📦 caamer20/telegram-drive

Telegram Drive

You can transform your Telegram account into an unlimited, secure cloud storage drive using a clever open-source application. This desktop tool turns your saved messages and private channels into a fully functional file explorer. Instead of digging through endless chat logs, you get an intuitive dashboard where you can drag and drop files, stream media directly without downloading, and organize your data into custom folders. Everything is powered locally using your own Telegram connection, keeping your data private without third-party servers. It is a brilliant way to turn a messaging app into your own personal, cost-free cloud.

🆔 @hackernewsgithubprojects

📦 ddlmanus/macoptimizer

MacOptimizer

Clean up your cluttered Mac and reclaim lost disk space with MacOptimizer, a slick, lightweight utility built entirely in native SwiftUI. Instead of paying a subscription for bloated cleaner apps, this open-source tool tackles the single most frustrating macOS problem: leftover junk. Its deep-cleaning scanner finds the hidden preference folders, caches, and orphaned containers that standard app deletions leave behind. It even includes a file explorer with quick terminal integration and live system monitoring to kill runaway processes. Grab the free download to instantly sweep away gigabytes of digital dust and keep your machine running light.

🆔 @hackernewsgithubprojects

📦 noamseg/interview-coach-skill

Interview Coach Skill

Interview-coach-skill is the terminal-based assistant that transforms how you prepare for job interviews by acting as an adaptive, highly personalized coach right in your editor. Instead of just giving you a generic list of questions, this system scores your answers across five distinct dimensions and actually diagnoses the psychological root causes behind your weak spots, like narrative hoarding or conflict avoidance. It builds a custom storybank, runs you through constraint and pushback drills, and even tracks how well its mock scores predict your real-world outcomes. It is a complete, self-correcting game plan that gets smarter the more you practice.

🆔 @hackernewsgithubprojects

📦 dennisklappe/cloudmeet

cloudmeet

cloudmeet is the open-source meeting scheduler that runs entirely on Cloudflare's free tier. This clever tool serves as a self-hosted substitute for booking platforms, letting you coordinate appointments without monthly subscription fees. By leveraging lightweight edge workers, it connects directly with your Google Calendar and Outlook accounts to check your availability in real time and automatically generate Google Meet or Microsoft Teams links. It even handles custom availability rules and background email reminders. It is the perfect weekend project for anyone wanting full control over their calendar links with zero hosting overhead.

🆔 @hackernewsgithubprojects

📦 vanilagy/turbores

turbores

turbores is the browser-based video decoder that finally decodes Apple ProRes video at blistering speeds. Built entirely from scratch using Zig and TypeScript, this lightweight tool compiles to WebAssembly to bypass heavy hardware requirements and decode professional-grade video formats right inside a standard web browser. While typical solutions struggle with latency, this library runs over twice as fast as native FFmpeg, achieving hundreds of frames per second even on complex 4K files. By using clever memory allocation and parallel processing, it brings high-performance desktop video editing workflows directly to your everyday web applications. Try it out to experience instant, stutter-free video decoding on the web.

📰 https://news.ycombinator.com/item?id=48722042

🆔 @hackernewsgithubprojects

📦 0xmassi/pocketdev

pocketdev

Spin up a secure, dedicated cloud development environment with a single command so you can code directly from your phone. pocketdev provisions a private Hetzner server, locks it inside your personal Tailscale network with zero public ports open, and instantly installs your favorite AI coding helpers like Claude Code or Cursor. Since your configuration is safely stored on your laptop, you can securely access your setup from anywhere. Just scan a QR code on your phone to connect your favorite SSH client, lock in a persistent session, and start building on the go. It is a brilliant way to write code on a real machine right from your pocket.

🆔 @hackernewsgithubprojects

📦 hellod035/now_you_see_that

Now You See That

Humanoid robots can now successfully navigate highly complex obstacles using nothing but raw visual data from on-board cameras. A research project called now_you_see_that has developed a system that lets these robots walk over gaps, climb steep stairs, and step over debris by processing raw depth pixels directly into physical movement. Traditionally, translating computer simulation to real-world cameras is difficult because real sensors are noisy and imperfect. By simulating realistic camera errors and training the artificial intelligence to ignore visual noise, researchers managed to deploy this single control system across different robot designs. It proves that walking machines can safely navigate unpredictable environments using basic vision.

🆔 @hackernewsgithubprojects

📦 mig-nju/evoembedding

EvoEmbedding

EvoEmbedding is the text retrieval model that updates its own memory on the fly to help computers search through massive histories without losing track of time. Standard models look at chunks of text in isolation, totally ignoring what came before and after. This model processes text sequentially and updates a continuous memory queue to generate embeddings that are highly sensitive to chronological order. This makes it incredibly good at answering timeline-based questions, like identifying what happened first or last in a long conversation. If you are building memory-heavy AI tools, this project proves that smart retrieval is all about preserving context.

🆔 @hackernewsgithubprojects

📦 visomasterfusion/visomaster-fusion

VisoMaster Fusion

You can now edit expressions, swap multiple faces, and run high-end video enhancements directly on your desktop without touching a single line of code. visomaster-fusion is a powerful desktop application that puts pro-level face editing, swapping, and restoration into a highly polished graphical interface. Instead of wrestling with complex command-line scripts, you get a clean workspace featuring multi-face tracking, color matching, and deep model controls that run locally on your own graphics card. It simplifies the entire workflow by letting you manage saved face profiles and preview edits in real time before rendering. Grab the launcher, run the setup, and start creating seamless edits instantly.

🆔 @hackernewsgithubprojects

📦 akin-ozer/cc-devops-skills

cc-devops-skills

Supercharge your AI development workflows by automatically generating and validating production-ready infrastructure configs. The cc-devops-skills repository provides a practical pack of thirty-one specialized tools designed specifically for Claude Code and Codex. Instead of struggling with manual syntax errors, this project gives your AI assistant sixteen automated generators to instantly scaffold pipelines, alongside fourteen intelligent validators to run local security audits, linting, and dry-run checks for platforms like Terraform, Kubernetes, and GitHub Actions. It bridges the gap between raw AI code generation and actual, working deployments, giving developers a safe, rapid way to build and verify their operations pipelines.

🆔 @hackernewsgithubprojects

📦 kulaxyz/self-learning-skills

Self-Learning Skills

self-learning-skills is the standing-instructions utility that finally lets your AI coding agents learn from their own mistakes. Every developer knows the frustration of watching an AI assistant struggle through a complex debugging loop, only to completely forget the solution once the session ends. This clever tool monitors your coding sessions, automatically detects when the agent hits a hard-won breakthrough, and saves that exact workflow into a reusable markdown skill. The next time you start a session, your agent instantly loads this local memory, skipping the dead ends and immediately following the proven golden path to success.

📰 https://news.ycombinator.com/item?id=48713137

🆔 @hackernewsgithubprojects

📦 lightmode-laboratories/spacesuit

Spacesuit

Organize your thoughts on an infinite digital workspace without the hassle of rigid folders or complex hierarchies. A desktop app called spacesuit gives you a visual canvas where you can place resizable note cards, connect different ideas, and even pull in live website snippets that stay visible without constant reloading. It brings together built-in artificial intelligence assistants to help you process your notes, alongside sensory micro-interactions to help you pause and refocus. Best of all, your data stays entirely on your own device or syncs safely to your private cloud. It is a peaceful, flexible hub designed to make sense of your personal projects.

🆔 @hackernewsgithubprojects

📦 palemoky/chinese-poetry-api

chinese-poetry-api

Build a high-performance database of classical Chinese poetry with both simplified and traditional characters automatically generated. This tool parses raw data files to build a highly optimized SQLite database complete with authors, dynasties, and poetic genres. Even better, it serves all this rich historical data instantly through both standard web endpoints and modern graph queries. It is perfect for developers who want to build educational apps or search engines without worrying about custom database schemas or slow performance. Set up this service to instantly bring thousands of years of classic literature into your next project with ease.

🆔 @hackernewsgithubprojects

📦 auriti-labs/geo-optimizer-skill

Geo Ready GEO Optimizer

Check whether AI search engines like ChatGPT, Perplexity, and Google AI Overviews are actually citing your website. The geo-optimizer-skill toolkit runs a quick audit on your site to score its AI search readiness from zero to one hundred. It scans your robots.txt, checks for an llms.txt file, and analyzes your schema markup to make sure AI bots can easily crawl and quote your content. You can even run live queries to see if your brand gets mentioned when users ask questions. It is a brilliant way to make sure your business does not become invisible in the age of AI search.

🆔 @hackernewsgithubprojects

📦 unimpor/t3

t3

AI assistants struggle when they do not have all the facts, often making bad moves because they lose track of what they actually know. An open-source project called t3 fixes this by training language models to actively seek out missing information and maintain an accurate picture of their environment during multi-step tasks. By introducing clever training methods that reward models for uncovering new details and penalize them for repetitive actions, it stops agents from getting trapped in unproductive loops. If you want to see how reinforcement learning can build smarter agents that actually know how to ask the right questions, this codebase is well worth a look.

🆔 @hackernewsgithubprojects

📦 gtrg55/trellis2-mlx

Trellis 2 MLX

Trellis 2 MLX is the open-source pipeline that finally brings high-resolution, image-to-3D generation natively to Apple Silicon. Instead of relying on power-hungry, cloud-hosted Nvidia graphics cards, this project runs Microsoft's massive four-billion-parameter model locally on your Mac using the MLX framework. The coolest part is how it handles textures. While most local generators produce messy, semi-transparent points, this tool uses smart chunked sampling to construct incredibly detailed, solid 3D meshes with fully baked color and roughness. You can export complete, textured objects straight into Blender or game engines without needing a massive server rack.

🆔 @hackernewsgithubprojects

📦 homemadegarbage/selfrisingrobot

SelfRisingRobot

An incredibly simple two-axis toy robot can learn to stand back up on its own after being knocked down in any direction. This project, called selfrisingrobot, makes reinforcement learning accessible by bridging virtual simulation and physical hardware. It uses a physics engine to train a virtual model of a tiny robot with just two micro servos. Once the computer brain learns to wiggle and push itself upright in the simulator, that identical code is flashed onto a real microcontroller to run on a physical, three-dimensionally printed robot. It is a fantastic hands-on demonstration of simulated intelligence translating perfectly into physical action.

🆔 @hackernewsgithubprojects

📦 thewakesystems/riverone

RiverONE

Compress multimodal AI models to a fraction of their size by treating neural network weight reduction as a simulated quantum computing problem. RiverONE runs entirely on classical hardware, translating quantum mechanics concepts into practical machine learning pipelines. It collapses continuous model weights into discrete, quantum-inspired codebooks, entangles adjacent layers in vision transformers so they share parameters through simulated unitary gates, and recovers lost accuracy using an optimization loop modeled after the variational quantum eigensolver. It is a wildly creative way to shrink heavy vision-language models down to pocket-sized, runnable software.

🆔 @hackernewsgithubprojects

📦 map-raysup/raysup

RaysUp Feature Upsampling

AI models generate incredibly smart feature maps, but they are often too blurry and low-resolution to be useful for precise visual tasks. RaysUp solves this problem by using a clever geometry-aware ray representation that upsamples these complex neural features with extreme precision. Instead of relying on heavy, slow neural networks to guess the missing pixels, this method uses lightweight geometric rays to guide the upsampling process. This makes the entire operation ultra-light and incredibly fast, while keeping the structural details perfectly sharp. It is a massive win for tasks like semantic segmentation and video tracking, proving that we can get high-definition AI features without the massive computational cost.

🆔 @hackernewsgithubprojects

📦 shootthesound/comfyui-angelo

ComfyUI Angelo

The comfyui-angelo tool completely replaces your complex, friction-heavy image editing workflows with a single, highly interactive node. Instead of jumping between mask editors, pad nodes, and separate samplers to fix a detail, you simply click, paint, or drag directly on your image preview to edit. It lets you refine specific areas immediately, auto-detect faces for quick fixes, and expand your canvas with smart outpainting while keeping the rest of your image bit-exact. You can load and restore old photos using a simple slider that perfectly balances original details with fresh, high-resolution textures. It is the ultimate hands-on playground for precise image refinement.

🆔 @hackernewsgithubprojects