📦 opensquilla/opensquilla

opensquilla

Large language model agents consume massive amounts of data, running up massive API bills just to keep track of their own conversation history. This project solves that problem by acting as a smart context manager that packs maximum reasoning capability into a strict token budget. By constantly optimizing what the model remembers, compacting tool outputs, and pruning unnecessary clutter from the chat history, it gives you a highly capable assistant without the skyrocketing costs. It even features a built-in terminal interface and a web dashboard to track your token savings in real time. Grab the code and build smarter, leaner AI assistants today.

🆔 @hackernewsgithubprojects

📦 zjunlp/lightmem-ego

LightMem-Ego

Stream first-person camera and audio directly from your smart glasses or web browser to build a continuous, searchable digital brain. LightMem-Ego processes your daily real-world experiences in real time, breaking everything you see and hear into current scene understanding, short-term events, and long-term habits. It solves the massive headache of trying to recall fleeting details by turning raw video and audio feeds into a structured, queryable knowledge base. It is like giving yourself perfect recall for names, conversations, and lost items. You can easily deploy this setup to chat with your own history and never forget a detail again.

🆔 @hackernewsgithubprojects

📦 frostlinx/searcheyes

SearchEyes

Build intelligent agents that can browse, search, and reason across both images and text by simulating an entire search world. While typical training pipelines struggle because they treat training data, search environments, and rewards as separate pieces, this project unifies them using a single knowledge graph. It connects data synthesis with a self-contained search engine and introduces a smart reward system to optimize how agents search. By keeping structural metadata intact, it helps smaller models match the search capabilities of models more than three times their size. It is the perfect setup for building next-generation visual search assistants.

🆔 @hackernewsgithubprojects

📦 filippawlicki/nanovsr

NanoVSR

NanoVSR is the video super-resolution model that actually runs in real time on small, low-power edge devices. Up-scaling blurry footage normally requires massive, power-hungry graphics cards, but this project solves that by ditching heavy transformers and complex optical flow algorithms. Instead, it uses a clever mathematical trick to collapse multiple network branches into a single, highly efficient stream during playback. This means you can get crisp, high-definition video at nearly thirty frames per second on a small, twenty-five watt hardware board. It is a brilliant, lightweight way to bring high-quality video enhancement to smart cameras and portable devices without draining their batteries.

🆔 @hackernewsgithubprojects

📦 vibedrift/vibedrift

VibeDrift

Keep your AI coding agent from making a mess of your codebase as it grows. AI assistants write logical code, but they have no memory of the custom design patterns you already settled on. One file gets a clean repository pattern while the next gets raw SQL, slowly turning your project into a chaotic mix of conflicting styles. This tool learns the dominant coding patterns your project already uses by taking a majority vote of your files, then automatically flags any new code that deviates. It stops style drift in its tracks right on your machine.

🆔 @hackernewsgithubprojects

📦 harmont-dev/hyper

Hyper Distributed MicroVMs

Spin up secure, isolated virtual machines in seconds without the bloat of traditional virtualization. Built on the Erlang virtual machine, hyper is a distributed orchestrator that manages Firecracker microVMs across multiple host nodes. Instead of relying on slow, full disk snapshots, it implements copy-on-write disk layering to let you fork running virtual machines instantly with zero performance penalty. You can load standard container images and run them inside isolated sandboxes while scale scheduling optimizes where your workloads run based on shared resources. It is the perfect tool for building secure, multi-tenant cloud platforms with raw bare-metal performance.

📰 https://news.ycombinator.com/item?id=48849161

🆔 @hackernewsgithubprojects

📦 ahekot/vnccs_easy-install

vnccs_easy-install

Build your own fully customized visual novel character sprites without touching a single line of messy configuration code. The vnccs_easy-install repository packages up a complete character production pipeline inside a portable Windows installer. It bundles ComfyUI with special tools to let you design outfits, swap poses in a 3D studio, generate natural emotion sheets, and even clone existing art. Instead of fighting python dependencies or downloading models manually, you just run one file to get a ready-made suite. Grab the package, run the setup script, and start building your own reusable character library today.

🆔 @hackernewsgithubprojects

📦 js-lee-ai/dart

DART

You can now dynamically adjust how much time an AI spend thinking before it answers, and you do not need to retrain a single model to do it. A clever project called dart lets you route questions between fast, cheap models and deep, reasoning models on the fly by checking if they agree on an initial draft. If the quick model nails it, you save money and time; if it struggles, the heavy-duty thinker automatically kicks in. It is basically a smart thermostat for your AI budget, keeping things fast without losing accuracy. It is a brilliant way to stop wasting brainpower on easy questions.

🆔 @hackernewsgithubprojects

📦 js-lee-ai/skin-deep

skin-deep

Test how easily an AI assistant will bypass its own safety rules before anyone even tries to attack it. While a model might block harmful requests during standard checks, its safety alignment is often surprisingly fragile and can break down completely after just a tiny bit of normal fine-tuning. This project, skin-deep, solves that problem by analyzing the internal mathematical activations of a language model. It calculates a single score showing if the model's safety is truly deep or just a temporary mask. Use this diagnostic tool to find and patch weak spots in your model before deploying it.

🆔 @hackernewsgithubprojects

📦 mjunaidca/polymarket-skills

Polymarket Skills

Polymarket-skills is the open-source trading toolkit that lets artificial intelligence agents safely navigate prediction markets using a strict paper-trading-first approach. Built as modular skills for AI assistants, it connects to live market data to scan for pricing inefficiencies, analyze momentum, and simulate trades without risking actual capital. The standout feature is its built-in safety constitution, which forces the system to run realistic paper trading sessions to prove a consistent mathematical edge before ever unlocking live execution. It translates complex blockchain market structures into clear, manageable steps so any curious developer can experiment with systematic trading safely.

🆔 @hackernewsgithubprojects

📦 harimoradiya/androidtvremote

Android TV Remote

androidtvremote is the Android application that rescues you when your physical TV remote inevitably slips between the couch cushions. It instantly transforms your smartphone into a fully functional remote control over your local Wi-Fi network. What makes this project stand out is how it effortlessly handles the heavy lifting of pairing protocols, secure handshake encryption, and network discovery right from your phone. You get smooth navigation, volume control, and even a synced keyboard that lets you type search queries directly on your phone instead of agonizingly clicking one letter at a time on the screen. It is completely free, privacy-focused, and keeps your living room smart.

🆔 @hackernewsgithubprojects

📦 reliefeai/browser-relay

browser-relay

Your AI agents can now work directly inside your everyday Chrome browser tabs, fully authenticated with your active logins, cookies, and extensions. Instead of spinning up a fresh, empty browser window that has none of your saved sessions, browser-relay connects your personal browser to local or remote AI agents over a simple command line, HTTP API, or Model Context Protocol. By using a lightweight Chrome extension and local server, agents can interact with your open tabs, click elements, capture network requests, and read page layouts safely. It is a fantastic local-first tool for developers who want their AI assistants to easily interact with real-world SaaS dashboards without annoying login hurdles.

🆔 @hackernewsgithubprojects

📦 lucidrains/x-jepa

x-jepa

Build and play with Yann LeCun's vision of autonomous machine intelligence using x-jepa. Instead of just predicting the next pixel or word, this project lets you experiment with Joint-Embedding Predictive Architectures, or JEPAs, which try to model the world in a much smarter, abstract way. It implements code inspired by cutting-edge papers on world models and self-supervised learning, bypassing the usual messy hacks. It is perfect if you want to move beyond basic generative AI and actually understand how machines might learn to plan and reason. Grab the library and start building your own world model today.

🆔 @hackernewsgithubprojects

📦 mikepapadim/london-property-hunt-public

London Property Hunt Public

london-property-hunt-public is the automated search workflow that completely takes the headache out of finding a flat in the UK capital. Instead of wasting hours refreshing tabs, this project uses artificial intelligence to scan four major rental platforms twice a day, instantly filters out duplicates, and logs everything in a neat spreadsheet. Its most impressive trick is that it automatically drafts personalized outreach emails for the highest-priority properties and drops them straight into your inbox as ready-to-send messages. It means you can apply to prime listings seconds after they go live without lifting a finger, giving you a massive head start in a competitive market.

🆔 @hackernewsgithubprojects

📦 henryndubuaku/maths-cs-ai-compendium

Maths CS AI Compendium

Master artificial intelligence from the ground up by building a rock-solid foundation in mathematics and computing. The maths-cs-ai-compendium is an open, unconventional textbook designed to ditch dense notation in favor of pure, intuitive understanding. Instead of skipping steps, it connects the dots across twenty chapters, covering everything from basic vectors and GPU programming to advanced multimodal learning. It even includes a custom server to turn the entire guide into a local knowledge base for your AI coding assistants. Grab a coffee, dive in, and finally understand how the math behind AI actually works.

🆔 @hackernewsgithubprojects

📦 kangarooking/cangjie-skill

Cangjie Skill

You can now transform passive book knowledge into active, executable AI tools. The open-source project called cangjie-skill uses a clever seven-stage pipeline to distill books, podcasts, and long videos into structured, testable skill packs designed specifically for AI agents to execute. Instead of generating flat, easily forgotten text summaries, it extracts concrete frameworks, principles, and real cases, passing them through rigorous verification and pressure tests. This turns static knowledge from masterpieces like Warren Buffett’s letters or Sun Tzu’s Art of War into modular, callable skills that you can directly install into your coding environment, allowing your AI assistant to actively apply those battle-tested strategies to your real-world decisions.

🆔 @hackernewsgithubprojects

📦 aieradev/clypra

Clypra

Clypra is a modern video editor that finally brings professional, premium-grade capabilities to a completely free, open-source desktop and mobile app. If you want high-end editing tools without coughing up a monthly subscription or dealing with ugly watermarks, this is your solution. It runs on a lightning-fast Rust and Tauri engine, which means you get buttery-smooth, hardware-accelerated timeline scrubbing, multi-track audio, and smart frame-accurate trimming. The coolest part is how it uses predictive decoder prewarming to completely eliminate lag when you hit play. Download Clypra today and take full control of your video editing workflow.

🆔 @hackernewsgithubprojects

📦 chenhg5/cc-connect

cc-connect

You can now control the AI coding agents running on your local computer directly from your phone using Slack, Discord, or Telegram. A clever open source project called cc-connect acts as a secure bridge that links local tools like Claude Code, Cursor, and Gemini CLI to your favorite messaging apps. It allows you to check in on long-running tests, trigger code reviews, or run terminal commands while you are away from your desk, without needing a public IP address. It is like having a software engineer in your pocket who is ready to deploy your code on command. Grab it today and start chatting with your workspace.

🆔 @hackernewsgithubprojects

📦 floodtide/dom-docx

dom-docx

dom-docx is the smart library that converts semantic HTML fragments into completely editable Word documents. Instead of relying on sloppy layout hacks or static screenshots, it maps your standard headings, lists, tables, and paragraphs directly into native Word formatting that you can click and type in. The best part is its optional rasterization trick, which can automatically grab complex charts or canvas drawings and cleanly embed them as sharp, high-density images inside the final document. It is the perfect, lightweight tool for developers who need to instantly generate clean, professional reports straight from their existing web content without any heavy overhead.

📰 https://news.ycombinator.com/item?id=48891267

🆔 @hackernewsgithubprojects

📦 rejourneyco/rejourney

Rejourney

Find hidden bugs and friction points that are quietly draining your app's revenue by monitoring actual user sessions. Rejourney is a lightweight, open-source observability platform for web and mobile apps that automatically tracks critical business transitions like signups, checkouts, and trial renewals. It compares your active users against a healthy baseline to instantly highlight where people are getting stuck or running into errors. Instead of guessing why your conversion rate dropped, you can watch precise session replays, heatmaps, and api errors side-by-side to pinpoint the exact issue. Set it up on your own servers to start plugging those revenue leaks today.

📰 https://news.ycombinator.com/item?id=48904912

🆔 @hackernewsgithubprojects

📦 marvinschwaibold/mri-grid-viewer

mri-grid-viewer

Browse and annotate medical scans right on your device with complete privacy. This clever project lets you load sensitive medical images into a local-only interface that runs entirely in your web browser or as an iOS app. Instead of uploading patient data to a third-party server, it processes everything on your own machine. You can search metadata, filter by anatomy, zoom in on frames, and even draw freehand annotations or add comments directly on the images. It is a brilliant way to inspect complex scans securely without any network dependency. Check it out to build your own private medical imaging sandbox.

📰 https://news.ycombinator.com/item?id=48880239

🆔 @hackernewsgithubprojects