📦 sparticle62ops/pssa

PSSA: The AI That Remembers

PSSA is the small language model that actually remembers what it reads. Instead of re-reading your entire previous text every single time, it carries a fixed-size state forward and keeps a tiny memory bank to look up specific details. This means it learns faster than a standard transformer and generates text twelve times quicker on the same computer. The whole thing is written in Rust with no heavy frameworks, proving that a simpler architecture can outperform the heavy hitters. It is a research prototype, not a chatbot, but it shows that efficiency matters more than raw size.

📰 https://news.ycombinator.com/item?id=49903993

🆔 @hackernewsgithubprojects

📦 zju-omniai/spatial-interactor

Spatial-Interactor

The most surprising fact about the new spatial-interactor project is that it teaches AI to understand physical space by actually moving through it, rather than just looking at static photos. This tool solves the problem of robots or virtual assistants getting lost or confused about where objects are in relation to them. It works by feeding the computer a step-by-step video of a camera moving through real rooms and simulated houses. The system then learns to connect those visual changes to actual physical movements, like turning a corner or walking forward. This allows the model to build a mental map of the environment on its own.

🆔 @hackernewsgithubprojects

📦 latticeverse/latticeverse

LatticeVerse

Turn any property you want into a physical lattice structure. LatticeVerse is the umbrella for a whole pipeline that takes a target stiffness or strength and spits out a 3D design that actually works. It links geometry modeling, physics simulation, and AI-driven inverse design into one flow. The coolest part is the versioned lattice sample. Every design carries its math, material, and manufacturing checks. So when a solver proposes a shape, you can trace exactly how it got there. No more guessing if a lattice is just math or actually printable. It’s a full research map for building things that behave exactly as you ask.

🆔 @hackernewsgithubprojects

📦 facebookresearch/context-language-models

Context Language Models

The most surprising thing about Context Language Models is that they let an artificial intelligence manage its own memory like a human editor, actively rewriting a file to keep only what matters. This project solves the problem of information overload by giving the model unrestricted control over its own context, allowing it to condense old thoughts and prioritize new data without human intervention. It is fascinating because it turns memory from a passive log into an active tool, enabling the system to handle long, complex tasks with far less wasted computation than standard methods.

📰 https://news.ycombinator.com/item?id=49915539

🆔 @hackernewsgithubprojects

📦 sparkcpark/synthetic_hospital

Synthetic Hospital

Synthetic Hospital turns abstract medical study questions into living digital patients with full medical histories, allowing researchers to test artificial intelligence systems on real clinical reasoning. Instead of static text files, this project simulates a complete hospital system where AI agents must browse charts, retrieve evidence, and summarize patient conditions over time. It solves a critical problem for developers who need safe, reproducible environments to verify if their models truly understand complex medical scenarios without risking real patient data. The platform even acts as a reinforcement learning environment, giving models a score based on how well they diagnose and summarize.

🆔 @hackernewsgithubprojects

📦 xuanwo/jiandao

Jiandao

Jiandao is the browser extension that quietly translates your web pages without any visual clutter. Unlike heavy tools that crowd your screen with bubbles, sidebars, or floating buttons, this extension keeps things incredibly clean. You just press a simple shortcut to reveal beautifully integrated, bilingual translations sitting neatly under each paragraph, and press it again to hide them. It connects directly from your browser to your preferred language model, meaning your data never touches a middleman server. It is the ultimate lightweight reader for anyone who wants a distraction-free way to browse the multilingual web.

🆔 @hackernewsgithubprojects

📦 composio-community/open-dot

Open Dot: Personal AI Agents on Your Mac

Open Dot puts a personal AI agent that works on its own, right on your own computer. Instead of being locked into a premium subscription, this open source Mac app lets you run background tasks using your own API keys or even open models. The coolest part is that each agent gets its own private browser that stays logged in. It can watch your inbox, check your calendar, and even run code, all while you work. You can set rules so it asks before doing anything risky, like sending an email, but handles the busywork automatically.

📰 https://news.ycombinator.com/item?id=49912211

🆔 @hackernewsgithubprojects

📦 urosidoki/htn_planner

HTN Planner

Compile your game AI logic directly into native C code before the game even starts. That is exactly what the HTN Planner does. Instead of making your computer think on the fly, which is slow and unpredictable, you describe your character's goals in a simple declarative language. The tool then bakes those rules into highly efficient native code. During gameplay, the system just runs that pre-built logic instantly. It solves the problem of laggy AI by removing all the heavy parsing and decision making from the actual frame loop. You define the behavior once, and the engine handles the rest at lightning speed.

🆔 @hackernewsgithubprojects

📦 1038lab/comfyui-qwenasr

ComfyUI-QwenASR: The Cleanest Local Speech-to-Text Node

Turn messy voice recordings into clean, timestamped subtitles without touching a single server. ComfyUI-QwenASR plugs the Qwen speech model directly into your local ComfyUI setup, handling the heavy lifting of transcription so you can just connect an audio node and get readable text. The coolest part is the smart cleanup: it automatically fixes spaced-out letters, converts spoken numbers into digits, and keeps your custom industry terms intact without needing a restart. It is perfect for podcasters or developers who want precise word-level timing for long-form audio, all running privately on your machine.

🆔 @hackernewsgithubprojects

📦 bilalyazicioglu/tincan-cli

tincan-cli

You can now host secure voice and text chats directly from your terminal with absolutely no servers, no signups, and no port forwarding. A clever command line tool called tincan-cli makes this possible by turning the first person who opens a room into the coordinator, while voice traffic flies directly between participants as encrypted, peer-to-peer data streams. Written in Rust and styled beautifully with a copper-toned terminal interface, it features built-in neural noise suppression that instantly filters out background typing and fan noise. It is the ultimate lightweight, serverless alternative to Discord that runs entirely on your own terms.

🆔 @hackernewsgithubprojects

📦 pengchujin/livecanvas

One Sentence To Animated Live Photo: LiveCanvas

LiveCanvas turns a simple sentence into a polished, animated Live Photo with a custom cover. You type a single prompt into your favorite AI coding assistant, and the project handles the rest. It researches the topic, plans a storyboard, and generates a short video using Remotion. The result is a shareable, moving graphic you can use anywhere. It removes the need for heavy editing software or manual design work. You get a finished, professional-looking piece of media in minutes. If you want to share a fact or a chart, this tool makes it easy to create a visual that actually moves.

🆔 @hackernewsgithubprojects

📦 sethrobinson/patchy

Patchy: Photoshop's Open Source Cousin

An open source editor named Patchy actually opens, edits, and saves Photoshop files without a single cent of licensing fees. It reads the complex layer structures and blend modes native to Adobe, meaning you can grab a file created in the paid version and keep working on it here. This solves the lock in problem by giving you full ownership of your digital canvas. You get the familiar workflow without the subscription anxiety. It is a powerful tool for anyone who wants creative freedom without the price tag. If you are tired of paying for access to your own art, this is your alternative.

🆔 @hackernewsgithubprojects

📦 elliotttate/wind-waker-recomp

Wind Waker Recomp: Zelda on Your iPhone

Wind Waker Recomp is the static recompilation project that finally lets you play the original GameCube classic on your phone without the usual emulation lag. Instead of interpreting old console code on the fly, it translates the entire game into native machine instructions before you even start, which means it runs like a modern app rather than a simulation. You get sixty or even one hundred and twenty frames per second, widescreen support, and modern controls like mouse and keyboard support. It is not a simple port; it is a deep technical lift where every instruction is checked to work on your specific device.

🆔 @hackernewsgithubprojects

📦 twitech-lab/devchain

DevChain: Run a Team of AI Agents on Your Computer

DevChain is the local-first app that lets you run a coordinated team of A.I. coding agents directly on your own computer. Instead of juggling separate chat windows for different tools, you get a shared board and live terminals where agents plan, build, and review code side-by-side. The most compelling feature is that you can move your entire project to a virtual machine on your server. This means your agents keep working and syncing files even after you close your laptop. It’s a practical way to build software with a digital crew that never sleeps, keeping all your data private and under your control.

🆔 @hackernewsgithubprojects

📦 isledecomp/racers

LEGO Racers: The Source Code

The entire source code for the 1999 hit LEGO Racers is now available for anyone to read and edit. This project, simply called racers, is a decompilation effort that reverses the original game’s machine code back into readable programming language. The goal is to match the original instructions perfectly so the code can be studied, modified, or ported to modern systems. It is not a mod or a fan game, but a faithful recreation of the actual engineering behind the toy car racing classic. It is a rare look under the hood of a nostalgic masterpiece, proving that even old software can be fully understood and improved today.

🆔 @hackernewsgithubprojects

📦 atomdrift-project/scan

scan

This scanner catches malware that traditional tools miss, finding zero day threats in your software supply chain. It uses machine learning to inspect over a hundred file types, from compiled binaries to plain text scripts, spotting malicious code before you run it. The system updates its rules daily using reinforcement learning, meaning it adapts to new attacks as they appear in the wild. You can plug it into your build process or scan individual files with simple commands. It gives you precise control over sensitivity, so you can balance security against false alarms. Because it runs locally, your code never leaves your machine, keeping your data private.

🆔 @hackernewsgithubprojects

📦 human37/open-wispr

OpenWispr: Private Voice Typing for Mac

OpenWispr is the macOS voice tool that keeps your words on your machine. It solves the privacy gap in standard dictation by running entirely locally using Apple Silicon hardware. Instead of sending your voice to the cloud, it transcribes audio right on your Mac using Whisper. You simply hold the Globe key, speak, and release to type. No account, no subscription, and zero data leaves your computer. This project makes professional-grade voice typing private and free. If you care about data control, this is the tool to watch.

🆔 @hackernewsgithubprojects

📦 alpcanaydin/tusk

Tusk: The Database Client That Shows Its Work

Tusk is a free, native Mac app that lets you manage twenty different database types using just your keyboard, and its built-in AI assistant actually shows you the queries it writes so you can review them before running them. Most database tools are slow, clunky web apps, but Tusk is built from scratch in Rust to feel instant and local, like a pro tool that respects your privacy. It handles everything from connecting to servers to editing data, all without sending your secrets to the cloud. If you are a developer tired of juggling browser tabs for data work, this is the one tool you need to check out.

🆔 @hackernewsgithubprojects

📦 2022hpsk/speakermemr1

SpeakerMem-R1: The AI Memory That Knows Who Said What

SpeakerMem-R1 is the memory system that finally fixes how AI handles group chats. Most AI assistants forget who said what the moment three people start talking at once. This project solves that by splitting memory into two distinct tracks. One track stores the raw conversation exactly as it happened. The other organizes those details into specific profiles for each person and the group. This means the AI can finally tell the difference between what you said and what someone else said about you. It turns messy group text into clear, usable facts.

🆔 @hackernewsgithubprojects

📦 screwys/rufin

Rufin: Your Music, Everywhere

Turn every music source into one unified, native player. Rufin connects your local files, Plex, Navidrome, Jellyfin, Emby, and even WebDAV shares directly into a single, fast application built entirely in Rust. It does not rely on a web browser or a generic wrapper; instead, it creates a true desktop experience that respects the unique features of each server you use. The standout feature is its ability to merge all these different sources into one global library, meaning you can drag tracks from a remote server into a playlist with your local files seamlessly.

🆔 @hackernewsgithubprojects

📦 tugrawork-creator/saas-motion-kit

SaaS Motion Kit: 100 Ways to Show Your Software

SaaS Motion Kit is the motion design framework that refuses to let your product demo look generic. Most software promos feel identical because creators reuse the same flashy transitions over and over, but this toolkit forces you to mix things up by offering one hundred unique visual themes and a variety audit that catches repeated effects before you build. It works with an AI assistant to walk through a strict seven stage production process, ensuring every cut feels intentional and distinct. You get a complete creative playbook that turns a simple idea into a polished video without the usual creative fatigue.

🆔 @hackernewsgithubprojects