📦 liunian06/ko-lesson

ko-lesson

College students can now convert messy, chaotic class slides and PDFs into a structured study system for Obsidian. Instead of just summarizing documents, ko-lesson works as an interactive study assistant that maps out a customized learning path based on your real course materials. It reorganizes dense slides into readable markdown files, builds concept flashcards, tracks your weak points, and even generates practice exams with customized follow-up tests for your mistakes. Whether you want to study one lesson at a time with active feedback or generate a massive exam cram package overnight, it transforms confusing materials into an organized workspace you can actually use to ace your finals.

🆔 @hackernewsgithubprojects

📦 nerd-lang/nerd-lang-core

Nerd Lang Core

Write backend logic with fifty percent fewer tokens using a language designed for AI models instead of humans. That is exactly what nerd-lang-core does. Since machines write nearly half of our code today, this project ditches braces, semicolons, and complex syntax in favor of plain English words that tokenize perfectly for LLMs. This drastically cuts down token usage and lowers API billing costs while keeping everything easily auditable by humans. It even compiles straight to native code using LLVM, giving you lightning-fast performance alongside built-in agent capabilities like direct Claude API calls and HTTP requests. Let the bots write the code and save a fortune on your next LLM project.

🆔 @hackernewsgithubprojects

📦 christopherkarani/espresso

Espresso

You can now run and train machine learning transformers directly on Apple's Neural Engine, bypassing CoreML entirely to get a massive speed boost. A clever project called espresso uses reverse-engineered private Apple APIs to compile models straight to the silicon. By staying completely on the hardware and avoiding the overhead of standard Apple frameworks, it runs token generation up to four times faster than CoreML. It utilizes zero-copy memory buffers and compiles the model once to reuse across steps, which stops the processor from constantly re-analyzing the network. It is a brilliant way to unlock the true potential of your Mac.

📰 https://news.ycombinator.com/item?id=48794803

🆔 @hackernewsgithubprojects

📦 loredcast/filewizard

filewizard

filewizard is the self-hosted media hub that finally runs all your heavy-duty file conversions and transcriptions locally through a single dark-mode dashboard. Instead of uploading your private documents to random internet converters, you can spin up this lightweight container to handle everything on your own hardware. Under the hood, it elegantly ties together classic command-line giants like FFmpeg, LibreOffice, and Pandoc, while adding modern machine learning tools for speech-to-text and optical character recognition. It simplifies complex multi-tool workflows into a drag-and-drop web interface where you can convert documents, transcribe audio, or extract text from scanned images with complete privacy.

🆔 @hackernewsgithubprojects

📦 danielpodolsky/ownyourcode

OwnYourCode

Most AI coding tools are designed to write your code for you, but they can easily leave you with an application you do not actually understand or know how to maintain. A new tool called ownyourcode completely flips this dynamic by turning the AI into a mentor rather than a coder. Instead of writing the implementation, the assistant guides you through Socratic questioning, strict quality gates, and structured design reviews. You are forced to predict code behavior and write the actual lines yourself, ensuring you build real mental ownership of the project while maintaining a clear web dashboard of your progress.

🆔 @hackernewsgithubprojects

📦 brikerman/skilless.ai

Skilless AI

Equip your favorite AI agents with the practical, real-world capabilities they need to actually get things done. While models like Claude or Cursor are great at writing and thinking, they quickly get stuck when you ask them to browse the live web or analyze local video files. skilless.ai solves this by acting as a lightweight, zero-configuration toolkit that integrates into your agent's workflow. It instantly gives your assistant the ability to perform semantic web searches, parse messy web pages into clean text, extract transcripts from online videos, and even convert media formats. It sets up in a single command, running entirely in an isolated environment without cluttering your system, making...

🆔 @hackernewsgithubprojects

📦 ent0n29/polybot

Polybot

An open-source trading toolkit called polybot reverse-engineers high-frequency trading strategies for Polymarket prediction markets. Developed as a multi-service Java system, it lets you dissect how successful traders operate, run simulated paper trading, and execute live market-making strategies. The system tracks complete-set arbitrage opportunities in binary markets, using a real-time event pipeline with ClickHouse and Redpanda to ingest trade data. It automatically balances inventory between opposing market outcomes to hedge risk while managing order replacements dynamically. It is a fascinating look at the high-speed infrastructure powering modern prediction markets, offering a complete framework for strategy research and execution.

🆔 @hackernewsgithubprojects

📦 evan-kolberg/prediction-market-backtesting

Prediction Market Backtesting

Test your forecasting strategies against historical trading data using this sophisticated simulator. It plugs right into a powerful trading engine to let you simulate how your ideas would have performed in real-world environments, specifically targeting prediction platforms like Polymarket. The project handles the complex math of order book depth, trading fees, and realistic order matching, so you can see if your market predictions actually translate into profitable trading rules. It also features stunning portfolio charting to visualize your returns and drawdowns, making it incredibly easy to see exactly where your strategy succeeded or struggled before you risk a single real dollar.

🆔 @hackernewsgithubprojects

📦 chekusu/mails

mails

Give your AI agents their very own email addresses to send, receive, and read messages programmatically. This project lets you spin up a custom mailbox for an AI assistant using a simple command-line tool or software kit. It handles the tricky parts of email by routing messages through Cloudflare, automatically parsing files, and even extracting verification codes from incoming text. You can save everything locally in a small database or search through messages in the cloud. It is a brilliant way to let digital assistants communicate with the real world using the universal standard of email.

🆔 @hackernewsgithubprojects

📦 adam-s/car-diagnosis

Diagnose Car Problems by Sound with AI

An open-source project called car-diagnosis uses a clever machine learning pipeline to identify mechanical car issues using nothing but audio clips from a phone. By feeding noisy audio through a multi-stage cleaning cascade, it automatically filters out speech, wind, and music to isolate the raw mechanical sounds. The system then runs this clean audio through a frozen contrastive language-audio pretraining model and small linear heads to estimate what is wrong, where the fault is, and which parts are likely failing. Instead of overconfident guessing, it is calibrated to return an uncertain status when the audio quality is too low to make a reliable call. It is a highly practical, honest...

📰 https://news.ycombinator.com/item?id=48749868

🆔 @hackernewsgithubprojects

📦 emilycodes-cmd/claude-code-ollama-local

Claude Code Ollama Local

claude-code-ollama-local is the Windows setup helper that lets you run Anthropic's terminal assistant completely offline without paying for an API key. Instead of sending your code to external servers, this clever configuration redirects everything to your own machine using Ollama. It bundles a simple launcher script to swap the default cloud connection with lightweight, local AI models like Qwen. It even includes handy troubleshooting tips for forcing CPU mode if your graphics card throws a fit. It is the perfect way to get smart terminal help locally, keeping your code private and your wallet happy.

🆔 @hackernewsgithubprojects

📦 penligent/ai2pentesttool

ai2pentesttool

Setting up a security testing lab can be a massive headache of broken dependencies and manual package troubleshooting, but ai2pentesttool fixes this by using a smart AI agent to configure and install your entire security suite in a single click. Instead of blindly running shell scripts, this tool uses GPT-4 to analyze your specific operating system, generate a custom installation plan for fifteen essential security tools, and dynamically rewrite commands if a download fails. If you are offline, it instantly falls back to a robust built-in configuration database. It turns a tedious afternoon of system troubleshooting into a flawless, hands-off setup.

🆔 @hackernewsgithubprojects

📦 dzhng/skills

Skills

Skills is the modular library that helps your AI agents build software autonomously instead of just writing lines of code. It changes how you work with AI by breaking massive development goals into small, independently verifiable slices, almost like clearing a fog of war. The project gives your agent structured procedures for everything from interviewing you to map out project unknowns, to performing visual screenshot reviews, and even rewriting its own development plans mid-run when things change. It turns AI assistants into systematic software factories that safely test and verify their own work until the goal is truly done.

🆔 @hackernewsgithubprojects

📦 meituan-longcat/sglang-fluentllm

Sglang-fluentllm

Sglang-fluentllm is the high-performance inference engine that finally stops hardware bottlenecks from slowing down your largest AI models. By combining model design directly with system engineering, this clever tool refactors how computers process text generation. It merges multiple decoding steps into a single, unified execution path on the graphics card, which dramatically cuts down on wasted processing overhead. This means your large language models can handle massive conversation histories and complex tasks without hitting a performance wall or lagging mid-sentence. If you want to run massive open-source models at maximum efficiency, this is the smart engine to deploy.

🆔 @hackernewsgithubprojects

📦 storybloq/storybloq

Storybloq

Keep your AI coding assistants from forgetting yesterday's work by turning your session history into active context. While most AI models start every session completely blank, this tool creates a dedicated folder in your repository to store tickets, issues, handovers, and lessons learned as simple files. It works as an MCP server and a CLI tool, allowing assistants like Claude Code to read your active roadmap, pick up where you left off, and even run autonomous workflows through planning and testing. It helps you build on top of past decisions instead of resetting your progress every single time you start coding.

🆔 @hackernewsgithubprojects

📦 scanopy/scanopy

Scanopy

Scanopy is the network documentation tool that automatically keeps your infrastructure diagrams up to date. Instead of spending hours manually sketching network setups in drawing apps only for them to go stale a week later, you just run a single background scanner. It automatically builds and schedules continuous updates for four distinct maps, showing you everything from physical switches and logical subnets to active workloads and application dependencies. It is perfect for anyone wanting a true, live picture of their home lab or company servers without installing software on every single device. Best of all, you can easily export the results to share with your team.

🆔 @hackernewsgithubprojects

📦 numman-ali/n-skills

n-skills

n-skills is the plugin marketplace that finally lets you write AI agent skills once and run them on any platform. While AI tools are evolving fast, they all use different formats, which means your custom prompts and workflows get locked into specific ecosystems. This repository solves that by offering a curated, cross-platform library of high-quality plugins using a single universal standard. You get pre-made skills for browser automation, multi-agent coordination, and repository maintenance that work natively with Claude Code, Copilot, Cursor, and Codex. It is the perfect setup to instantly give your local AI agents superpowers without rewriting code.

🆔 @hackernewsgithubprojects

📦 bytedance-seed/depth-anything-3

Depth Anything 3

You can now reconstruct incredibly detailed 3D scenes from a handful of random images without even knowing the camera angles. Depth anything 3 achieves this by treating the visual space as a simple field of light rays, eliminating the need for complex, heavy architectures. By using a single plain transformer, this model predicts highly consistent multi-view depth, estimates camera poses, and can even output high-fidelity 3D Gaussians for virtual flythroughs. It is incredibly useful for developers building spatial computing tools, robotic vision, or virtual reality projects. Check it out to easily bring flat images into the third dimension.

🆔 @hackernewsgithubprojects

📦 prat011/awesome-llm-skills

Awesome LLM Skills

Teach your favorite AI assistants how to handle specialized, repetitive tasks with customized workflows. The awesome-llm-skills repository is a curated collection of pre-made templates and guides that let you inject fresh capabilities directly into AI command-line tools and desktop apps. Instead of writing massive prompts every single time, you can equip your digital assistants with specific blueprints for organizing messy files, generating changelogs, or even analyzing meeting transcripts. This means your tools can automatically trigger the right workflow for the job, making your daily automation feel truly seamless. Grab a skill, drop it in your local folder, and watch your assistant get smarter.

🆔 @hackernewsgithubprojects

📦 shepherd-agents/shepherd

shepherd

Run your AI agents inside a reversible, Git-like playground where you can safely preview every file change before it ever touches your real directory. That is exactly what this clever library makes possible. Instead of letting an AI agent loose on your actual codebase, it runs the task inside a secure sandbox and saves the work as a reviewable proposal. You get a durable execution trace that lets you inspect, run, select, or discard the proposed changes with simple commands. It makes AI development incredibly safe and stress-free because you remain the ultimate gatekeeper of your code.

🆔 @hackernewsgithubprojects

📦 ekwek1/soprano-factory

Soprano Factory

You can now train a custom text-to-speech model that generates voice audio at up to two thousand times real-time speed on a standard graphics card. A clever new training toolkit called soprano-factory packages this process into a tiny six-hundred-line script. It lets you feed in your own voice recordings to teach a lightweight, eighty-million parameter model how to speak like you, a friend, or any character you want. The resulting voice is crystal clear and runs on a normal laptop using almost no memory. If you want to build super fast, custom voice models on your own hardware, this is the perfect place to start.

🆔 @hackernewsgithubprojects