📦 inclusionai/ming-3.0-tts

Ming 3.0 TTS: The Open Source Voice Clone That Actually Works

Ming 3.0 TTS lets you describe a voice in plain English, like a grumpy old man or a bright young woman, and it generates that specific sound without needing a single audio sample. This open source project solves the problem of robotic, one size fits all text to speech by letting you control speed, pitch, and even laughter tags directly in your text. It’s a family of models, with a tiny version that runs on your local machine, so you can build apps that sound genuinely human.

🆔 @hackernewsgithubprojects

📦 appariciojunior/video-editor-skills

Video Editor Skills: Your AI Co-Editor

Video Editor Skills is the AI co-editor that finally handles the boring parts of your video cut. It teaches the AI to open your footage in DaVinci Resolve, find the dead air, and slice it out while keeping your natural rhythm. The coolest part is that it actually watches your silent clips to find real story transitions instead of just guessing. You make the creative choices, but this tool handles the tedious cleanup and audio fixes so you can just relax. It is free to use as long as you do not sell it, making it a handy partner for your next vlog.

🆔 @hackernewsgithubprojects

📦 gilvt-wtj/gilvt

Gilvt: The Terminal That Knows Your AI Is Waiting

Gilvt is the native macOS terminal that finally tells you which coding agent needs your attention. If you run multiple AI sessions in parallel, the hard part is never writing code, but knowing which window is stuck waiting for approval or just finished its task. Gilvt solves this by adding a smart sidebar that lists every session and automatically bubbles up the ones that need you to the top. It also gives you a clear view of exactly what each agent changed, so you can review the work right where you are. It feels less like checking tabs and more like having a helpful assistant watching your back.

🆔 @hackernewsgithubprojects

📦 mugar123/papertok

PaperTok: Your Personal Science Feed

PaperTok is the scientific feed that finally stops you from searching for things you don't even know exist. It takes papers from sources like arXiv and PubMed, then builds a scrollable stream based on what you actually read and like. The coolest part is how it adapts in real time. Skip a boring paper and the algorithm quietly learns your taste. Save a fascinating one and similar research pops up. It turns the chaotic world of academic publishing into a simple, personalized scroll that gets smarter every time you look at it. No more keyword guessing.

🆔 @hackernewsgithubprojects

📦 dracu-lah/tmplayer

TMPlayer: Watch Telegram Videos Before They Finish Downloading

TMPlayer lets you play videos from your own Telegram chats on Android, Windows, macOS, and Linux even while they are still downloading. Most players force you to wait for the file to fully save before you can hit play, but this app streams the data as it arrives so you can start watching immediately. It works across phones, desktops, and TV devices like Fire TV, making it a handy way to view media without filling up your storage or wasting time. If you send large video files in chats, this tool removes the annoying wait time by treating the download process as a live stream rather than a finished file.

📰 https://news.ycombinator.com/item?id=50161458

🆔 @hackernewsgithubprojects

📦 tencent-hunyuan/hy-nomas

Hy-Nomas: Create Interactive Worlds

Paint a world that actually moves. Hy-Nomas lets you type a sentence and watch a coherent video scene unfold, reacting to your words in real time. It’s not just a static image generator; it’s a dynamic environment where you can steer the camera and change the story on the fly. The coolest part? It stays consistent. You can add a robot arm or a cinematic style, and the world holds together across multiple turns. It solves the headache of disjointed video clips by generating a continuous, interactive experience. If you want to build worlds instead of just watching them, this is the tool to see.

🆔 @hackernewsgithubprojects

📦 nyankomintsu/world-execute-me-lyric-mv

World Execute Me Lyric MV: How It Works

Render every frame of a fan-made music video for Mili's World Execute Me directly in your browser using only code. This project isn't a video file; it is a live JavaScript engine that draws the entire animation on a canvas, turning a chat with an A.I. character into a visual story. Every pixel is calculated in real time, meaning no footage was ever filmed or edited. The coolest part is that every single line of this complex animation code was actually written by an A.I. assistant based on detailed human prompts.

🆔 @hackernewsgithubprojects

📦 salesforceairesearch/beagle

Beagle: The Evolution Engine

Beagle is the training framework that treats your AI agents like biological species. Instead of manually tweaking code, it runs a natural selection process where an evolver agent tests different variations of your system and automatically keeps the versions that perform best. Think of it as a fitness tracker for your software, ensuring every change actually helps without breaking existing features. It recently helped push agents to the top of major leaderboards by finding hidden improvements we humans might miss. If you are building autonomous tools, this is how you make them genuinely better over time.

📰 https://news.ycombinator.com/item?id=49650988

🆔 @hackernewsgithubprojects

📦 terse-ai/terseai

Turn Your Desktop Into a 3D Particle World

Turn your desktop into a living 3D particle field that displays your music, notifications, and AI agent activity in real time. Terse takes your wallpaper picture and rebuilds it as a floating cloud of particles, then writes text directly into the field so it appears to float in space. The most surprising feature is that it syncs song lyrics to the beat, drawing each line out of particles as the music plays. For developers, it also gives your AI coding agents a shared space to communicate and saves money by filtering unnecessary output.

🆔 @hackernewsgithubprojects

📦 zbndev/tidemark

Tidemark: Know When Your AI Quota Resets

Tidemark keeps watching your AI service limits in the background, even when the app window is closed, so your desktop tray always shows exactly how much credit you have left. It solves the annoying mystery of hitting a hard stop in the middle of a long task by clearly showing when each rate limit window resets. The tool draws a small marker on every progress bar that tells you instantly if your current usage pace will finish before that reset time or if you are about to run out. It is built in native Rust, so it runs quietly and uses very little memory compared to heavy web-based apps.

🆔 @hackernewsgithubprojects

📦 ahren09/scienceutopia

Science Utopia: AI Simulates Science

Science Utopia uses large language models to simulate entire scientific communities, where AI agents act as researchers choosing projects, writing papers, and peer-reviewing each other's work. The project solves the problem of how to study the complex dynamics of research without waiting decades for real-world results. It is useful for anyone curious about how funding incentives and review policies shape scientific progress, allowing users to run configurable experiments on how these rules affect outcomes over time.

🆔 @hackernewsgithubprojects

📦 google-ai-edge/ml-drift

Run Big AIs On Your Phone

Run massive language models directly on your phone or browser without sending data to a server. This is the big idea behind ML Drift, a new engine from Google that makes heavy AI work feel instant on your own hardware. The coolest part is how it builds custom math instructions right on the fly for your specific chip, squeezing huge models into tiny devices. It is basically the next step for private, offline intelligence. You get real-time answers while keeping your data safe and your battery alive. No cloud required. Just you and your device thinking fast.

🆔 @hackernewsgithubprojects

📦 aivrar/vllm-windows-build

Run vLLM on Windows without Linux

Run high-performance language model servers directly on Windows without ever touching a virtual machine or container. This project builds the popular VLLM inference engine natively for Windows, so you can skip the usual Linux workarounds entirely. It takes the standard software and fixes it to run smoothly on your native system, giving you fast, local AI serving with full graphics card acceleration. You get a simple installer and a pre-built package that just works, making it easy to test and deploy models on your desktop or laptop. No more fighting with compatibility layers or complex setups. Just download, install, and start generating text.

🆔 @hackernewsgithubprojects

📦 caldalis/minicpm-math-harness

Run a Math Proof Swarm on Your Laptop

MiniCPM Math Harness runs a swarm of artificial agents on your laptop to verify mathematical proofs without massive hardware. It uses a compact model to spawn parallel provers that attack problems with different strategies, ensuring they don't make the same mistake twice. The system then uses adversarial verifiers and executes brute-force Python scripts to check the math, settling disagreements with hard evidence rather than guesswork. This turns a single small model into a rigorous proof checker you can run locally. It is a fascinating look at how test-time compute can boost accuracy on everyday hardware, proving that you don't need a data center to do serious math reasoning.

🆔 @hackernewsgithubprojects

📦 epistates/mcpsafari

MCP Safari: Your AI Agent Finally Has Eyes in Safari

Let your AI agent drive your actual, logged-in Safari on Mac. MCP Safari connects your favorite AI assistants to the real browser session you already use. No headless Chrome, no separate login windows. It operates your live tabs, preserving your cookies and session history, so your agent can test forms, click buttons, and read pages exactly as your users see them. Built with Swift and a native Safari extension, it bridges the gap between modern AI tools and the reliable, familiar interface of Safari. It’s not a new browser; it’s a direct line into the one you already trust.

🆔 @hackernewsgithubprojects

📦 michaelshimeles/rec-rs

Rec-Rs: The Lightweight Rust Screen Recorder

A screen recorder built in Rust can stream and record in forty percent less space than the industry standard while using barely any power. Rec-Rs captures your Mac's screen, microphone, and webcam directly from the hardware, skipping the heavy software layers that slow down traditional tools. It sends that raw feed straight to your graphics card for encoding, which means your computer stays cool and your battery lasts longer. The result is a tiny, fast app that handles live streams and high-quality recordings without bogging down your system. If you want a recording tool that respects your machine's speed and energy, this is the one to try.

🆔 @hackernewsgithubprojects

📦 togethercomputer/hybridagent

HybridAgent: Thinking You Can See

Watch a language model think in plain text you can actually read. HybridAgent, created by researchers at Stanford and Together AI, lets a large language model switch between fast, private inner thoughts and normal written answers. The most surprising part is that you can look inside its head. Instead of a black box, the system renders its internal reasoning as readable words, showing exactly how it uses tools like search or a calculator to solve a problem. You get to see the step-by-step logic before it gives the final answer. This makes artificial intelligence far easier to trust and understand.

🆔 @hackernewsgithubprojects

📦 oil-oil/oil-find

Find Any File on Your Mac in Milliseconds

Stop waiting for slow search results on your Mac. Oil Find gives you instant file search by keeping a fast, in-memory map of every name on your disk. Instead of hunting through folders, you press a shortcut and results appear as you type. It uses clever coding to scan millions of files in milliseconds, even catching live changes in real time. This tool turns a chaotic computer into a neat, organized library, letting you find that one file you forgot exists in a blink. It is free, open source, and works entirely on your machine without sending data anywhere.

🆔 @hackernewsgithubprojects

📦 earendil-works/pi-durable-cloudflare

Pi Durable Cloudflare

Pi Durable Cloudflare is the durable web app that lets you fork conversations like branches of a story. It runs on Cloudflare so you can open the same chat in multiple browsers, see who is typing, and jump back to an earlier message to start a completely different path without losing the original. If your internet drops or you close your laptop, your place is saved exactly where you left it. This is a working example of how to build resilient, multiplayer AI experiences that survive restarts. Try it out to see how state persistence changes the way we chat with machines.

🆔 @hackernewsgithubprojects

📦 oaslananka/kicad-mcp-pro

KiCad MCP Pro

Artificial intelligence can now inspect and modify real printed circuit board layouts right inside open source engineering software. KiCad MCP Pro connects intelligent digital assistants directly to electronic design workflows, letting them run error checks, analyze parts lists, and generate manufacturing files. This bridges the gap between conversational models and physical hardware creation, keeping humans in complete control of every final edit before production. It turns tedious design verification into a fast, cooperative conversation that helps bring hardware ideas to life safely.

🆔 @hackernewsgithubprojects