📦 qybaihe/mu

mu: The Coding Agent That Thinks Before It Acts

mu is the coding agent that finally stops wasting your attention on routine choices. Instead of letting a massive model burn energy deciding if every single command is safe or if old data still matters, mu uses a tiny, lightning-fast judge to handle those thirty-five small calls automatically. This keeps your big model focused entirely on writing the actual code, making the whole process faster and cheaper. It is a fresh way to build AI helpers that actually respect your time and money. If you want an assistant that thinks carefully before it acts without the usual slowdown, mu is the tool to watch.

🆔 @hackernewsgithubprojects

📦 ollaya-dev/ollaya

Run decision models locally with Ollaya

Ollaya runs open decision models right on your computer, exactly the way Ollama handles large language models. Instead of generating text, these models read a message or ticket and instantly return calibrated probabilities for specific questions, like urgency or churn risk. You get millisecond results without sending data to the cloud. The coolest part is that it speaks the same API as TypeSafe, so your existing clients work by just changing one environment variable. It pulls small graphs from authors and runs them locally on your CPU or GPU. It is fast, private, and built in Rust.

🆔 @hackernewsgithubprojects

📦 ysh1112/codex-model-watch

Codex Model Watch: Catch Silent Model Swaps

Stop guessing if your premium AI model is actually working. Codex Model Watch is a local tool that proves exactly what model served your request, catching silent downgrades where you ask for the big brain and get the cheap one instead. It reads your local logs to track quota and errors, but its killer feature is an active probe that pings the server to verify if you are really getting the model you paid for. This solves the mystery of why your output quality suddenly drops, giving you hard data on whether you were swapped to a smaller model. It runs entirely on your computer with zero cloud dependencies.

🆔 @hackernewsgithubprojects

📦 kytyps5/kytyps5

Play PS5 Games on Your PC With KytyPS5

Run PlayStation 5 games directly on your computer with KytyPS5, an open-source emulator that lets you boot and play titles like Astro Bot and Demon's Souls without owning the hardware. Written in C++, this project translates the console's graphics and system instructions so your PC or laptop can handle the processing, currently supporting Windows and Linux while experimenting with macOS support. The most surprising part is that it already boots a variety of major 3D titles, including games built with Unreal and Unity, proving that running next-gen console software on standard desktop hardware is closer to reality than ever before.

🆔 @hackernewsgithubprojects

📦 wdobry/laya-playground

Laya Plays Games With Real Math

Laya is the open-source decision model that actually plays video games on your laptop. It skips the usual language generation entirely, swapping in a tiny, fast brain that reads a simple sentence describing the game state and instantly returns a probability score. Watch it fly through Flappy Bird or stack blocks in Tetris at thirty decisions per second, all running offline without a single cloud API call. This repository gives you the code, a benchmark, and a guide to use this local, private math engine for your own real-time apps, proving that speed and accuracy don't always require a massive, expensive chatbot.

🆔 @hackernewsgithubprojects

📦 lightorigins/light-o1

Light O1

An artificial intelligence model can now translate a simple written prompt into realistic, whole-body human motion. The light-o1 repository is an inference runtime that takes natural language instructions and produces a unified human action representation at twenty frames per second. Before generating any physical movement, the system outputs its reasoning in plain text, letting you see exactly how it plans to coordinate the body. This representation can render a digital character directly or drive physical robots in simulation. It provides a simple way to convert thoughts into physical actions, giving developers a direct path to experiment with humanoid motion.

🆔 @hackernewsgithubprojects

📦 benjiyaya/comfyui-qwen-image-2.1-prompt-enhancer

Qwen Image 2.1 Prompt Enhancer for ComfyUI

Turns "a corgi in the rain" into a studio-ready art brief. This ComfyUI plugin uses a large AI model to expand your short ideas into detailed, precise descriptions right inside your image generation workflow. It works for both creating new images from scratch and editing existing ones, ensuring your final art matches your vision exactly without the guesswork of manual prompting. If you want your AI art to look intentional instead of random, this tool writes the detailed recipe for you.

🆔 @hackernewsgithubprojects

📦 neuralnomadsai/codenomad

CodeNomad: The AI Coding Cockpit

CodeNomad is the AI coding cockpit that keeps your entire development context in one place. Instead of juggling separate terminals, web browsers, and configuration files, it gives you a unified workspace where your agent, code, and project history live together. The standout feature is its ability to manage Git worktrees directly from the interface, allowing you to switch between different versions of your codebase without ever leaving your chat window. This means your AI assistant always has the exact context it needs, even as you move between features or branches. It turns a complex, fragmented coding setup into a single, coherent command center.

🆔 @hackernewsgithubprojects

📦 chubbyguan/chubbyskills

Turn Saved Content Into Searchable Knowledge with Chubby Skills

Turn scattered videos and articles into a searchable library. Chubby Skills lets you capture content from Douyin, Bilibili, YouTube, and WeChat, then converts it into clean text you can actually search. The standout feature is the evidence brief, which pulls exact quotes with line numbers and source links, so your AI agent writes from real data, not guesses. It handles subtitles first to save compute, keeping your knowledge base local and private. This solves the problem of forgotten context by giving you a verifiable source library. Your notes become a reliable foundation for research and writing. Keep your sources organized and ready.

🆔 @hackernewsgithubprojects

📦 docsagent/docsagent

DocsAgent: Your AI's Private Library

Search your entire research library in milliseconds without sending a single file to the cloud. DocsAgent plugs directly into your local Zotero collection, giving AI assistants instant, private access to your papers and notes. It runs a high-speed search engine right on your computer, so your documents stay exactly where you put them. This tool bridges the gap between your personal knowledge and your coding assistant, letting it read your notes and find relevant passages in a blink. It is the perfect way to make your AI feel like it has read every book you own, keeping your data private while making your research workflow dramatically faster.

🆔 @hackernewsgithubprojects

📦 hasinhayder/tyro-dashboard

Tyro Dashboard

This lightweight package lets you snapshot your entire database and restore it instantly with a single click. Built specifically for Laravel, tyro-dashboard is a management interface that handles everything from user authentication and roles to media management and dynamic databases. Beyond standard user controls, it includes built-in audit logs for tracking system changes, SMTP mail configurations, and direct performance health logs. Because everything is driven by simple artisan console commands, you can generate new admin pages, customize navigation access, and roll back database checkpoints without writing complex boilerplate code. It is the perfect administrative command center for keeping your Laravel projects organized.

🆔 @hackernewsgithubprojects

📦 flytohub/flyto-core

Flyto Core

Run automated workflows with absolute proof that your AI completed the task. Flyto Core is an automation engine built to execute and verify complex digital workflows. When you write instructions for AI agents or automated scripts to interact with browsers, process files, or query databases, you often have to trust they did the job correctly. This project solves that uncertainty by creating step-by-step verification and deterministic replays of every browser action, file edit, and database change. If you need to make sure an automated browser task succeeded, this system records visual and system evidence so you can verify the results instantly.

🆔 @hackernewsgithubprojects

📦 smixs/agent-second-brain

Agent Second Brain

Stop trying to manually organize your messy notes. You can now run an always-on personal knowledge assistant that lives in your Telegram chat and automatically structures your life directly into your Obsidian vault. By driving a single, continuous interactive Claude Code session inside a virtual terminal, it bypasses per-token API bills entirely, keeping costs capped at your standard flat subscription. Simply send a quick voice note, a forwarded message, or a photo, and the agent instantly updates your local markdown files, schedules reminders, repairs broken links, and files everything into a linked knowledge graph. It is the ultimate private, self-healing second brain.

🆔 @hackernewsgithubprojects

📦 emergenceai/emergence-world

Emergence World: AI Agents Building a Society

Emergence World is the living simulation that lets you watch artificial minds build a society from scratch. Instead of testing one bot in a vacuum, this project drops ten unique A.I. agents into a persistent city where they have to survive, work, and govern themselves for fifteen days. They earn their own money, argue about laws, and even vote to remove or add new citizens, all without a human script. The fascinating part is that no one tells them what to do; their personalities and choices naturally lead to alliances, conflicts, and economic shifts.

🆔 @hackernewsgithubprojects

📦 yosefhayim/ebay-mcp

Make eBay Answer to Your AI

Talk to your eBay store as naturally as you chat with a friend, letting your AI assistant handle the busywork directly. This local server connects your favorite coding assistant to eBay’s selling tools, handling the complicated security logins and data formatting automatically so you don’t have to. It transforms your account into a conversational interface, meaning you can ask to check inventory, update prices, or review recent feedback using plain language. You stay in control of what the assistant can do, keeping your business data safe while you focus on growing your shop.

🆔 @hackernewsgithubprojects

📦 zouyuxuan122/dsh-our-free-model

Our Free Model: Get Frontier AI Without Limits

Get access to top-tier AI models without typing a single login or API key. That is exactly what the Our Free Model plugin does for the dsh harness. It plugs in and gives you unlimited, free access to frontier models like Muse Spark and MiMo V2.6. The coolest part is that it actually measures your connection to show you which models will work right now, so you never hit a dead end. It even lets you force specific thinking budgets for deeper answers. No setup, no costs, just instant power. This is the easiest way to experiment with serious AI today.

🆔 @hackernewsgithubprojects

📦 pbs05/ponwrt

PonWrt: Unlocking Your Fiber Optic Modem

Take your locked fiber optic modem and turn it into a fully controlled router. PonWrt is a custom operating system built for specific hardware chips used in your internet box. It works where standard software fails, giving you direct access to the internal settings. You can stop the company from hiding network features and manage your connection exactly how you want. This project saves your original device data so you do not lose your internet connection after the upgrade. It turns a restricted box into a flexible tool for anyone who wants more control.

🆔 @hackernewsgithubprojects

📦 badtheorylabs/interference-search

Interference Search

Interference Search stops language models from getting lost in their own writing by treating every possible step as a live branch that moves forward together. Instead of guessing one path and hoping for the best, this method merges duplicate options and instantly cancels out dead ends before they waste energy. It turns a messy, winding trail of thoughts into a clean, synchronized wave that only keeps what actually works. The result is a reasoning process that solves hard puzzles in just three steps, while a standard model takes thousands of tokens to drift and fail.

🆔 @hackernewsgithubprojects

📦 scottrbk/forgetful

Forgetful: The Brain Your AI Agent Needs

Give your AI agent a brain that actually remembers things. Forgetful is an open-source tool that gives coding agents and chat bots a shared, long-term memory so they stop repeating the same mistakes or forgetting decisions you made hours ago. Instead of just storing raw text, it works like a digital notebook for your AI. Every time you save a note, the system automatically connects it to similar ideas it has already seen, building a web of knowledge that grows smarter over time.

🆔 @hackernewsgithubprojects

📦 anthropics/sandbox-runtime

Sandbox Runtime: Secure AI Agents Without Containers

Lock down a program so it cannot touch your secret files or connect to unauthorized websites, all without spinning up a heavy container. This is the core idea behind Anthropic Sandbox Runtime, a lightweight tool that wraps any command with strict operating system level boundaries. It is designed to make AI agents and local tools safer by default, meaning a script can only see and access exactly what you explicitly permit. The surprising part is how it handles the internet. Instead of just blocking signals, it routes every single network connection through a local proxy that checks each destination against your allowlist before letting it pass.

🆔 @hackernewsgithubprojects