📦 billy-enrizky/openbrowser-ai

OpenBrowser: The AI That Codes Your Way Around the Web

OpenBrowser is the browser automation framework that lets an AI write its own Python code to do the clicking for you. Instead of guessing where a button might be, the large language model generates actual code to interact with the page in a persistent workspace. This means it can navigate complex sites, fill out forms, and scrape data with the precision of a developer, not just a script. It works directly with your existing browser, so you can watch it think, type, and extract exactly what you need.

🆔 @hackernewsgithubprojects

📦 sharziki/semdecide

SemDecide: grep for meaning in your terminal

Pipe text into SemDecide and get a typed semantic decision back, right in your terminal. It acts like grep for meaning, letting you ask if a log entry describes an account takeover or if a customer is about to churn. The tool returns a clear verdict with a probability score and a stable exit code, so your shell scripts and CI pipelines can handle AI judgments without messy parsing or flaky glue code. It keeps uncertainty visible, so you never accidentally treat a low-confidence guess as a hard fact. It’s a slick way to bring semantic intelligence into Unix workflows without writing prompts.

🆔 @hackernewsgithubprojects

📦 openrsi-foundation/openrsi-index

OpenRSI Index: Can AI Do Real Science?

Test whether AI can actually do original research, not just follow instructions. OpenRSI Index is a growing collection of hard, real-world scientific tasks designed to see if AI agents can push the frontier of knowledge instead of just repeating known results. The coolest part? It runs these tasks on massive clusters of GPUs, like hundreds of them, to handle genuinely complex problems like optimizing training for huge language models or improving video understanding. Think of it as a stress test for AI's ability to be a real researcher. It’s not about speed or simple math; it’s about sustained, high-level scientific progress.

🆔 @hackernewsgithubprojects

📦 contrastive-lm/clm

CLM: The AI That Decides at Light Speed

CLM is the fast decision engine that picks the right action in milliseconds without the usual heavy lifting. Instead of a massive model thinking through every step, it uses a clever trick to instantly match the current situation with the best possible move. This makes it perfect for real-time tasks like playing games or quickly checking if your code is correct. You can even plug it into your own projects to make your bots smarter and way cheaper to run. If you need an AI that acts on instinct rather than slow reasoning, this is the tool to watch.

🆔 @hackernewsgithubprojects

📦 mihailescu2m/llama.cpp

Run Huge AI Models On Your Mac

llama.cpp is the C engine that lets your laptop run massive AI models that are far too big to fit in its memory. While most software demands expensive cloud servers for these tools, this project streams the model's heavy parts directly from your hard drive as you need them, letting you run a two hundred billion parameter model on a standard Mac with just sixty four gigabytes of memory. By treating your disk speed as the real limit rather than your memory, it opens the door for local, private, and massive language processing on hardware you already own.

🆔 @hackernewsgithubprojects

📦 midspiral/lemmascript

Lemmascript: Proving TypeScript Code

Lemmascript lets you write ordinary TypeScript while proving your code is actually correct. You add simple comments to your files describing what should be true. The tool then turns your TypeScript into a format that a formal proof checker can analyze. It hands off your logic to serious verification engines to confirm that your functions behave exactly as intended. If the math checks out, you get a guarantee that your logic is sound. This is huge for developers who want certainty without rewriting everything in a new language. It bridges the gap between daily coding and deep verification.

📰 https://news.ycombinator.com/item?id=47849102

🆔 @hackernewsgithubprojects

📦 geastack/examples

GeaStack Examples: Run One App Everywhere

Take a single piece of code and run it on a browser, a tiny microcontroller, or a custom operating system without touching a single line. GeaStack Examples is a collection of ready-made demo apps that prove exactly how that works. The coolest part is the manifest file inside each app. It is just a simple list that tells the system where the app lives and what hardware it can run on. No more juggling separate builds or rewriting logic for different screens. One project, one set of rules, infinite targets. If you want to build software that truly doesn't care where it runs, this is your cheat sheet.

📰 https://news.ycombinator.com/item?id=49807021

🆔 @hackernewsgithubprojects

📦 samyost1/3dicon

Make Animated 3D Icons with AI

Drop in a single text prompt and get a finished, transparent WebP icon that loops perfectly inside your app. This tool is a skill for Claude Code that handles the entire workflow, starting with a generated image and using a clever trick to make the video loop seamlessly by matching the first and last frames. It also handles the tricky background removal mathematically to ensure soft edges stay clean, so you are not left with ugly halos around your icon. You stop for approval after every major step to avoid wasting money, giving you total control over the final result.

🆔 @hackernewsgithubprojects

📦 agent-substrate/env

env: The Remote Workspace That Wakes Itself

env is the environment manager that lets you run code in remote workspaces that pause and resume themselves. It solves the problem of keeping expensive compute jobs running without wasting resources. Instead of a server that stays on constantly, this system checkpoints your work and sleeps the machine. The surprising part is that it wakes up automatically. When you send a simple command to a paused workspace, the system boots it back up in the background. No manual restart buttons required. This creates a flexible setup for developers who need to run long tasks. The takeaway is simple.

🆔 @hackernewsgithubprojects

📦 nahid-sparktales/agent-dispatcher

Agent Dispatcher: Your AI's Project Brain

Stop making your AI guess which files matter. Agent Dispatcher acts as a smart router that scans your codebase to pull up only the exact lines of code and project notes your AI actually needs. It pairs every task with a specialized role, so debugging a bug gets a debugger, while writing docs gets a writer. This kills the chaos of context overload, ensuring your AI works with precision instead of confusion. If you want your AI to work smarter, not just harder, this is the missing piece.

🆔 @hackernewsgithubprojects

📦 vmvlab/us-vla

US-VLA

Scan a human liver and kidney with a robot arm, guided entirely by live video. US-VLA is a new model that teaches a robot to perform abdominal ultrasound exams by watching the screen in real time. It combines standard camera views with the actual ultrasound feed to understand exactly where the probe needs to move. This means the robot can capture the five standard medical images on its own, without needing a human doctor to manually guide the probe. By turning a complex medical task into a simple visual goal, it paves the way for automated, consistent scans.

🆔 @hackernewsgithubprojects

📦 driceroland/search

Search: The 3 MB Browser That Actually Respects You

Search is the tiny web browser that finally treats your Mac like a tool, not a trap. It skips the bloated Chrome engine entirely to ride on WebKit, which means the whole app weighs just three megabytes and opens before you can blink. The coolest part? It runs Chrome extensions right on Safari’s engine, so you get your favorite add-ons without dragging in a second, heavy browser. There is no account to sign into, no cloud sync, and no data phoning home. It is just a fast, quiet window where you can hide annoying overlays, float videos, and read articles without the noise.

📰 https://news.ycombinator.com/item?id=49828120

🆔 @hackernewsgithubprojects

📦 fank1/foot-cream

Footcream: Read books in your units

Footcream is the KOReader plugin that finally lets you read English books in the units you actually think in. It scans your book once and converts every measurement, turning phrases like six feet into point one eight meters right on the screen. The best part is that it runs entirely on your device, so you need no internet and no account. It gets smart enough to tell the difference between pounds of weight and pounds of money, ensuring it only converts what matters. This small tool removes the mental math from your reading, letting you enjoy the story without pausing to calculate distances or temperatures in your head.

📰 https://news.ycombinator.com/item?id=48760335

🆔 @hackernewsgithubprojects

📦 met4citizen/headaudio

HeadAudio: Real-Time Browser Lip-Sync

HeadAudio makes your 3D character talk in real time by simply listening to any audio stream right inside the browser. Instead of needing a text transcript or server calls, this tool analyzes the sound waves directly to figure out which mouth shapes to draw. It runs entirely on your device, so it’s fast, private, and works with any voice source you have. You just connect the audio and let the software handle the rest. It’s a clever way to bring avatars to life without the heavy lifting. If you want your characters to speak naturally, give this a try.

🆔 @hackernewsgithubprojects

📦 zeronsh/zeron

Control All Your Coding Agents From One Place

Zeron lets you run and steer multiple independent coding agents from a single native interface, even on a server without a screen. It handles complex background work on remote machines while you simply read the output or send new instructions from your laptop or phone. This design keeps your local files private by default, with optional secure syncing across your devices. You get a unified view of all your automated tasks, solving the chaos of managing separate terminal windows for every different tool. It turns fragmented agent management into a single, coherent dashboard. If you use multiple coding assistants, this is how you keep them all in check.

🆔 @hackernewsgithubprojects

📦 atomicstrata/llm-wiki-compiler

llm-wiki-compiler

Raw text files do not naturally remember where facts come from, but this compiler fixes that by turning loose documents into a permanent, interconnected map. It reads your source materials once and builds durable wiki pages that permanently link every claim back to its exact origin line. Instead of your AI tool guessing through a pile of messy files every time you ask a question, it queries this pre built structure for answers that are already verified and cited. You get a stable knowledge base that stays accurate even as new information arrives. It is not just a summary generator; it is a memory system that knows its own evidence.

🆔 @hackernewsgithubprojects

📦 op7418/humanizer-zh

How to Remove the Robot Voice from Your Chinese Text

This repository is a specialized editing guide that identifies and removes the specific verbal tics that give away text was generated by a machine. It targets the empty hype words, repetitive sentence structures, and generic praise that characterize typical machine output. By applying a detailed checklist of thirty one common patterns, it strips away the robotic polish while strictly preserving the original facts and tone. This makes it a practical tool for writers who want their articles to feel genuinely human. The result is cleaner, more natural text that reads like a person actually wrote it. You get back your voice without losing the substance of your work.

🆔 @hackernewsgithubprojects

📦 alchemy-run/alchemy

Alchemy: Cloud Infra Without the Glue

Stop writing separate configuration files for your cloud servers. Alchemy lets you define your entire infrastructure and your application logic in one single, type-safe program. This tool eliminates the messy glue code that usually connects your code to cloud services like S3 or databases. Instead of juggling multiple formats, you simply describe your setup using a specific library called Effect. This approach ensures that errors are caught early in your code rather than failing during deployment. It supports major providers like AWS and Cloudflare right now. The result is a cleaner workflow where your local development matches your live production environment perfectly.

🆔 @hackernewsgithubprojects

📦 aayushch/laya

Laya: Your AI Command Center

Laya takes the noise from your workday and turns it into clear, ready-to-approve actions. It sits on your computer, listening to messages from Slack, email, and project trackers. Using local AI, it reads every notification, figures out what matters, and even drafts replies or fixes. Instead of checking ten different apps, you get one calm dashboard with cards for the things that actually need your touch. The coolest part is that it runs entirely on your machine, so your data stays private. You just approve the actions Laya prepares, keeping your focus where it belongs. It is a quiet way to take control back from the endless ping.

🆔 @hackernewsgithubprojects

📦 ibrahimqureshae/mdflux

MDFlux: Turn Any Document Into AI-Ready Markdown

MDFlux is the local desktop app that turns messy scanned PDFs and office files into clean, AI-ready Markdown without touching the cloud. It solves the problem of feeding raw documents to language models, which usually requires expensive vision models or manual cleanup. By processing everything on your machine, it cuts token usage by up to six times while keeping your data private. The standout feature is its ability to read handwritten scans and batch-convert entire folders offline, making it a powerful tool for researchers and developers. You get polished text instantly, ready for analysis. Try it next time you need to feed a large document into an AI workflow.

🆔 @hackernewsgithubprojects