📦 conglu1997/local-autoresearch

local-autoresearch

Let a local language model run autonomous experiments on your MacBook. The project called local-autoresearch uses a model named Flint to iteratively improve code for geometry, scheduling, and machine learning tasks. It proposes changes, tests them against strict limits, and keeps the best results. One standout result is a thirty percent reduction in bike rental prediction error. This runs entirely offline on your machine. It shows how automated research works locally. You get a clear view of the model's reasoning. Try it to see local AI improve code step by step.

🆔 @hackernewsgithubprojects

📦 d2i-group/tsflab

TSFLab: The Ultimate Forecasting Leaderboard

TSFLab is the forecasting platform that puts every model on the same playing field. It solves the messy problem of comparing different prediction tools by running over one hundred and ninety of them on identical data. The coolest part is the public leaderboard, where you can see exactly how well each model performs on real-world problems like electricity usage or stock prices. It is built for agents, meaning it helps computer programs pick the best tool for the job automatically. This gives you a clear, honest look at what actually works, without the marketing fluff. It is the ultimate way to know if your prediction method is actually good.

🆔 @hackernewsgithubprojects

📦 we-are-mono/verso

Verso: A Modern Web UI for OpenWrt

Replace your router's dated interface with a modern web dashboard. Verso is a new front-end for OpenWrt that swaps out the old LuCI for a clean, card-based view of your network. Instead of digging through confusing menus, you see exactly what is happening on your connection, ports, and system resources at a glance. It runs directly on your router as a simple package, giving you a fast, secure way to manage your home network without any heavy software. It is an experimental tool, so back up your settings before installing. This project makes router management feel like using a modern app rather than reading a manual.

🆔 @hackernewsgithubprojects

📦 luiscarlosjayk/tiza

Draw on your code with Tiza

Grab your code and sketch right on top of it. Tiza turns your terminal into a whiteboard where you draw boxes, arrows, and notes directly over the source text. This solves the age-old problem of explaining complex logic without switching to a separate presentation tool. The coolest part? The drawings stick to the text. If you scroll or resize your window, the annotations move with the code, keeping your explanation perfectly aligned. It is built for teaching, debugging, or recording tutorials without the usual layout headaches. Next time you need to walk someone through a tricky function, try this.

🆔 @hackernewsgithubprojects

📦 sardorml/captureflow

CaptureFlow: The Screen Recorder For Your AI Agents

Your AI agent just finished a task, and instead of trusting its word, you get a video proof. That is exactly what CaptureFlow does. It is an open-source screen recorder built specifically for agentic workflows, letting you capture what your AI is doing on your screen in real time. The coolest part is that the share link is ready the moment you stop recording because the upload happens in the background while you are still recording. It solves the trust gap with AI by giving you visual evidence of its actions, and you can self-host it on your own servers to keep your data private.

🆔 @hackernewsgithubprojects

📦 occupysaas/zeroauthy

Zeroauthy: The Self-Hosted Auth0 Replacement

You can point your existing Auth0 SDKs at your own server and they just work. Zeroauthy is an open-source, self-hostable identity server that recreates the Auth0 Authentication API from public documentation and open standards. It is a clean-room implementation, meaning it is not affiliated with Auth0 or Okta, but it speaks the same language. If you are tired of relying on a hosted provider, this lets you take control of your authentication infrastructure. It handles everything from login pages and token issuance to user management and event streaming. The project is built in Go, ensuring it is fast and efficient.

🆔 @hackernewsgithubprojects

📦 ctxd-dev/ctxd

ctxd: Making Your Files Visible to AI

Your AI agents currently cannot see the documents on your computer. The tool called Ctxd fixes this by turning your local folders into a searchable context that artificial intelligence can actually read and use. You simply point the software at the directories you want shared, and it tracks those files, exposing them through a secure, local service. This allows an agent to search your specific work files for keywords, read exact line ranges, or list directory contents, all without sending your private data to a distant cloud. It essentially builds a bridge between your local file system and the intelligence running on top of it.

🆔 @hackernewsgithubprojects

📦 romainguy/vibrance

Vibrance: Real-World Paint Mixing for Code

Vibrance makes digital colors mix exactly like physical paint. Standard software blends light, but this library simulates subtractive mixing, so blue and yellow actually create green instead of muddy gray. It uses a tiny neural network trained on real pigment data to calculate these natural gradients instantly on your phone or computer. The result is visuals that feel organic and vibrant, not flat. If you want your app’s colors to behave like an artist’s palette rather than a screen’s light, this is the tool to explore.

🆔 @hackernewsgithubprojects

📦 automattic/static-site-importer

Static Site Importer

You can drop a plain zip file of old HTML pages into a WordPress site and watch it instantly become a fully editable block theme with working navigation and proper permalinks. That is exactly what Static Site Importer does. It takes static HTML files, pasted code, or a simple folder of pages and converts them into native WordPress content without needing to manually rebuild your site in a visual editor. This is useful because it turns dead, static code into live, manageable pages where you can actually edit text and update links later. It handles the messy work of linking pages and setting up headers automatically.

🆔 @hackernewsgithubprojects

📦 sanderland/ctok

CtOk Counts Claude Tokens Offline

CtOk counts exactly how many tokens a string uses in Claude, right on your machine, without calling any API. That is the surprising part. It works by guessing how Anthropic's system breaks your text into pieces, using a set of rules learned from real data. This is useful if you need to know your costs or limit lengths before you hit the server. It is not perfect because the exact internal logic is a secret, but it gets you the right number almost every time. It is a clever hack for anyone trying to predict their bill. Try it out and see if it matches your actual usage.

📰 https://news.ycombinator.com/item?id=49313865

🆔 @hackernewsgithubprojects

📦 lucianoneo/pokemon-pokopia-pt-br

Pokémon Pokopia Brazilian Translation

A single person has translated over thirty thousand text entries for the Pokémon Pokopia game into Brazilian Portuguese. This unofficial mod replaces every piece of interface text, menu label, and item description with local Portuguese, while keeping official Pokémon names in English. The project is useful because it allows Portuguese-speaking players to understand the game entirely without switching languages, and it includes a simple Windows installer that automatically backs up original files. The mod covers both game versions, ensuring full coverage of every cataloged string. It is a precise community effort that makes a niche game accessible through careful manual review and automation.

🆔 @hackernewsgithubprojects

📦 windows-gpu-unlock/windows-gpu-unlock

Windows GPU Unlock: Boot RTX Cards Without a Reset

Windows GPU Unlock is the experimental boot loader that finally lets specific Nvidia graphics cards run on Windows without needing a second monitor or a full system reset. The project works by carefully guiding your computer through a special startup process that releases the graphics card from its initial locked state, allowing it to hand off control directly to your operating system. This solves the frustrating problem of needing extra hardware or complicated manual steps just to get your graphics card working correctly at boot. It is a fascinating look at how deep hardware tinkering can go, turning a rigid system process into something much more flexible for enthusiasts.

🆔 @hackernewsgithubprojects

📦 anemll/flash-moe

Flash-MoE: Running a 400B Model on a Mac

Flash-MoE is the inference engine that streams a massive language model from your solid state drive. Instead of fitting the entire network into memory, this pure C project reads only the specific parts needed for each word. It processes a model with four hundred billion parameters on a MacBook Pro by pulling data directly from the disk. This design avoids the need for expensive server clusters. It proves that local hardware can handle very large artificial intelligence tasks. You get real-time responses without the cloud. This approach changes how we think about local computing power. It makes high-end language models accessible to everyone.

🆔 @hackernewsgithubprojects

📦 shenysun/skills-manager

How to Manage All Your AI Agents with One Command

Skills Manager treats your AI helper abilities like a shared library that updates itself without asking permission. Instead of manually copying files between different coding tools, you install one central hub and let your chat agent handle the rest. The project solves the messy problem of keeping skills in sync across dozens of different applications. You simply tell your chat bot what to install, and it manages the files behind the scenes. This approach saves time and prevents outdated skills from causing errors. It is a simple way to organize your AI toolkit without technical overhead.

🆔 @hackernewsgithubprojects

📦 super-phenix/superphenix

Superphenix: Your Own Cloud

Superphenix is an open-source platform that lets you build a full cloud provider on the hardware you already own. It runs on Kubernetes to manage everything from virtual machines and storage to networking and managed services, all running as first-class workloads on top. This solves the problem of vendor lock-in and high costs by giving you complete control over your infrastructure and data. Whether you are building a public cloud or a private one, it is designed to be resilient and scalable across multiple datacenters. It is already being used in production by companies looking for independence and total control.

📰 https://news.ycombinator.com/item?id=49714404

🆔 @hackernewsgithubprojects

📦 git-fed/museai-skills-guide

Museai Skills Guide: The Offline Audit Ledger

Track every move an A I agent makes without needing secret server logs. This guide provides a private, offline tool that records actions locally on your device, so you can see exactly what happened during a task. It solves the problem of blind trust by giving you a simple ledger to log evidence one step at a time. Instead of guessing whether a tool worked or what data was sent, you get a clear, permanent record stored right in your browser. This project is a practical field guide for anyone who wants to understand how these assistants actually operate. It keeps your data private and your questions answered.

🆔 @hackernewsgithubprojects

📦 tencentyouturesearch/palm-infra

Run 100B+ AI Models on Your Laptop with Palm-Infra

Run massive one hundred billion parameter AI models directly on your laptop or a single workstation. Palm-Infra solves the memory bottleneck of large language models by using a clever SSD offloading technique. Instead of forcing your computer to hold the entire model in its limited fast memory, it keeps the core data there and streams only the specific parts needed for each word from your hard drive. This allows you to run complex expert models that would normally require expensive data center hardware, right from your local machine. The project wraps this logic in a simple C++ engine that converts standard models into a single portable file.

🆔 @hackernewsgithubprojects