📦 smg-project/smg

Shepherd Model Gateway: The LLM Traffic Controller

Shepherd Model Gateway connects any language model to any client through a single fast interface, so you can mix open-source tools with cloud services without rewriting code. Its smartest move is cache-aware routing, which watches how much memory each server is holding and sends new requests to the one that already has similar data ready, saving time and money. Built in Rust, it handles heavy traffic with built-in safety nets like circuit breakers and keeps your conversation history private. Instead of juggling dozens of different APIs, you get one clean door to run your AI apps efficiently.

🆔 @hackernewsgithubprojects

📦 heartached/noctis

Noctis Music Player Review

Play your entire collection in pristine audio quality without sacrificing your privacy or sanity. Noctis is a beautiful, cross-platform music player that finally gets it right. Forget clunky interfaces and forced updates. Instead, enjoy a stunning dark theme, smooth animations, and a library that actually stays organized. But the real magic happens with the lyrics. Watch as word-by-word karaoke lines appear in perfect sync, just like a professional video clip. It even lets you edit those lyrics yourself and share them as cool image cards. Whether you are streaming from a home server or just managing local files, Noctis feels fast and responsive.

📰 https://news.ycombinator.com/item?id=48099435

🆔 @hackernewsgithubprojects

📦 noobygains/claude-pulse

claude-pulse: The AI Usage Dashboard

Claude-pulse is the status bar dashboard that finally keeps your AI usage visible and under control. It sits right in your terminal, reading data directly from Claude Code without needing any extra API keys or complex setups. You get instant, color-coded progress bars for your session and weekly limits, so you always know how much of your budget you have left. It tracks your spending in real time, shows your active model, and even highlights when your context window is getting full. Instead of guessing or checking dashboards later, you see exactly what’s happening while you work.

🆔 @hackernewsgithubprojects

📦 qwenaudio/qwen-audio-agent

Qwen Audio Agent

Qwen Audio Agent is the voice frontend that keeps AI agents present while they work. It handles real-time voice chat so you can talk naturally, while the agent runs background tasks like coding or research. Instead of waiting for a loading spinner, the agent just listens, chimes in when it’s done, and hands control back. You can use it with major tools like Claude Code or OpenClaw, or just stick to voice mode. It runs locally, respects your privacy, and turns your computer into a voice-first workspace. Check it out if you want your agents to actually feel alive.

🆔 @hackernewsgithubprojects

📦 kyuz0/strix-halo-ds4-toolbox

AMD Strix Halo ds4 Toolbox

AMD Strix Halo ds4 Toolbox is the developer container that finally lets you run powerful local AI models on AMD’s new laptop chips. It solves the headache of complex hardware setup by providing a pre-built environment that gets your integrated graphics running deep learning software with just a few commands. The coolest part is how it handles massive AI models by treating your solid-state drive like extra memory, letting you run huge language models on laptops with only a single hundred二十八 gigabytes of RAM. You simply pull the container, grab the model files, and start chatting or coding right away.

🆔 @hackernewsgithubprojects

📦 paritok-official/paritok-4b-v1

Cut Coding API Costs by 95% with Paritok

Coding agents burn through your API budget fast. Paritok is the first open-source compression model built specifically to fix that. It sits between your coding assistant and the API, shrinking massive conversation histories down to twenty-five percent of their original size. This means you keep almost all the solving power but pay a tiny fraction of the token costs. The best part is it never deletes your code forever. If the AI needs the full original text back, it can instantly recall it from the compressed data. It is a smart, self-hostable way to make long coding sessions affordable and efficient.

🆔 @hackernewsgithubprojects

📦 bmad-code-org/bmad-module-creative-intelligence-suite

Creative Intelligence Suite

Get unstuck and spark fresh ideas by letting specialized AI agents guide you through proven creative exercises, turning that mental block into actionable solutions for brainstorming, design thinking, or problem solving. It is basically a personal creativity coach that walks you through structured techniques instead of just dumping a random answer. The coolest part is the Brainstorming Coach, which actively steers you through frameworks like SCAMPER to force you to look at your problem from seven completely different angles until a real breakthrough clicks. It works because it makes you do the thinking, not just read it.

🆔 @hackernewsgithubprojects

📦 tonyd2wild/deepseek-v4-flash-0731-dspark-1m-nvfp4-kv-2x-dgx-spark

DeepSeek V4 Flash on DGX Spark: 1 Million Token Context

This setup runs the DeepSeek V4 Flash model across two dedicated graphics cards to deliver an impressive one million token context window while maintaining smooth, high-speed responses. It solves the common problem of AI models losing track of long conversations or documents by using a special memory-saving technique that keeps huge amounts of text accessible without slowing everything down. The repository provides a ready-to-use recipe that automatically handles the complex configuration needed to keep the system stable under heavy loads. You get reliable performance for tasks like processing large codebases or analyzing lengthy reports without the usual glitches or delays.

🆔 @hackernewsgithubprojects

📦 lixiaochuan2020/agentic-context-management

Agentic Context Management

Agentic Context Management is the open-source toolkit that finally lets AI agents remember long conversations without forgetting the details. Traditional AI models hit a memory wall and lose track of early clues when tasks get complex, but this project teaches a model to actively manage its own knowledge by compressing old data into retrievable chunks and keeping a permanent backup on disk. Instead of blindly deleting chat history, the agent learns to summarize what it knows and store it safely for later reference. This post-training approach turns a forgetful model into a reliable researcher that can handle multi-step problems without losing context.

🆔 @hackernewsgithubprojects

📦 frontisai/openrsi

OpenRSI: Training AI to Improve AI

Build smarter AI by teaching it to code, test, and fix itself through a loop of trial and error. This project turns the abstract idea of self-improvement into real, measurable code where machines learn from their own mistakes. You get an open toolkit to train models using actual execution feedback, letting them refine their own software engineering skills over time. It is not just a benchmark result, but a complete system for making AI research faster and more efficient. You can run experiments on your own GPU to watch the model evolve its own coding abilities from scratch.

🆔 @hackernewsgithubprojects

📦 daly2211/autoretrieval

Autoretrieval: Self-Improving AI Search

Autoretrieval is the autonomous AI agent that optimizes document search pipelines without human intervention. Most retrieval systems are static and brittle, requiring manual tuning to work well with your specific data. This tool changes that by letting an AI agent experiment on its own. You point it at your documents, and it iteratively tweaks chunk sizes, embedding models, and keyword filters. The agent runs evaluations, checks if accuracy improved, and keeps only the successful changes. It repeats this loop overnight, waking up to a significantly better search setup. You get a tailored retrieval system that actually works for your content, built through trial and error rather than guesswork.

📰 https://news.ycombinator.com/item?id=48995907

🆔 @hackernewsgithubprojects

📦 ddcat-ai/open-ai-canvas

Open AI Canvas: Your Infinite AI Film Studio

Open AI Canvas turns your computer into an infinite film studio where artificial intelligence actually builds the scenes for you. Instead of wrestling with confusing timelines or manual editing, you get a massive blank workspace that grows as you create. You can drag in images, text, and prompts, then connect them with smart agents that automatically generate video clips and storyboards based on your ideas. It handles the heavy lifting of turning your loose creative thoughts into a structured visual narrative, letting you focus on the story rather than the technical setup.

🆔 @hackernewsgithubprojects

📦 jsgrrchg/neverwrite

NeverWrite: Your AI-Powered Markdown Workspace

NeverWrite lets you build a personal knowledge vault that actually works with artificial intelligence agents. It combines a flexible text editor with a powerful review system that tracks every change an AI makes so you can inspect, accept, or reject edits before they stick. This keeps your writing in your control while still giving you the help of smart assistants like Claude or Codex. You can clip web pages into your notes, view documents in a visual graph, and keep all your data stored locally on your own computer.

📰 https://news.ycombinator.com/item?id=49131545

🆔 @hackernewsgithubprojects

📦 mokshablr/gander

Gander: The Ultimate Offline Android File Viewer

Gander lets you open any file on your Android phone without giving away a single permission. It handles PDFs, Word docs, Excel sheets, PowerPoint slides, photos, and videos all in one tiny app. The coolest part is how it works: it asks for zero permissions because it only accesses files you explicitly share with it. No cloud tracking, no internet access needed, and your data stays completely private. It even searches inside documents and lets you browse folders you grant access to. If you want a simple, secure way to view files without the bloat, this is the tool you need.

📰 https://news.ycombinator.com/item?id=49119425

🆔 @hackernewsgithubprojects

📦 ferrumox/rabbit

How Rabbit Runs a 2.8 Trillion Parameter Model on a Laptop

How Rabbit Runs a 2.8 Trillion Parameter Model on a Laptop

rabbit is a pure Rust engine that lets you run massive frontier AI models, like the 2.8 trillion parameter Kimi K3, directly on a standard consumer computer with just 128 gigabytes of RAM. Instead of trying to fit the entire model into memory, it keeps the essential parts resident in RAM while streaming the specific neural network experts onto the processor only when they are needed. This clever approach of loading thousands of tiny sub-networks from disk on demand means you can run some of the largest open-weight models available without needing expensive professional hardware.

📰 https://news.ycombinator.com/item?id=49093910

🆔 @hackernewsgithubprojects

📦 yu-fangxu/w2s-opd

Weak-to-Strong On-Policy Distillation

Make a strong AI smarter by teaching it from weaker models. That sounds backwards, right? Usually, you train a student using a teacher that is clearly better. But W2S-OPD flips the script. It takes a weak positive model and a weak negative model, compares their differences, and uses that contrast to build a fake, stronger teacher right inside the student's own math space. This lets the student improve without needing a massive, expensive expert model to guide it. It is a clever workaround for hitting the ceiling when you run out of bigger teachers to learn from. It is smart, efficient, and totally changes how we think about training.

🆔 @hackernewsgithubprojects

📦 h-embodvis/road

ROAD: 3D Shape Generation That Actually Makes Sense

Road is the 3D generation project that finally teaches computers to understand shape by aligning what they see with what they build. Most AI tools just guess geometry, but Road fixes that by locking a detailed 3D scanner onto the generation process. It matches every single point on the new model against a known perfect reference until they perfectly align. This means instead of blurry blobs, you get crisp, accurate textures that actually match the prompt. It is a clever way to force quality through strict point matching. Go check out the code if you want to build better 3D models.

🆔 @hackernewsgithubprojects

📦 anxmuy/agenticasr

AgenticASR: Fixing Speech Recognition in Real Time

AgenticASR actually rewrites speech transcripts as they happen to fix mistakes. Most speech tools just type what they hear, even when the speaker changes their mind or stumbles. This project adds a smart layer that listens closely and corrects the text on the fly. It removes filler words, fixes self-corrections, and updates earlier parts of the sentence when new information arrives. It works with any speech-to-text engine and handles both English and Chinese. You can use it for one-time cleanup or live streaming. The code includes everything needed to train, run, and test this refining process. It turns messy spoken language into clean, accurate written text automatically.

🆔 @hackernewsgithubprojects

📦 597358816/distilled-rl

Distilled RL: Smarter AI Learning

Distilled Reinforcement Learning is the AI training framework that teaches smaller models by using a smarter teacher to fine-tune every single word choice. Standard reinforcement learning usually just looks at the final answer to decide if the model succeeded, which is a blunt tool. Distilled RL changes the game by having a larger, expert model whisper corrections at every step, guiding the student’s grammar and logic in real time rather than just grading the result. This selective coaching helps tiny models understand complex math and reasoning far better than standard methods. You get impressive accuracy boosts without needing massive computing power.

🆔 @hackernewsgithubprojects

📦 epfml/looped-moe

Looped MoE: Share Expert Work, Cut Compute Costs

looped moe lets you slash the training cost of large language models by reusing the same brain cells across multiple layers. instead of teaching every single layer from scratch, it ties them together so they share weights, cutting parameter counts by up to four times without losing smarts. it works with popular models like olmoe and qwen three, proving that recycling knowledge is a clever hack for efficient ai. check it out if you want to train faster and cheaper.

🆔 @hackernewsgithubprojects

📦 borkdude/cljbang.el

Write Clojure in Emacs

Run Clojure code directly inside Emacs without ever leaving the editor or spinning up external processes. Cljbang lets you write familiar Clojure syntax that instantly compiles into native Emacs Lisp and executes right where you are. You get the elegant data structures and macros you love while keeping every function fully accessible to Emacs itself. It feels like magic because the performance is nearly identical to writing pure Lisp. This tool bridges two worlds so you can use modern syntax for everyday tasks while staying firmly inside your favorite environment.

📰 https://news.ycombinator.com/item?id=49004149

🆔 @hackernewsgithubprojects