📦 miaai-lab/qwen3.8-flash-next-single-dgx-spark

One Card, Giant Model

Run a massive, multimodal AI model on a single workstation GPU. This project lets you serve a giant vision-language model that handles text, images, and video using only one piece of hardware. The real magic is the memory trick. Instead of forcing the entire model into the graphics card, it parks the huge lookup tables in system memory and fetches them on demand. This simple trick lets a 99-gigabyte model fit inside a device with just 121 gigabytes of unified memory. It turns a modest desktop chip into a capable server for serious AI tasks without buying a rack of data center gear.

🆔 @hackernewsgithubprojects

📦 cloudgeni-ai/opengeni

OpenGeni: Run AI Agents on Your Own Hardware

OpenGeni is the self-hostable runtime that finally lets you run AI agents on your own hardware without losing control. It solves the trust gap in enterprise AI by keeping your data, credentials, and audit logs entirely inside your own infrastructure. The standout feature is the connected machine model, where you enroll your laptop or server as a first-class compute target. Instead of sending code to a vendor's cloud, the agent dials out to your machine to work directly with your files and git repos. It’s a durable, replayable session API that treats your hardware as an equal to managed sandboxes.

🆔 @hackernewsgithubprojects

📦 hwdsl2/docker-whisper

Docker Whisper: Your Private AI Transcriber

Docker Whisper is the self-hosted server that finally lets you process audio without sending it to the cloud. It runs OpenAI’s powerful speech recognition model directly on your own hardware, turning any voice recording into accurate text. The coolest part is speaker diarization, which automatically identifies exactly who is talking in a group conversation. This means your meeting notes know which person said what, all while keeping your data completely private. It even speaks the same API language as OpenAI, so you can switch your existing apps over with a single line change.

📰 https://news.ycombinator.com/item?id=47732384

🆔 @hackernewsgithubprojects

📦 yejy53/editable-design

Editable-Design: AI Design That Never Flattens

Editable-Design is the visual design tool that refuses to flatten your work. Instead of giving you a single, static image where changing one word breaks the whole picture, it builds your design using real, separate text layers and independent images. This means you can select any element and tweak it exactly as you need. It uses AI to handle the tricky parts of layout and style, but the result is fully editable code. So you get professional-level visual polish without being locked into a single, unchangeable screenshot. It is a smarter way to create graphics that actually work for your specific needs.

🆔 @hackernewsgithubprojects

📦 gnosil/semantix

Semantix: The Memory Engine for AI Agents

Semantix is the memory kernel that finally stops your AI coding agent from forgetting everything the moment a session ends. It solves a massive, invisible problem: most agents treat every new chat as a blank slate, constantly re-learning your project's specific quirks and build rules from scratch, which burns through your token budget and slows down workflows. The clever part is how it fixes this without you changing your existing tools. It quietly extracts reusable insights from your past conversations and stores them in a local library.

🆔 @hackernewsgithubprojects

📦 owainlewis/machinist

Machinist: The AI Code Factory

Machinist turns your local computer into a controlled factory for running AI coding agents. You can queue up complex software tasks and let tools like Codex or Claude Code handle the heavy lifting, all without ever letting a single API key leave your machine. The system records every single step, from the initial prompt to the final code output, so you can watch exactly what the AI did and why. It hands you a clean pull request at the end, keeping the final decision to merge entirely in your hands. This gives developers a safe, transparent way to automate their workflows while keeping full control over their credentials and data.

🆔 @hackernewsgithubprojects

📦 youichi-uda/godot-mcp-pro

Godot MCP Pro: AI Directs Your Game Editor

Your AI assistant can now play your game, press buttons, and fix its own code while you watch the editor. Godot MCP Pro connects Claude directly to the Godot editor, giving it live control over scenes, physics, and animations. This isn't just for writing scripts; it lets the AI inspect a running game, simulate player inputs, and run automated tests to find bugs before you do. You can literally ask it to fix a collision issue and verify the fix by watching the game play. It turns your AI from a coder into a QA partner. If you want your AI to actually test your game, this is the tool.

🆔 @hackernewsgithubprojects

📦 dicklesworthstone/ultimate_bug_scanner

Ultimate Bug Scanner: The AI Agent's Safety Net

Spot a thousand hidden bugs before your code even runs. The Ultimate Bug Scanner acts as a safety net for AI coding agents, catching over a thousand specific bug patterns across ten different languages in under a second. It scans your code for real problems like memory leaks and security holes, then gives the AI a clear pass or fail signal. This means your coding assistant actually learns from mistakes and stops making them. It is the missing piece of quality control for the AI era. Your code will finally be as safe as it is fast.

🆔 @hackernewsgithubprojects

📦 agent-next/polymarket-paper-trader

Polymarket Paper Trader: AI Betting Simulator

Turn your AI agent into a risk-free trader for Polymarket. This tool gives your bot ten thousand dollars in fake cash to bet on real-world events using live order books. It isn't a toy that uses random numbers; it simulates actual exchange mechanics, including fees and slippage. Your agent places real bids against real liquidity, so the profit and loss is accurate. It even includes a backtesting engine to test strategies against historical data. The best part is the leaderboard, where you can pit different AI strategies against each other to see who wins. No real money is lost, but the trading logic is genuine.

🆔 @hackernewsgithubprojects

📦 oliver-zehentleitner/keep-the-why

Keep the Why: The AI Skill That Remembers Why Code Exists

Your AI coding assistant keeps forgetting why you rejected a specific solution, so it suggests the exact same broken idea again and again. Keep the Why fixes this by quietly saving the reasoning behind your decisions right inside your project files. It turns those scattered thoughts into simple text notes that travel with your code, meaning your assistant finally understands your past choices. You stop repeating old mistakes, new team members onboard faster without asking old colleagues, and legacy projects stop feeling like black boxes. It is not a separate documentation site to maintain, but a lightweight layer that grows naturally as you work.

🆔 @hackernewsgithubprojects

📦 hex/claude-council

Claude Council

Stop guessing when your code breaks. Claude Council asks five different AI models the exact same question and puts their answers right next to each other. You see where they agree and where they clash, instantly exposing the blind spots that usually trip up single-model advice. It turns a solo coding session into a group review, helping you spot flaws before they become bugs. No more wondering if your logic is sound. Just ask the council and trust the consensus. Your next architecture decision has never been safer.

🆔 @hackernewsgithubprojects

📦 ashemag/human-atlas

Human Atlas: Explode the Human Body in 3D

A web app called Human Atlas lets you take a full adult human body apart into more than two thousand separate, clickable pieces right in your browser. Instead of staring at a static diagram, you slide a control to pull the organs and bones away from each other, turning the anatomy into a floating inventory so you can see exactly where every part sits. It is built as an educational tool, meaning it is not a medical device, but a way for students and curious people to visualize the complex layout of fifteen different body systems.

🆔 @hackernewsgithubprojects

📦 petergpt/3d-prompt-collection

3D Prompt Collection: 63 World-Building Ideas

Three D Prompt Collection is the prompt library that turns text into massive, explorable three-dimensional environments. It solves the problem of running out of ideas for complex virtual scenes by offering sixty-three detailed descriptions ready to copy. Instead of struggling to explain how to build a city, you simply paste a prompt like the one for a voxel Ancient Rome or a living ocean reef. The repository is organized into themes like natural spectacles and cosmic finales, making it easy to find the perfect visual concept for your next project.

🆔 @hackernewsgithubprojects

📦 dannymac180/astra-advisor

Astra Advisor: The AI That Hires Its Own Team

Astra Advisor is the coding plugin that finally gives your main AI assistant the power to hire its own specialized team. You hand it a complex task, and it stops doing everything alone. Instead, it dynamically decides when to spin up temporary, specialized subagents to handle specific, isolated pieces of the work while it continues planning and verifying the core architecture. It picks the exact right brain for each job and keeps a strict, transparent ledger of every single action.

🆔 @hackernewsgithubprojects

📦 theagencymge/aero-dock

Aero Dock: The Windows Glass Taskbar

Hovering over an app icon makes it grow, and what pops up is not a fake image but a live, moving picture of the actual window behind it. This is Aero Dock, a free, open-source replacement for the standard Windows taskbar that brings back the glossy, translucent look from years past. It sits on any edge of your screen, letting you drag apps directly from your file explorer onto the glass pane. Because it tracks your real windows, you can see exactly what you are doing in every open program without ever switching focus.

🆔 @hackernewsgithubprojects

📦 shadowarcanist/netviz

Netviz: Animate Your Network Architecture

Netviz is the browser tool that turns static network diagrams into living, animated stories. Instead of just drawing boxes and arrows for servers and databases, you can build timed animations showing exactly how a request flows through your system. It solves the problem of explaining complex infrastructure by letting you visualize data movement, failovers, and load balancing without writing code. The standout feature is its deterministic playback engine, which ensures every animated effect follows the exact path you drew, so your presentation looks perfect every single time. If you need to explain how your backend works, Netviz makes it visual, clear, and easy to understand.

🆔 @hackernewsgithubprojects

📦 nikolai-vysotskyi/trace-mcp

Trace MCP Cuts Token Usage by 90%

Trace MCP is the code intelligence tool that finally stops wasting tokens on redundant context. Instead of forcing your AI assistant to read every single line of a massive pull request, it maps the actual structure of your code so the model only sees the parts that matter. This means you can review complex changes in eight percent of the usual space, all while running entirely on your local machine so your data never leaves your computer. It is like giving your AI a precise map instead of a blindfold.

📰 https://news.ycombinator.com/item?id=49581857

🆔 @hackernewsgithubprojects

📦 jpezzulli/sglang-rtxpro6000

sglang-rtxpro6000

sglang-rtxpro6000 is a personal, optimized runtime that lets you run large Qwen language models on a single RTX Pro 6000 Blackwell graphics card. This is not a toy project; it is a specialized build of the SGLang inference engine, tuned specifically for the SM 120 architecture found in that specific professional-grade GPU. The reason this matters is that running these heavy models usually demands expensive multi-GPU clusters or massive cloud bills, but this configuration squeezes serious performance out of one piece of hardware. It solves the problem of high inference costs by leveraging the specific strengths of Blackwell silicon.

🆔 @hackernewsgithubprojects

📦 sapientinc/hrm-text

HRM-Text: Train an AI for $1000

Sapient Inc just proved you can build a billion parameter language model for under one thousand dollars, which completely breaks the usual rules for AI development. This project uses a specialized architecture that repeats its internal logic to squeeze more intelligence out of fewer chips, cutting the required compute power by hundreds of times compared to standard methods. Instead of buying expensive hardware, you get a full training framework that runs on just a few high-end graphics cards to produce a model that handles math and reading comprehension surprisingly well. It is not just a theoretical idea but a complete toolkit with data processing and testing built in.

🆔 @hackernewsgithubprojects

📦 shengsheng90/dsh-taskboard

How AI Agents Manage Real Tasks

DSH Taskboard is the local plugin that gives AI agents a real job board. While most AI tools just chat, this project lets coding agents actually claim tasks, do the work, and submit results for your approval. It runs locally on your machine using a simple database, so your data stays private. The coolest part is that agents can't mark their own work as finished. They have to send it to you, a human, for final sign-off. It turns your AI assistant into a productive coworker that you can actually trust with real projects.

🆔 @hackernewsgithubprojects