected bots. Think of it as the ultimate operating manual and system map for your virtual team. It sets up an elegant folder structure on your server to document runbooks, track agent roles, and map out secure connections without ever exposing raw API keys. You can easily scale from a single assistant to a fully coordinated team routing tasks through a shared file system. It is the clever way to keep your agents secure, organized, and running smoothly.

🆔 @hackernewsgithubprojects

📦 pasteurlabs/mosaic

Mosaic

Mosaic is the open-source library that finally brings standardized benchmarking to differentiable physics solvers. If you have ever tried to train a machine learning model inside a physical simulation, you know the absolute nightmare of getting accurate gradients from different engines. Mosaic solves this by wrapping diverse solvers, from fluid dynamics to structural mechanics, inside standard containers with a unified interface. Instead of just checking if the physics look right, it rigorously evaluates both gradient accuracy and computational speed across frameworks like JAX, PyTorch, and Julia. It is basically the OpenAI Gym for the physics world, making it incredibly easy to see which solver actually holds up under real optimization...

🆔 @hackernewsgithubprojects

📦 runfusion/fusion

Fusion

You can now coordinate an entire fleet of specialized AI software developers across multiple machines to build software simultaneously. Fusion is a multi-node agent orchestrator that lets you spin up a distributed workspace where agents—from backend engineers and product managers to adversarial reviewers—collaborate on complex tasks. It takes care of the hard stuff, handling task lineage, managing multi-project workflows, and keeping agent nodes in sync even during long-running operations. Instead of dealing with single-file code assistants, you get a visual workspace with desktop and mobile interfaces to map out, run, and watch your coordinated AI team build whole systems together.

🆔 @hackernewsgithubprojects

📦 filliptm/comfyui-fl-qwen3tts

FL Qwen3 TTS

You can now clone any voice, design custom speakers from plain text descriptions, and fine-tune your own speech models directly inside ComfyUI. The comfyui-fl-qwen3tts project integrates Alibaba's powerful Qwen3 text-to-speech architecture into your visual workflows. It completely removes the need for complex command-line scripts by giving you a beautiful, real-time training dashboard with live loss charts and instant validation audio playback. Whether you want to generate natural speech across ten different languages or train a specialized voice model from a small folder of audio clips, this node pack turns high-end voice synthesis into a simple drag-and-drop experience.

🆔 @hackernewsgithubprojects

📦 amd-agi/micro-world

micro-world

Control the camera movement of AI-generated videos using actual keyboard and mouse commands. This is made possible by micro-world, an open-source project from AMD that turns video generators into interactive, playable environments. By training on a Minecraft dataset, it learns how standard gaming actions like pressing keys or moving a mouse should realistically change the camera perspective. You can describe a scene with text, or upload an image, and then feed in a sequence of movement keys to generate a video that moves exactly how you want. It makes virtual worlds feel responsive and immersive. Check it out to build your own playable AI worlds.

🆔 @hackernewsgithubprojects

📦 rafal-qa/slopo

Slopo

Slopo is the command-line tool that catches the hidden, non-exact code duplicates other tools completely miss. Standard tools are great at finding identical copy-pastes, but they fail when developer-written snippets are slightly modified, renamed, or scattered across different directories. This tool solves that by generating vector embeddings for your functions and analyzing their mathematical similarity. It groups closely related blocks of code into clusters, ranks them by similarity and physical distance, and outputs a neat report. This makes it incredibly easy to hand off the grouped results to an AI assistant for automatic cleanup and refactoring. Run it on your codebase today to clean up your architecture.

📰 https://news.ycombinator.com/item?id=48762038

🆔 @hackernewsgithubprojects

📦 google-antigravity/antigravity-cli

Antigravity CLI

Antigravity-cli is the terminal interface that finally puts a smart coding agent directly into your command line workflow. It runs on the same engine as the main desktop application, meaning it can understand your codebase, suggest multi-file edits, and run commands right where you already work. What makes it incredibly cool is how it handles remote work; it automatically detects SSH sessions and prints a local login link, keeping you authenticated without breaking your flow. It is built for speed and keyboard efficiency, letting you manage complex programming tasks with zero window switching. You get a lightweight, keyboard-first assistant that keeps you fast, focused, and in control.

🆔 @hackernewsgithubprojects

📦 kevin-thu/storymem

StoryMem

Generate cohesive, multi-shot narrative videos from a simple text script. While standard video generators often lose track of what characters look like from one clip to the next, storymem solves this by using a clever memory-conditioned system. It generates videos shot by shot, automatically saving visual details like character appearance and setting as memory, and then uses that memory to guide the next clip. This keeps the entire sequence looking consistent and cinematic over a full minute of footage. It is an exciting step forward for creators looking to turn written scripts into visually connected, multi-shot stories.

🆔 @hackernewsgithubprojects

📦 nousresearch/autonovel

autonovel

The autonomous AI agent pipeline that actually writes, edits, and produces a complete, print-ready book. Instead of just spitting out a mountain of text, autonovel uses co-evolving layers to coordinate world-building, characters, and outlines, tracking the propagation of changes like a real author would. It runs a mechanical scan to catch cliché AI writing patterns and runs an adversarial editing loop where separate model personas act as tough literary critics. The system even generates cover art, formats the manuscript into LaTeX, and builds a multi-voice audiobook. Its first fully generated novel reached over seventy-nine thousand words, proving that AI can handle long-form narrative structure from start to finish.

🆔 @hackernewsgithubprojects

📦 skillmatic-ai/awesome-agent-skills

Awesome Agent Skills

Awesome-agent-skills is the curated directory that finally organizes the entire universe of modular AI capabilities in one place. Instead of stuffing massive instructions into an AI prompt and wasting precious memory, developers are using standardized markdown packages to let agents load specific capabilities only when needed. This repository gathers every major tutorial, official skill library from major tech players, and academic paper detailing this clever approach to context management. It is the perfect starting point to understand how modern AI assistants learn new tricks on the fly without slowing down, giving you a clear roadmap to build your own modular AI agent.

🆔 @hackernewsgithubprojects

📦 maxritter/pilot-shell

Pilot Shell

Pilot Shell is the command-line companion that finally makes AI coding agents production-ready. While tools like Claude Code write code incredibly fast, they often skip tests, lose critical context, and ignore your team's unique guidelines. Pilot Shell steps in to enforce strict test-driven development, run automated quality checks, and compress your context to slash token costs by up to ninety percent. It even gives you a local web dashboard to manage your sessions and track automated background jobs. It is the ultimate way to turn chaotic AI generation into structured, reliable software engineering. Check it out to build smarter.

📰 https://news.ycombinator.com/item?id=48775746

🆔 @hackernewsgithubprojects

📦 shobhit99/claude-watch

claude-watch

Control your Claude Code terminal sessions right from your Apple Watch while walking away from your desk. This clever project sets up a local bridge server on your Mac that routes events from your active AI coding terminal directly to your wrist. Instead of staying glued to your monitor, you can watch the terminal run commands, edit files, and use tools in real-time. Most impressively, when the AI needs your permission to execute a script or write a file, you get an interactive prompt on your watch to approve or deny the action. Grab the code, hook up your watch, and keep your build moving.

🆔 @hackernewsgithubprojects

📦 nearcyan/vibecraft

Vibecraft

Vibecraft is the 3D visualization app that turns your local Claude Code terminal sessions into a lively, interactive workshop. Instead of staring at dry terminal lines, you watch an animated robot avatar run around a virtual hex grid in real time. When Claude reads a file, it walks over to a physical bookshelf; when it runs a search, it heads to a satellite antenna; and when it edits code, it tinkers at a workbench. With built-in spatial audio and the ability to spin up multiple parallel agent zones, this project makes watching your AI work feel like playing a classic real-time strategy game.

🆔 @hackernewsgithubprojects

📦 devops-interviews/devops-interview-questions

devops-interview-questions

Ace your next platform engineering loop by practicing with a curated collection of one hundred and fifteen real, hands-on challenges. This repository gathers actual technical screening tasks from heavy hitters like Google, Amazon, and Meta, covering everything from debugging broken Kubernetes pods to managing complex Git histories and AWS cloud architectures. Instead of dry trivia, you get practical scenarios complete with step-by-step video walkthroughs showing you exactly how to fix the systems. It is the ultimate hands-on cheat sheet for mastering real-world infrastructure problems. Grab this guide, practice the actual engineering scenarios, and walk into your next interview with complete confidence.

🆔 @hackernewsgithubprojects

📦 zerolu/awesome-nanobanana-pro

Awesome Nanobanana Pro

Generate hyper-realistic AI images using specialized prompt templates from a curated community collection. awesome-nanobanana-pro gathers the most effective text instructions and structured JSON formats specifically designed for the Nano Banana AI model. Instead of trial and error, you get precise formulas for complex challenges like rendering realistic skin textures under sunset lighting or recreating nostalgic retro camera styles. This library solves the problem of unpredictable AI art by providing proven blueprints gathered from top creators across social media. It is a fantastic guide for anyone wanting to master prompt engineering and see what this specific visual engine can truly do.

🆔 @hackernewsgithubprojects