299 subscribers
14K videos
14.6K links
Download Telegram
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ sshah03/recalc

recalc turns a spreadsheet into a live dashboard for your AI agents

recalc turns a standard spreadsheet into a live dashboard where every single cell can be a terminal window running an artificial intelligence agent. You can set up a grid where an agent writes its code in one cell and a test script in the cell right next to it. If the agent claims the code works but the test fails, the cell turns red instantly so you can spot the lie before it ships. It is built for people who use coding agents but hate losing track of what they are doing. Instead of juggling five different chat windows, you watch them all work on one shared screen.

πŸ“° https://news.ycombinator.com/item?id=50413970

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ shuttle-browser/shuttle

Shuttle: The Browser Built for AI Agents

Shuttle is a stripped-down Chromium fork that gives every artificial intelligence agent its own logged-in identity and its own set of tabs right inside your regular browser window. It solves a massive headache where agents used to steal your cookies or get locked out of your accounts entirely. Now, each AI gets its own persistent storage and a built-in connection layer, so it stays logged in and works in total isolation. You can watch the agent browse, grab a tab, or kick it out of your session without any extra software. It is the first browser that actually treats AI agents as first-class citizens.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ yihaohu0118/viss

ViSS: The Video Model That Decides What To Watch

A vision language model can now decide exactly which video frames to inspect before answering a question. This project, called Viss, separates what a model knows from how it looks at the evidence. Instead of forcing the system to process every single frame of a long video, it starts with the recent moments and only fetches specific past clips if it is unsure. This saves massive computing power while keeping the model accurate. It is a fresh approach to making streaming video understanding efficient and reliable. If you are building systems that watch live feeds, this is the one to study.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ pmndrs/klipp

Klipp: The Cinemachine for the Web

Direct your own cinematic shots in any browser-based three-dimensional scene. Klipp brings the power of game engine virtual cameras to the web, letting you define multiple camera angles with specific priorities. When you switch between them, it doesn't just jump; it blends the position and rotation smoothly, creating those high-end transitions seen in AAA titles. Think of it as a director for your code, handling the tricky math of following targets and framing groups so your camera feels alive instead of static. It runs on pure math, so it works with any renderer you choose.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ cx-artlab/agent-robot-avatar

Give Your AI a Face With Agent Robot Avatar

Drop a living robot face into your web app that actually feels alive. Agent Robot Avatar is a tiny piece of code that turns a blank space into a character with eyes that follow your mouse and a body that squishes when you poke it. The coolest part is how it reacts to your AI’s work. When your chatbot is thinking, the robot looks away and waits. When it finishes, it celebrates or looks sad if something went wrong. It is built with simple web tools and no heavy software, so it loads instantly.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ shawnthebeachy/return-affinity-colors

Get Your Affinity Colors Back

Affinity Designer for Version Three launched with every tool icon stripped of its color, leaving your interface feeling stark and empty. A developer named Shawn The Beachy built a tool called Return Affinity Colors to fix this. It works by digging into the app's core files and swapping those blank gray symbols for the vibrant icons from the previous version. You can even dump the original graphics, tweak them, and import your own custom set. It is a simple command line utility that restores the visual charm many users missed.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ qinggo/engram-peft

Engram-PEFT: The 240x Memory Trick for AI

Engram-PEFT lets you inject massive amounts of knowledge into large language models without slowing them down at all. This unofficial implementation of the DeepSeek Engram paper uses a clever trick called sparse retrieval to add a high-capacity memory layer that sits alongside the model. The magic is that it only activates a tiny fraction of that memory during each step of inference, meaning you get huge storage for facts without the usual computational overhead. It is essentially a way to give an AI a giant library of specific data without making it think slower.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ uclanlp/openvlharness

OpenVLHarness: The Visual Reasoning Upgrade

OpenVLHarness is the visual reasoning harness that turns any vision model into a detective. It works by giving the model access to a toolkit of specialized services, like a zoom lens, a text reader, and even a web browser. This allows the AI to look closer at details, read signs, and search for context online before answering. The real magic happens in its memory system. Unlike standard models that forget previous steps, this system keeps track of everything it sees, like boxes and images, so it can build on its own discoveries. The result is a massive jump in accuracy without retraining the model.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ cairn-framework/cairn

Cairn: The Architectural Memory for AI Agents

Cairn is the architecture memory that stops your AI agent from forgetting your codebase. Instead of re-reading every file from scratch each session, your agent drafts a plain-text map of your project structure while you work. Cairn keeps this map in sync with your actual code, flagging any drift before it causes problems. This means every new session starts with full context, respecting decisions made weeks ago rather than breaking them. It works locally, keeps your data private, and turns architectural knowledge into simple, reviewable files you own. No more explaining your system over and over.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ ak375456/privacycam

PrivacyCam: Keep Your Photos Private

Scrub sensitive details like faces, license plates, and credit card numbers right out of your photos, videos, and PDFs before you share them. PrivacyCam is a mobile app that runs all of that detection and redaction locally on your phone, meaning your images never touch a cloud server or a developer's computer. It automatically finds the stuff you probably want to hide, but you stay in control of exactly what gets covered up. It’s the digital equivalent of checking your rearview mirror before you drive off. If you value your privacy, this is a tool worth keeping in your back pocket.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ chiyunnoh/trail-odom

TRaIL-Odom

TRaIL-Odom is the motion tracking system that adapts radar data to weak LiDAR geometry. This project helps robots navigate through tunnels and open fields where standard sensors fail. It looks at each laser scan to find directions where the robot is unsure. Then it boosts the radar signals specifically for those weak spots. This adaptive weighting cuts positioning errors by eighty-six percent in tricky areas. The result is a robot that knows exactly where it is, even when the environment is confusing. It is a precise tool for reliable autonomous movement.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ rxjfighting/scenemosaic

SceneMosaic: From Photo to 3D

SceneMosaic takes a single indoor photo and builds a fully editable three-dimensional scene you can actually use in robotics simulations. It doesn't just guess where the sofa goes; it uses AI agents to actively debate the layout, checking for collisions and ensuring chairs face the table correctly. This hybrid approach turns a flat image into a structured digital twin, solving the massive problem of needing manual modeling for every new environment. If you're building autonomous systems, this bridge from real-world photos to simulation-ready assets is a huge time saver. The result is a diverse set of scene variants ready for your next test run.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ cocacola-lab/recursia

Recursia: The Robot That Writes Its Own Code

Teach a robot to pick up a mug, and then watch it figure out how to pick up a carton of milk by writing its own software. Recursia is a system that allows a physical robot arm to improve its abilities not by adjusting its brain, but by literally writing Python code. It runs on a Franka arm, testing its new programs against real objects to ensure they work before saving them. If the code fails, the robot fixes it. If it works, that code becomes a permanent skill. This means the robot learns through executable logic rather than fuzzy math.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ annchenn/manga2animate

Manga2Animate: Turning Comics into Faithful Animation

Turn static comic pages into a coherent video that actually looks like the original art. Manga2Animate solves a tricky problem: most AI video tools get the story wrong or lose the specific character designs when you feed them comic panels. Instead of guessing, this tool acts like a professional animation studio. It writes a detailed script and draws a visual storyboard first, then uses AI to generate the final video from those plans. The result is an animation that stays true to the source material’s style and story, giving you a much more faithful and consistent experience than standard video generators.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ vals-ai/valkyrie

How Valkyrie Tests AI Agents at Scale

Valkyrie lets you spin up hundreds of isolated cloud environments in seconds to stress test AI agents against complex real-world tasks. Testing an AI agent is messy because every task requires a fresh, clean computer environment to prevent data leaks between runs. Valkyrie solves this by managing the entire infrastructure for you, automatically spinning up and tearing down these isolated spaces so you can run thousands of evaluations simultaneously. You simply define your agent, pick a benchmark, and let the platform handle the heavy lifting of cloud orchestration. This makes it easy to verify that your AI actually works under pressure before you release it.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ seerray-lab/xiaomi-lens-0

Xiaomi Lens Zero: The AI That Actually Sees

Xiaomi Lens Zero is the compact vision model that finally lets an AI point at exactly what it means. Built on a four billion parameter base, this tool solves a frustrating problem where language models can describe objects but cannot reliably show you which specific pixel belongs to them. The single most impressive feature is its ability to generate pixel-perfect masks directly while reasoning, meaning it can visually isolate a specific apple in a pile or track a moving car in real time. Instead of just guessing coordinates, it physically highlights the region it is discussing, making its visual understanding concrete and verifiable.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ adatutorank/adatutorank

AdaTutoRank: Smarter AI Research

AdaTutoRank reorders search results to find answers faster. It fixes how A I systems pick documents by teaching them to rank entire sets of sources, not just single pages. This matters because standard search often misses the full context needed for complex questions. The project uses a new training method called Adaptive Tutoring Optimization to improve accuracy. It helps A I assistants understand what is missing in their search results. This leads to better answers for deep research tasks. The code is open source and ready to use. You can test it on real search problems. It shows how better ranking leads to smarter A I.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ sandlbn/polyend-tracker-re

OpenTracker: 16 Tracks on Original Polyend Tracker

OpenTracker breathes new life into the original Polyend Tracker by running a completely open-source firmware directly on the hardware. The biggest surprise is that it adds a full sixteen-track sequencer to a machine originally built for eight. This clean-room rebuild lets you load projects from the newer Mini model and play three custom synth engines, all without using a single line of proprietary code. It is a fascinating look at how far community reverse engineering can push dedicated music hardware. You get modern features on vintage gear without any vendor support. It is a perfect example of what happens when developers take control of their tools.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ willwaltz/globes

Watch Water Slosh in a Glass Globe

Globes is the web demo that actually lets you shake a glass sphere of water and watch it slosh, splash, and settle in real time. It runs the entire fluid simulation directly on your graphics card using millions of tiny particles, meaning the water behaves with genuine physical weight rather than just looking like a video. You can tilt the globe to spill the water, drop food for the koi, or even place the window on your desktop to bend your wallpaper through the glass.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ paulomarques742/imago

Imago: The Photo Editor That Never Messes Up Your Originals

Imago lets you fix your photos without ever permanently changing the original file. Instead of saving a new image every time you tweak the lighting or color, it records your changes as a simple recipe you can tweak or delete at any moment. This matters because it keeps your photo library clean, avoiding the messy pile of duplicates that usually happens after editing. It runs entirely on your own phone or computer, so your private pictures never get sent to a stranger's server. It is built for people who care about their data and want to edit safely without losing the original shot.

πŸ†” @hackernewsgithubprojects