📦 renatoworks/oh-my-reddit

oh-my-reddit

Stream your favorite Reddit threads live inside your terminal with beautiful styling and smooth animations. This Go-built terminal application lets you browse subreddits, read posts, and watch new comments glide in one by one as they are published. It securely grabs your existing browser login session, removing the need for annoying OAuth setups or password entry. If you are on a Mac, it can even read the incoming feed out loud using natural system voices. You can also connect your OpenAI key to instantly generate live sentiment cards or ask questions directly about the thread.

📰 https://news.ycombinator.com/item?id=48563294

🆔 @hackernewsgithubprojects

📦 awslabs/cli-agent-orchestrator

CLI Agent Orchestrator

You can now run a team of elite AI coding assistants in parallel, right from your command line. cli-agent-orchestrator is an ingenious tool that lets a single supervisor agent coordinate multiple specialized worker agents across different tools like Claude Code, Gemini, and GitHub Copilot. It runs each developer agent in its own isolated background terminal using tmux, allowing them to collaborate, hand off tasks, or run parallel code reviews. Since they run as full command-line processes, they keep all their native features and authentication. You can even attach to any background session to guide them manually whenever they need a hand.

🆔 @hackernewsgithubprojects

📦 deepreinforce-ai/ornith-1

Ornith-1

Ornith-1 is the self-improving coding model that actually learns how to build its own step-by-step problem-solving guides. While typical AI models simply generate code, this open-source project teaches itself by optimizing both its coding decisions and the scaffolding that structures those decisions. By using reinforcement learning to explore better thinking paths, it regularly beats much larger models on challenging programming benchmarks. Whether you want a compact model to run locally or a massive mixture-of-experts model, it delivers top-tier performance for automation tasks. Check out Ornith-1 online and start building smarter AI programming assistants today.

📰 https://news.ycombinator.com/item?id=48722052

🆔 @hackernewsgithubprojects

📦 itsthelore/rac-core

rac-core

rac-core is the project context manager that stops AI coding assistants from repeating mistakes your team already solved. When working with AI, we often waste time watching agents write code that violates architectural decisions we made months ago. This tool fixes that by turning team decisions, designs, and roadmaps into structured knowledge that agents can actually read and follow. It plugs directly into your workflow to enforce product rules during development. If you want your AI tools to behave like a smart team member instead of a forgetful intern, check out rac-core and start keeping your agent grounded.

📰 https://news.ycombinator.com/item?id=48714880

🆔 @hackernewsgithubprojects

📦 leeyeel/sketch2motion

Sketch2motion

Sketch2motion is the animation tool that turns your flat, boring doodles into smooth hand-drawn videos. Instead of trying to animate frame-by-frame, you just upload any sketch or photo, and this clever tool converts it into vector paths and uses a math-rendering engine to draw it right before your eyes. You can tweak the speed, choose a drawing style like smooth or wiggly, and download the finished clip. It even pauses on the final frame automatically so it looks super polished. It is basically a quick way to bring your casual whiteboard ideas or notebook doodles to life with zero effort.

🆔 @hackernewsgithubprojects

📦 calesthio/breakoutanalysis

BreakoutAnalysis

An automated market scanner called breakoutanalysis wakes up every morning to scan thousands of stocks, identify explosive breakout moves, and deliver visual trading intelligence directly to your Discord or email. Instead of staring at charts all day, this tool filters out penny stock noise by demanding high relative volume and verifying multi-year price strength through historical data. Once a true mover is detected, an integrated artificial intelligence model automatically reads the latest news to write a clear explanation of why the stock is surging. It even snaps a live chart screenshot so you get a complete, effortless visual briefing before the market opens.

🆔 @hackernewsgithubprojects

📦 danmoreng/qwen-tts-studio

Qwen-TTS Studio

The local text-to-speech app that finally runs high-quality voice generation entirely on your own machine without a complex Python setup. Qwen-TTS Studio is a slick desktop interface for Windows and Linux that gives you a fast C++ speech engine powered by your CPU or Nvidia card. Instead of sending your voice data to some corporate cloud, you can generate natural speech, play with tone instructions like whispering, and even clone custom voices from a tiny audio clip, all completely offline. It is the ultimate playground for running advanced speech models locally without the headache. Give it a spin and take back your voice privacy.

🆔 @hackernewsgithubprojects

📦 pham-tuan-binh/leslider

LeSlider

You can now expand a standard six-axis robot arm into a seven-degree-of-freedom system using a single extra servo motor and some 3D-printed parts. The leslider project is an open-source hardware and software add-on designed for the SO-101 robot arm. By mounting the arm on a linear V-slot slider, it introduces a seventh travel axis that integrates with the LeRobot framework. It supports both velocity-based steering and a calibrated position mode, allowing you to control linear travel alongside the arm using a keyboard. It lets you teleoperate, record unified datasets, and train AI policies on the entire seven-joint setup.

🆔 @hackernewsgithubprojects

📦 taylorhuston/local-life-manager

Local Life Manager

Have Claude Code run your entire day by transforming your terminal into a personalized command center. This framework links your local notes and code repositories to custom AI agents, using dozens of tailored skills to orchestrate your daily routine. Instead of juggling generic prompts, you can trigger deep-dive research, manage daily journals, structure learning sessions, or even have the AI read and summarize your RSS feeds and YouTube subscriptions directly into your Obsidian vault. It acts as a clever coordinator that remembers your preferences, plans your tasks, and acts as a dedicated study partner. It is a brilliant way to build a private, localized second brain.

🆔 @hackernewsgithubprojects

📦 gabrieletrovato01/koreshelf

KoreShelf

Organize your digital books in a stunning virtual room and turn them into custom brain food for artificial intelligence. By dragging your electronic documents into this local environment, you get a beautiful virtual bookshelf complete with moving bookmarks and a built-in reader. But the coolest trick is the export button, which strips away noisy formatting from your books to create a clean, structured document. You can drop this directly into your favorite artificial intelligence companion to chat with the text, ask about the plot, or get quick summaries. It is the ultimate bridge between your personal library and smart tech.

🆔 @hackernewsgithubprojects

📦 capitan01r/comfyui-krea2t-enhancer

ComfyUI Krea2T Enhancer

Make AI images follow written prompts with much higher accuracy. A new custom extension called comfyui-krea2t-enhancer hooks directly into the text-fusion path of Krea2 diffusion models to solve the common issue of ignored details. Instead of messing with text prompts or adding complicated extra nodes, this extension applies a controlled conditioning adjustment directly inside the model during sampling. You simply drop the node between your model loader and the sampler to blend the enhancement strength to your liking. It is a clean, lightweight way to get precise control over your image generation pipeline without cluttering your workflow.

🆔 @hackernewsgithubprojects

📦 gazingstars123/anima-standalone-trainer

Anima Standalone Trainer

Train personalized AI art styles on your own computer with a dedicated, lightweight environment that focuses strictly on the Anima model. This training tool packages everything you need into a single desktop interface, solving the headache of complex command lines and messy dependencies. It gives you a clean web control panel to manage your training jobs, adjust model paths, and fine-tune your parameters. What makes it really special is how it optimizes multi-GPU setups on Windows and Linux, letting multiple graphics cards share the heavy lifting to speed up your training. Now you can easily create custom styles without fighting complicated software configurations.

🆔 @hackernewsgithubprojects

📦 avadiii/rewardspy

rewardspy

rewardspy is the plug-in debugger that catches reinforcement learning reward hacking before it ruins your model's training run. When you are training a model, the reward curve might look like it is climbing straight to success, but your model might actually just be exploiting a loophole, like over-writing its own tests or inflating response lengths. With one decorator import, this tool observes your reward function without modifying it, tracking statistical anomalies like variance collapse or component dominance in real time. It displays everything on a live terminal dashboard and can even automatically halt training if it detects cheating.

🆔 @hackernewsgithubprojects

📦 spikelab-jhu/isaac-sim-event-camera-plugin

isaac-sim-event-camera-plugin

isaac-sim-event-camera-plugin is the physics simulator extension that generates high-speed event camera data without slowing your computer to a crawl. Standard virtual event cameras require rendering hundreds of frames a second, which completely tanks simulation performance. This clever tool solves that by rendering only a few key frames and using smart motion-vector math to fill in the gaps. It gives you hyper-realistic, thousand-hertz event data while cutting down actual rendering work by up to eight times. It even features a live preview window so you can watch simulated robot eyes react in real time. Download this plugin to instantly add fast, efficient event-camera sensing to your robotics projects.

🆔 @hackernewsgithubprojects

📦 artokun/comfyui-mcp

comfyui-mcp

comfyui-mcp is the Claude Code plugin and Model Context Protocol server that lets you edit and run your live ComfyUI graph directly from an AI chat session. Instead of tangling with spaghetti nodes by hand, you can just tell the AI to build workflows, swap models, or generate art. It packs over a hundred tools and specialized skills for heavy-lifters like Flux and WAN video, meaning the AI actually understands sampler settings and resolutions without breaking your setup. It even embeds an autonomous agent panel right into your sidebar. It is the ultimate way to orchestrate complex AI generation with pure, conversational ease.

🆔 @hackernewsgithubprojects

📦 fredantb/spec-driven-development

Spec-Driven Development

spec-driven-development is the Claude skill that finally stops your AI coding assistants from contradicting each other. If you have ever used Claude Code, Cursor, and Copilot on the same project, you know the pain of one agent completely overwriting another agent's progress because they do not share a single source of truth. This clever tool fixes that by interviewing you first, generating three unified markdown files for your requirements, design, and tasks, and then writing matching configuration files for all your favorite AI editors. Now, every single assistant reads the exact same playbook, stopping architectural drift and wild code hallucinations instantly.

📰 https://news.ycombinator.com/item?id=48221805

🆔 @hackernewsgithubprojects

📦 harshuljain13/llm-inference-at-scale

llm-inference-at-scale

Serving a large language model is completely different from traditional software because every single response has a highly unpredictable cost. If you are trying to understand how to handle this, the project llm-inference-at-scale is a comprehensive, open-source handbook designed specifically to teach you how to serve these models in production. It breaks down the complex mechanics of GPU hardware, memory limits, and optimization strategies into easy-to-digest chapters with practical labs. You will learn how systems manage memory during conversations and how to handle massive traffic. Check out the repository to master the engineering behind running models efficiently.

📰 https://news.ycombinator.com/item?id=48424467

🆔 @hackernewsgithubprojects

📦 jia-lab-research/unityshots

UnityShots

Most video generators struggle to keep characters and settings consistent when switching between different camera angles, but unityshots fixes this by generating an entire multi-shot film with synchronized audio in one go. It uses a smart memory system to remember what characters look like and how the environment looks across up to nine different camera cuts. This means a character keeps the same clothes and face from shot to shot, while the background music and ambient sounds transition smoothly. It turns a single-shot video generator into a complete, multi-shot storyteller that keeps everything in perfect harmony.

🆔 @hackernewsgithubprojects

📦 porunc/codewiki

CodeWiki

You can now transform any messy codebase into an interactive, structured developer wiki with a single command. CodeWiki is a local intelligence tool that scans your repository, builds a complete code graph of definitions, imports, and calls, and uses a local database to generate deep, source-grounded documentation. It completely bypasses manual writing by mapping your systems, workflows, and API layers into a visual graph explorer and a searchable web UI. It even has a lightweight mode that runs entirely without AI, feeding fast symbol searches and call traces straight to your favorite editor or assistant. It is like having a sharp developer who already knows your entire system inside out.

🆔 @hackernewsgithubprojects

📦 ch-bas/threejs-sims-house-builder

threejs-sims-house-builder

Design and build your dream multi-floor home directly in your web browser. This project brings a complete grid-based architecture and interior design system to life using three-dimensional rendering. You can construct up to four floors, paint walls with custom patterns, lay down wood or tile flooring, and arrange dozens of different furniture pieces. There is even a first-person walkthrough mode to explore your creations, wandering computer-controlled characters, and a dynamic day-and-night cycle that casts realistic shadows. It is the perfect playground for anyone who loves design, offering a fully interactive, lightweight home builder without needing to install a thing.

🆔 @hackernewsgithubprojects