📦 typellm/typellm

TypeLLM: The End of LLM Hallucinations

Your large language model is about to stop making up facts. TypeLLM wraps standard AI models in a layer that guarantees every answer fits a specific structure you define. It works without changing the model's underlying brain, meaning you get accurate, structured data like exact numbers or strict categories instead of messy guesses. This is huge because it solves the biggest headache in AI development, which is trusting the output. You define the rules, and the model is forced to follow them. It’s like giving the model a form to fill out where it literally cannot write outside the lines.

🆔 @hackernewsgithubprojects

📦 jankeesvw/omarchy-meeting-recorder

Meeting Recorder: Your Private Meeting Transcriber

Grab a local transcript of every call you take without sending a single byte to the cloud. Meeting Recorder captures your microphone and computer audio as two separate tracks, ensuring you never miss a word, even if someone talks over background music. It runs entirely on your machine using open source tools, so your sensitive conversations stay strictly private. The coolest part is that it figures out who said what, assigns names to each speaker, and even lets you hook up your own scripts to automatically save notes to your favorite app. No bots join the call, and nothing leaves your computer.

🆔 @hackernewsgithubprojects

📦 sopersone/cabbage-trading-machine

Cabbage: The Honest Crypto Bot

Cabbage is the spot trading bot that finally makes its backtesting honest. It runs a simple strategy using Relative Strength Index and moving averages to decide when to buy, but the real value is in the verification. Before you risk a single euro, you can simulate years of historical data offline to see exactly how the logic performed. It also supports paper trading, letting you run live market data without placing real orders. This setup ensures you know the system works before you connect your actual exchange keys.

🆔 @hackernewsgithubprojects

📦 eolix/photosuite

PhotoSuite: The Open Source Photoshop Alternative

Edit images exactly like classic Photoshop, but on your own computer with total privacy. PhotoSuite is a desktop app that faithfully recreates the familiar interface and tools you know, right down to the keyboard shortcuts. The coolest part is that it saves files in the native Photoshop format, so you can work in this free tool and open your files in Photoshop without losing a single layer or effect. It runs entirely offline, meaning your photos never leave your machine. If you want the comfort of a professional editor without the subscription, this is your new go-to tool.

📰 https://news.ycombinator.com/item?id=49847025

🆔 @hackernewsgithubprojects

📦 majidmanzarpour/blender-game-skills

Blender Game Skills

An AI agent now measures your 3D models against your reference art to guarantee they fit your game before you even see them. Blender Game Skills is a collection of instructions that turns flat concept art into fully rigged, game-ready assets for Unity or Unreal. The coolest part is that it refuses to guess. Instead of trusting your eyes, it renders the model and measures the silhouette against your sketch, flagging errors like a head being twelve percent too tall. It fixes these numbers until the math is perfect. It’s like having a strict junior artist who only works with a ruler, saving you hours of rework.

🆔 @hackernewsgithubprojects

📦 embodiedswe/embodiedswe

How AI Robots Solve Real Tasks

A coding agent can now write the code to make a robot assemble a lightbulb, pack a carton, or fold a t-shirt, and EmbodiedSWE is the open framework that makes this possible. Instead of manually programming every movement, you simply ask an AI to solve a specific physical task. The system then takes that single verified solution and automatically generates thousands of variations to train broader, more capable robot policies. This turns one successful demo into a massive dataset for general learning, shifting the focus from manual coding to autonomous simulation. It is a fascinating look at how software agents are beginning to drive physical robotics development.

🆔 @hackernewsgithubprojects

📦 steveyegge/beads

Beads: The Memory Upgrade for AI Agents

Beads is a distributed graph issue tracker that gives AI coding agents persistent, structured memory. It replaces messy markdown plans with a dependency-aware database, allowing agents to handle long-horizon tasks without losing context. By using a version-controlled SQL backend, it enables multiple agents to work on different branches and sync their changes without merge conflicts. This means your AI assistant can remember project insights, track complex dependencies, and coordinate work across machines. It is not just a to-do list; it is a shared brain for your development workflow. If you are tired of agents forgetting what they just did, Beads provides a reliable way to keep them focused and productive.

🆔 @hackernewsgithubprojects

📦 geoffwatts/ninfer-v100

Ninfer V100: Running Massive AI Models on a Single Card

Run massive artificial intelligence models directly on a single graphics card instead of needing a cluster. Ninfer V100 solves the expensive hardware problem by carefully squeezing large language models into limited memory without losing speed. The real surprise is how it handles specific card types, allowing heavy models like Qwen to run smoothly on older hardware that usually requires expensive upgrades. It works by managing memory tightly and using optimized calculations to keep the process fast. This gives developers a practical way to test and deploy large models without a massive budget.

🆔 @hackernewsgithubprojects

📦 open-bfme/open-bfme-1

open-bfme-1: The Lord of the Rings Modding Revolution

open-bfme-1 is rewriting the legendary Lord of the Rings game from scratch, aiming to produce a source code file that compiles into the exact same file as the original retail executable. This byte-perfect recreation is the only way to unlock massive modern updates like sixty frames per second and support for multiple processors without breaking compatibility. The project is currently using artificial intelligence agents to systematically reverse-engineer the original code, verifying every single function before adding it to the open source codebase. It is a massive, collaborative effort to bring a classic title into the modern era with features you have only dreamed of.

📰 https://news.ycombinator.com/item?id=48973264

🆔 @hackernewsgithubprojects

📦 tamaratran/jev-pruner

Jev Pruner

Jev Pruner is the AI coding tool plugin that keeps your context window clean by removing unnecessary terminal noise before the model sees it. When you run a command that spits out thousands of lines of logs, this tool steps in to cut away the repetitive progress updates while protecting the actual errors and results you need. It uses a smart scoring system to decide what matters, so the AI focuses only on the important parts without wasting space on fluff. This makes your coding sessions faster and cheaper, ensuring your AI assistant has the clear, concise data it needs to help you work efficiently.

🆔 @hackernewsgithubprojects

📦 luc4n3x/levyra-deepsound

Levyra DeepSound: The Private Music Player That Actually Works

Play your music without handing your data to a corporation or logging into another account. Levyra DeepSound is an open source player for Android and Windows that keeps your library completely private while pulling in rich details like synced lyrics and high quality artwork. It even lets you stream live internet radio stations directly inside the app. The real magic is the recognition tool that identifies a song playing on your device just by listening to it. It is the rare music app that respects your privacy while still giving you a surprisingly smart listening experience.

📰 https://news.ycombinator.com/item?id=49848394

🆔 @hackernewsgithubprojects

📦 aowang-ai/jev-trade

Jev Trade: The AI Crypto Trader

An AI model named Jev sits in the driver's seat for real money on the Hyperliquid exchange, making instant buy, sell, or hold calls every two seconds. This repository runs a live trading bot for five major coins, where Jev analyzes the market and decides exactly when to act. The coolest part is that Jev often chooses to hold, meaning it actively decides to do nothing. You can watch every decision on a live dashboard, seeing the AI think through the market and execute real trades with actual wallets. It is a fascinating look at how artificial intelligence handles high-stakes financial decisions.

🆔 @hackernewsgithubprojects

📦 wildminder/awesome-qwen-image

Qwen Image 2.1: The One-Stop Hub

Alibaba's latest image model, Qwen Image 2.1, generates pictures and edits them using just ten reference images in a single step. This is where things get messy for developers, though. The model is huge, with a text encoder that takes up nearly eighteen gigabytes of storage on its own. The GitHub repository, Awesome Qwen Image, solves this by acting as a single map for every version of the model. It lists official downloads, lighter versions for your laptop, and tools that improve your prompts. You do not need to guess which file to download. The repository organizes everything by hardware and task, so you can find the right version quickly.

🆔 @hackernewsgithubprojects

📦 ameyanagi/reshiki

ReShiki: The Rust-Based Molecular Drawing Workspace

ReShiki is a standalone desktop application that lets chemists draw complex molecules and reaction schemes without needing to install Python or RDKit, handling all chemical logic entirely in Rust. It functions as a complete workspace where you can sketch structures, apply professional styling, and export high-quality figures directly from your computer. The real surprise is that it runs completely offline, meaning your data stays private while you work. This tool bridges the gap between quick sketching and publication-ready figures, making it a fascinating option for developers interested in high-performance scientific software.

🆔 @hackernewsgithubprojects

📦 harbor-framework/terminal-bench-2-1

Terminal Bench 2.1

Terminal Bench 2.1 is the benchmark suite that actually tests if your A.I. can finish hard jobs. It gives agents real container environments to fix bugs, assemble proteins, or patch security holes. This update fixes twenty-six previous tasks to stop agents from cheating to get high scores. It matters because we need to know if A.I. can handle messy, real-world work. If you care about A.I. that delivers results, this is the standard to watch. It is the honest test for the future of digital workers. See you in the next one.

🆔 @hackernewsgithubprojects

📦 liquid4all/liquid-audio

Liquid Audio: Real-Time Speech AI

Liquid Audio is the speech foundation model that finally lets AI talk to you in real time without lag. Built by Liquid AI, it handles the entire loop, hearing you and answering with natural voice instead of just reading text. The coolest part is how it processes sound and language together, so it can hold a conversation while still understanding exactly what you meant. It is lightweight enough to run on regular hardware, making interactive voice agents actually feasible for developers. If you have been waiting for an open model that handles audio like a first-class citizen, this is the one to check out.

🆔 @hackernewsgithubprojects

📦 clerk/skills

Clerk Skills

Clerk Skills is the cheat code that turns your AI coding assistant into a Clerk expert. It stops the model from guessing how to handle authentication by giving it a complete, organized library of proven patterns. The coolest part is that it actually thinks. When you ask your agent to add login to your Vue or Next dot JS app, a router skill instantly picks the right guide from the bunch. It knows exactly which hooks to use or how to protect your server routes. No more messy, half-finished code. You just get clean, production-ready authentication logic that fits your specific framework perfectly.

🆔 @hackernewsgithubprojects

📦 getsentry/skills

Sentry Skills: The Open Source Playbook for AI Coding

Sentry has published the exact instruction sets their engineers use to program their AI coding assistants. This repository, called Skills, acts as a shared memory bank that teaches artificial intelligence how to think like a Sentry developer. It covers everything from writing clean code to navigating their specific security standards. The most surprising part is that these are not just generic tips; they are precise, tested workflows that ensure every line of code generated matches the company's highest standards. By opening up these internal tools, Sentry gives every developer a free, professional-grade cheat sheet for training their own AI helpers.

🆔 @hackernewsgithubprojects

📦 mastra-ai/skills

Skills: The Knowledge Layer for Mastra

Skills is the knowledge layer that fixes outdated coding agent memory for the Mastra framework. You know how AI assistants often write code that looks right but fails because they rely on old training data? This repository solves that by giving agents a fresh, verified map of current APIs and workflows. Instead of guessing, the agent checks live documentation directly from your installed packages or remote sources before writing a single line of code. It ensures your AI assistant actually understands your specific version of Mastra, preventing frustrating errors and broken builds. This means you get accurate, production-ready code every time, without manually fact-checking the AI’s every suggestion.

🆔 @hackernewsgithubprojects

📦 mcollina/skills

Skills: Teaching AI to Code Node.js

Matteo Collina, the lead maintainer of the Node.js project, has published a public library of instructions called Skills that significantly boosts the accuracy of AI coding assistants. The repository contains carefully crafted text files that teach large language models specific best practices for building fast and secure Node.js applications. When you feed these instructions to an AI, the model stops guessing and starts writing code that follows professional standards. This is interesting because it shows that we can improve AI output simply by providing better context and clear guidelines. It turns a generic chatbot into a specialized helper that understands the nuances of server development.

🆔 @hackernewsgithubprojects