56 subscribers
7.27K videos
7.87K links
Download Telegram
This media is not supported in your browser
VIEW IN TELEGRAM
📦 addyosmani/agentic-seo

agentic-seo

The agentic-seo project is the auditing tool that finally grades your documentation on how well AI coding agents can read and digest it. See, tools like Claude Code and Cursor do not browse documentation like humans do; they rip out the HTML, count tokens, and discard your content if it is formatted poorly or restricted. This command-line utility runs checks across your site to score things like robot rules, token budgets, and structured indexes so AI assistants do not get locked out. Run the audit to get an immediate grade, build missing configuration files, and ensure your developer documentation is fully optimized for the artificial intelligence era.

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 echovic/blade-deepseek

Blade DeepSeek

Run a fully autonomous coding assistant right from your terminal by using blade-deepseek. This Rust-based tool acts as a local coding companion designed specifically to work with DeepSeek's advanced reasoning models. Instead of just answering questions, it runs a smart, multi-turn loop that can search your workspace, edit files, and even spin up child agents to delegate smaller tasks. What makes it incredibly handy is its persistent goal mode, which keeps the assistant working through complex, multi-step refactoring jobs even if you restart your session. It is the perfect terminal companion for developers looking to automate repetitive coding chores.

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 breschio/drawbridge

Drawbridge

You can now leave Figma-style comments on your live web app and watch your AI coding assistant instantly build the fixes. A clever Chrome extension called drawbridge lets you click on any interface element or draw a freeform rectangle over layout bugs to leave a note. Instead of keeping these feedback notes trapped in a browser, it automatically syncs them straight into your local codebase as clean markdown and structured data files. When you run a simple slash command in tools like Cursor or Claude Code, your AI assistant reads the task queue, grabs the cropped screenshots, and starts writing the code to fix the exact UI issues you just...

📰 https://news.ycombinator.com/item?id=48791105

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 apple/foundation-models-utilities

foundation-models-utilities

foundation-models-utilities is the Swift companion library that finally makes managing conversational context for local and hosted AI models a breeze. Instead of wrestling with massive prompt histories that break your context windows, this package introduces clever history management modifiers to gracefully summarize old conversations, drop completed tool calls, or roll active windows on the fly. It even features a dynamic skill system that lets models spin up custom style guides or calendar tools just-in-time via tool calls, preventing context pollution and keeping things lightning fast. It is the perfect toolkit to make your next Swift AI app much smarter.

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 boona13/glb-shrink

GLB Shrink

This brilliant tool can shrink bloated fifty-megabyte 3D models down to less than one megabyte in just seconds. Called glb-shrink, it is an open-source web application designed to turn heavy scans and AI-generated models into light, production-ready assets for web apps and games. You simply drag and drop any heavy file into your browser, adjust a simple slider between file size and visual sharpness, and watch the tool automatically simplify geometry, compress textures to WebP, and apply Draco encoding. With a live side-by-side comparison, it is the perfect way to optimize assets without opening complex modeling software.

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 study-overflow/mbench

mbench

Test how well your long-video AI models actually remember the world they are creating. While most benchmarks only look at single frames or quick clips, mbench evaluates if an AI can keep things consistent over time. It measures whether a character stays the same after walking off-screen, if a room layout changes when the camera spins around, and if physical actions continue naturally. It even uses a clever trigger system to stop models from cheating by just generating static, frozen videos. It is the perfect reality check for anyone building next-generation video generators, ensuring their virtual worlds do not fall apart the second you look away.

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 ai4s-research/open-science

Open Science

The open-source desktop app that finally gives researchers a local-first alternative to commercial AI science assistants. While typical AI tools restrict you to a simple chat window, this platform connects your literature searches, data analysis, and document drafting into a single verifiable system on your own computer. Its standout feature is automatic artifact traceability, which means every chart, table, or report generated by the AI is permanently linked to the exact code, data, and conversation history that created it. It even audits citations and flags unverified numbers to ensure your work is fully reproducible. Say hello to a more transparent way to do research.

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 rshah515/claude-code-subagents

claude-code-subagents

claude-code-subagents is the ultimate productivity booster that transforms your terminal assistant into an entire software company of hyper-specialized experts. Instead of relying on a single, generic artificial intelligence to handle your codebase, this project hands you a massive toolkit of over one hundred and sixty tailored personalities. Whether you need a wizard for Rust memory safety, a dedicated database architect for PostgreSQL, or even a HIPAA compliance expert for healthcare security, you can spin up the exact specialist you need instantly. It turns your development process into a coordinated, multi-agent powerhouse that solves complex, niche problems without the usual generic fluff. Grab your favorites and start building like a whole...

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 code3-dev/proxycloud

Proxy Cloud

You can now run a completely free, unlimited, and secure VPN on your device using a fully open-source application built on modern proxy protocols. This project, called proxycloud, is a beautiful Flutter-based client designed to bypass network restrictions using advanced V2Ray and MTProto technologies. Instead of relying on expensive, data-logging commercial providers, it puts you in control by letting you easily import, organize, and auto-select the fastest global servers based on real-time ping. It even features per-app tunneling and dedicated tools for Telegram. It is a brilliant way to reclaim your online privacy with a gorgeous, zero-data-collection application.

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 0xsero/parcels

Parcels

Parcels is the command line utility that lets you instantly migrate live coding agent sessions to a remote machine. If you are running an AI coding assistant on your laptop and need to close the lid, you do not have to lose your momentum. This project packages up your entire git repository, uncommitted changes, environment files, and the active AI session transcript, then ships it to a remote server over your secure network. The remote machine automatically restores the session inside tmux, allowing the AI to read your handoff notes and continue working right where it left off.

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 augani/dory

Dory

Dory is a free, open-source native macOS app that runs your Docker and Linux containers using almost five times less idle memory than traditional setups. Instead of spinning up a heavy, resource-hungry virtual machine for every single container, this clever tool boots one single, lightweight Linux micro-VM to handle them all. Built natively with Swift and SwiftUI, it integrates directly with your existing docker command line tools, boots up silently with virtually zero CPU strain, and even packs one-click Kubernetes support. It is the perfect, lightweight alternative to bloated container apps for anyone running Apple silicon.

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 0xjuanma/golazo

Golazo

Follow live football matches right from your command line without ever leaving your terminal or breaking your coding flow. This incredibly clever terminal tool delivers real-time score updates, minute-by-minute match timelines, and detailed stats like possession and starting lineups in a gorgeous, lightweight display. The absolute best part is that it embeds clickable goal replay links directly next to the scorer's name, so you can watch instant highlights of that world-class strike without opening a browser. It is the perfect, distraction-free way to keep up with over sixty leagues and even the World Cup while you work.

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 tronghieuit/valtec-tts

Valtec TTS

Clone any Vietnamese voice using just a brief three-second audio clip and run the entire setup directly on your everyday computer processor without needing an expensive graphics card. This project makes high-quality Vietnamese speech synthesis incredibly accessible by packing a full multi-speaker system and zero-shot voice cloning into a tiny model of under seventy-five million parameters. You can run it with just two lines of Python code, launch a local web interface, or download a pre-packaged app that works instantly on your desktop. It is the perfect lightweight tool for generating natural speech, matching local accents, and creating voiceovers on basic hardware.

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 claude-world/notebooklm-skill

notebooklm-skill

The automation pipeline that finally lets Claude use Google's NotebookLM for automated content creation. Instead of manually copying and pasting sources into Google's interface, you can feed web links or documents directly to this tool. It logs into Google securely, creates a temporary notebook, extracts cited research findings, and hands them to Claude to craft polished blogs, scripts, or social posts. It even automates downloading NotebookLM's famous generated podcasts and slides. You get the depth of Google's search engine combined with Claude's unmatched writing voice, all wrapped in a simple command-line interface or a tool for your editor.

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 archive228/loopkit

loopkit

Stop letting your coding agents write messy, half-finished code based on sheer guesswork. You can now equip tools like Cursor, Claude Code, and Gemini with a set of thirty-three highly specific, battle-tested skills that load only when needed. Instead of burning your token budget on massive system prompts, this minimal harness lets your agent pull in the exact instructions required for tasks like systematic debugging, security reviews, or test-driven development. It even runs a local validation loop to verify its own work before claiming a feature is done. Grab loopkit to make your AI coding assistant work like a disciplined developer.

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 0xroylee/getsuperpower

GetSuperpower

AI agents are incredibly powerful, but getting them to run a complex, multi-step engineering pipeline usually means baby-sitting them through every single step. GetSuperpower changes that by packaging an entire multi-agent workflow into a single callable skill. Instead of manually telling your agent to brainstorm, write a spec, run test-driven development, and archive the code, you just install a workflow and call one entry command. The tool automatically resolves and installs all required sub-skills, validates the manifest, and orchestrates the entire sequence behind the scenes. It makes complex, multi-step agent automation reliable and repeatable with a single command.

📰 https://news.ycombinator.com/item?id=48791847

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 chipmates/agoracosmica

Agora Cosmica

Imagine talking directly to Marcus Aurelius, Ada Lovelace, or Frida Kahlo, hearing their deepest insights in their own historically researched voices. This is exactly what agoracosmica does, bringing thirty legendary figures from science, philosophy, and art to life as interactive AI Echoes. Built by a German nonprofit, this open-source platform pairs hundreds of beautifully narrated historical episodes and multi-character debate councils with live AI conversation. What makes it truly remarkable is its absolute commitment to privacy: there is no signup, no tracking cookies, and your conversations are encrypted locally right in your browser. You can even run the entire AI voice and language stack completely on your own machine. It...

📰 https://news.ycombinator.com/item?id=48459913

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 zhongerxin/cowart

Cowart

Cowart is the native infinite-canvas plugin that finally lets you sketch, annotate, and generate AI images directly inside your Codex workspace. Instead of forcing you to jump back and forth between different browser tabs and image editors, it embeds a collaborative infinite canvas right into your workflow. The coolest part is how it handles edits. You can just draw red arrows and write notes directly onto an image, click a button, and the AI reads your visual feedback to generate a perfectly revised, clean version right next to the original. It is the ultimate visual playground for your workspace.

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 rushindrasinha/youtube-shorts-pipeline

youtube-shorts-pipeline

Turn a simple text topic into a fully finished, publish-ready vertical video for about eleven cents in API costs. This system searches the web for factual, up-to-date news, writes a high-retention script, generates matching vertical images, records a natural voiceover, burns in animated word-by-word captions, adds background music, and uploads the final video to YouTube. What makes this special is its niche intelligence, which shapes every step using pre-configured profiles so a true crime video automatically gets suspenseful music and dark visuals, while a cooking video gets upbeat audio and warm food imagery. It is the ultimate tool for running a complete content channel automatically.

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 yuzhworkhard-wq/image-to-code

image-to-code

You can now transform a flat mobile screenshot directly into pixel-perfect frontend code and fully editable Figma layers. The image-to-code repository is a specialized assistant workflow designed to solve the frustrating gap between static design images and actual development assets. Instead of just guessing where elements go, it maps out a precise coordinate system to slice high-resolution transparent icons, write adaptive styling, and even export actual vector paths and text layers back into Figma. This means you can easily reconstruct working websites and design files from a single reference image. Check it out to make your design-to-code workflow a breeze.

🆔 @hackernewsgithubprojects
Media is too big
VIEW IN TELEGRAM
📦 j-jaeyoung/bad-epoll

Bad Epoll

Escalate privilege all the way to root on Linux and Android devices using a tiny, hidden race condition that even advanced AI models completely missed. The project, bad-epoll, documents a brilliant zero-day vulnerability in a core Linux subsystem that cannot be turned off. It exploits a split-second collision when closing two objects at the exact same time, widening a microscopic six-instruction race window into a ninety-nine percent reliable exploit. It is a fantastic reminder that even when state-of-the-art automated security tools sweep a codebase, human ingenuity can still find the quietest, most elusive bugs hiding in plain sight.

📰 https://news.ycombinator.com/item?id=48792275

🆔 @hackernewsgithubprojects