56 subscribers
7.26K videos
7.87K links
Download Telegram
This media is not supported in your browser
VIEW IN TELEGRAM
📦 ruwwww/comfyui-speed

comfyui-speed

comfyui-speed is the ComfyUI custom node that speeds up image generation by progressively scaling up resolution during the denoising process. Instead of wasting heavy processing power on high-resolution noise at the start of a generation, this tool dynamically expands the image size in steps, matching the natural details as they emerge. By skipping unnecessary detail calculations in the early noisy phases, it cuts generation times by over thirty percent without losing any final visual quality. It plugs right into your existing workflows, giving you a clever mathematical shortcut to faster, beautiful AI art.

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 fergana-labs/stash

Stash

Stash is the shared memory hub that finally gives your team's coding agents a unified brain. When you run AI programmers like Claude Code or Cursor, they generate incredibly valuable session transcripts, but those insights usually stay trapped on a single computer, causing duplicated work and lost context. This project solves that by streaming every transcript and tool call into an agent-native virtual filesystem. Now, your agents can run semantic searches over past attempts, edit shared markdown pages in real time, and pick up exactly where another agent left off, speeding up engineering tasks by almost fifty percent.

📰 https://news.ycombinator.com/item?id=47882853

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

claude-code-workflows

claude-code-workflows is the repository of pre-built configurations that finally turns your terminal assistant into an elite, automated review team. Instead of manually checking code line-by-line, you get production-ready workflows for code quality, security vulnerabilities, and front-end design checks. It uses specialized subagents and slash commands to run deep, dual-loop reviews right in your terminal or automated on your pull requests. The coolest part is the design review workflow, which hooks up browser automation to interact with your live UI components and check them against world-class accessibility and design standards. It is like having a principal engineer, security lead, and UX designer reviewing your work on demand.

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 zexuanw958-svg/travel-plan-viz

travel-plan-viz

Your AI travel assistant can now package an entire vacation into a single, beautifully designed HTML file that works perfectly offline on your phone. This clever tool researches your destination, maps out a complete day-by-day itinerary with Leaflet, calculates a pre-trip ticket booking schedule, and embeds the entire plan as raw data right inside the page. Because everything is packed into a single responsive file, your itinerary reads perfectly even when you lose signal underground. If you want to make changes, you just hand the file back to your AI assistant to rearrange the days instantly.

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 xberg-io/alef

alef

alef is the polyglot binding generator that finally ends the misery of writing and maintaining multi-language glue code for your Rust projects. Instead of hand-writing separate libraries for Python, Node, Go, and Ruby, you write your core logic in Rust once. You define a single configuration file, and alef automatically spits out fully-typed, lint-clean bindings for up to eighteen different target languages. It even scaffolds the packages, builds type stubs, and generates matching test suites so your APIs never drift apart. It is the ultimate way to ship your high-performance Rust core to every popular developer ecosystem without losing your sanity.

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 programmersd21/bangen

Bangen

Spice up your terminal by turning plain text into vibrant, animated ASCII art banners with custom gradients. That is what bangen does. It is a command-line tool and terminal app that takes boring text and morphs it into eye-catching animated banners with effects like wave, pulse, and glitch. The coolest part is how it handles styling; you get a complete interactive interface right in your terminal to preview designs, adjust true-color RGB gradients, and even export your animated creations directly to transparent GIFs or PNGs to share with friends. It makes customizing terminal visuals incredibly simple and fun.

📰 https://news.ycombinator.com/item?id=47750564

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 pengrui-han/llm_modularity

llm_modularity

llm_modularity is the research pipeline that reveals how artificial neural networks organize themselves like biological brains. By tracking how neurons fire across dozens of reasoning tasks, this toolkit maps out dedicated neural neighborhoods inside popular language models. If you disable the specific physics neurons, a model can still write perfect grammar but its scientific logic completely falls apart. Conversely, turning off the language neurons leaves its logic intact while its grammar breaks down. This work offers exciting evidence that intelligence itself naturally organizes into specialized departments, whether it is shaped by biological evolution or code.

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

OpenHive

Host your own private team chat server in just a few minutes without being locked into expensive subscription platforms. This clever alternative automatically configures over twenty database tables, handles real-time messaging, and even manages file sharing using your own cloud storage. But the coolest part is that it includes built-in video and audio calls with screen sharing, meaning you do not have to rely on third-party services just to talk to your team. It is the perfect weekend project for anyone wanting full control over their team communication. Grab your own database keys and set up your private chat hub today.

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 basilysf1709/golos

Golos

Turn your voice into text instantly anywhere you can type. With this tool, you just hold down a hotkey, speak what is on your mind, and let go. The moment you release the key, your words are immediately pasted right where your cursor is sitting. This clever macOS tool bypasses heavy, bloated apps by running as a tiny, lightweight background utility. It streams your audio in real time to get fast, accurate transcriptions, and you can even set up custom word shortcuts. It is the perfect, distraction-free way to dictate your thoughts. Check out golos and give your keyboard a break.

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 gen-verse/latentmas

LatentMAS

The multi-agent framework that finally lets AI models collaborate without talking in slow, expensive text, latentmas is a brilliant shortcut for complex reasoning. Instead of forcing multiple AI agents to write out long, wordy paragraphs back and forth to solve a problem, this clever tool lets them pass their raw, internal thoughts directly to each other in the model's own hidden language. By skipping the slow process of turning thoughts into words, it slashes token costs by up to eighty percent and speeds up problem-solving by up to seven times. Grab the code and build lightning-fast, quiet AI teams today.

📰 https://news.ycombinator.com/item?id=48793308

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 zerx-lab/fluxdown

FluxDown

Grab files from almost any protocol at maximum speed using a modern desktop app that brings intelligence to your downloads. While classic downloaders split files into fixed parts regardless of your network, this utility runs a smart bandwidth probe before each download to dynamically calculate the perfect segment count based on your exact file size, CPU cores, and connection speed. Idle threads will even jump in to rescue slower segments on the fly. It combines a high-performance Rust engine with a clean, tracker-free interface that supports everything from standard web links to torrents and stream files, giving you a completely local-first power tool.

📰 https://news.ycombinator.com/item?id=48784543

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 soniqo/speech-swift

Speech Swift

You can now run a full suite of highly advanced speech models entirely on your Mac or iPhone without any internet connection. An open-source toolkit called speech-swift brings local text-to-speech, translation, and ultra-fast transcription to Apple Silicon using Apple's Neural Engine. It supports a mind-boggling array of over sixteen hundred languages by leveraging native machine learning libraries. Because everything runs locally on your device, you do not need API keys, and your audio data never leaves your hardware. It is a fantastic tool for developers wanting to build private, low-latency voice assistants.

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 yin-yizhen/sonic-topography

Sonic Topography

This desktop application turns the dynamic frequencies of your favorite songs into a glowing, responsive 3D terrain that ripples and dances to the beat. Named sonic-topography, this interactive music visualizer uses advanced frequency faders to transform sub-bass, mid-tones, and high-end air into actual physical waves, peaks, and sparkling edge shimmers. By acting as a visual effects mixer for the ground, it lets you customize exactly how different sounds shape the landscape. It is perfect for turning your music into a gorgeous, focus-friendly background clock or Pomodoro timer that breathes with the rhythm.

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 charonviz/text2geo

text2geo

You can now turn highly misspelled, multilingual place names into precise GPS coordinates completely offline without paying for a single API key. A clever little Python tool called text2geo bypasses rate limits and internet dependencies entirely by matching queries against local datasets of up to three hundred and fifty thousand locations. It uses fast fuzzy matching, so typing a heavily misspelled city name in English, Russian, or Ukrainian still instantly returns the correct latitude and longitude. It even resolves duplicates by prioritizing the larger city, so major hubs take precedence over tiny towns with the same name. Just install it, let it download its offline database once, and you can...

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 42lux/comfyui-42lux-hildegard-refiner

comfyui-42lux-hildegard-refiner

comfyui-42lux-hildegard-refiner is the ComfyUI extension that finally makes tiled image upscaling look completely seamless and natural. Instead of just chopping your image into simple grids, this clever tool generates a smart three-part layout for every single tile, combining a local crop, a positioning map, and a global thumbnail. This gives the underlying AI model full awareness of where each tile belongs in the bigger picture, preventing weird seams and mismatched details. You can even fine-tune the high-frequency details to let the model creatively restore noisy textures while keeping the core structure perfectly intact.

🆔 @hackernewsgithubprojects
Media is too big
VIEW IN TELEGRAM
📦 fewtarius/cachyllama

CachyLLama

CachyLLama reduces prompt evaluation times on lower-spec hardware from minutes to just one second by introducing a persistent on-disk cache. While running large local AI models on integrated graphics or handheld consoles usually means waiting through agonizing minutes of re-evaluating long system instructions, this project saves the conversation state directly to your solid-state drive. When you send a new message, it restores those thousands of background tokens instantly instead of recalculating them from scratch. It is the perfect upgrade for anyone wanting to run smart, multi-step local AI assistants on everyday devices without the painful lag.

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

ReceiptHero

Turn your messy pile of crumpled paper receipts into a clean, organized digital dashboard just by dragging and dropping them into your browser. This clever application uses an artificial intelligence model to scan your receipt images, instantly extracting the vendor, date, total cost, and payment method. It even automatically categorizes your spending into logical groups like groceries or dining out, converting foreign currencies on the fly. The best part is that it processes everything and stores the results directly in your local browser storage, keeping your financial data completely private without needing a database. It is the perfect weekend hack for stress-free expense tracking.

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 graphify-labs/graphify

Graphify

Graphify is the coding assistant tool that finally maps out your entire workspace into a queryable knowledge graph. Instead of watching your AI helper struggle to understand how your code, databases, and servers fit together, you can point this tool at your folder and watch it automatically build a complete map of your project. It connects your application logic, SQL schemas, and server setups into one visual network that your AI assistant can actually read and understand. It does all of this locally with zero AI cost, giving your assistant instant memory. Try it out and see your project's true shape.

📰 https://news.ycombinator.com/item?id=48794162

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 meituan-longcat/longcat-2.0

LongCat 2.0

Run massive artificial intelligence models on alternative hardware setups without losing speed or performance. The longcat-2.0 repository introduces a giant, one point six trillion parameter model that handles complex coding and agent tasks over incredibly long documents. To make this work smoothly, it uses a clever sparse attention mechanism that organizes memory access and shares indexing across layers, meaning it can process massive amounts of information without chugging. It is an incredibly smart way to run heavy-duty model reasoning on standard chips. If you want to check out how they optimized this massive architecture, you should definitely take a look.

🆔 @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
📦 suvink/cut-it-out

CutItOut

Your photos never actually leave your device with this entirely client-side background remover. Running completely inside your web browser using WebAssembly and a deep learning model, cut-it-out slices away image backgrounds instantly with zero external server calls. It features a smart magic brush that lets you restore or erase details based on color similarity, and a ghost overlay so you can easily see what you are painting back in. It is a completely free, privacy-first tool that delivers high-resolution transparent files without watermarks. Drag an image in, let the browser do the heavy lifting, and export clean graphics instantly.

🆔 @hackernewsgithubprojects