📦 grieferpig/esp32-s31-linux

Running Linux on the ESP32-S31

Run a full Linux operating system with virtual memory on a tiny microcontroller chip, turning a standard development board into a genuine computer. This project takes Linux version six point twelve and fits it onto the ESP thirty two S thirty one hardware, handling all the messy work of mapping memory and setting up the boot process so the system actually starts. The most fascinating part is that it executes code directly from flash storage without needing to copy everything into the slow external memory first, which saves precious space and keeps things moving.

📰 https://news.ycombinator.com/item?id=49087499

🆔 @hackernewsgithubprojects

📦 drewnekota/cetus

Cetus: The Desktop Assistant for Coding Agents

Cetus is the macOS desktop layer that turns your coding agents into an always-on assistant by summoning them from anywhere, scheduling background runs, and feeding them context from your screen. Instead of burying your work in terminal tabs, Cetus lets you launch Claude or Codex with one hotkey, capture what you see, and review results in a single board. It handles the boring parts like scheduling, history, and isolated coding workspaces so you can focus on the actual code.

📰 https://news.ycombinator.com/item?id=49084323

🆔 @hackernewsgithubprojects

📦 0xwilliamortiz/ratchet

Ratchet

Ratchet is the complexity control that finally makes coding agents keep their promises instead of just reading them. Every AI assistant drifts toward adding dependencies and writing extra code, assuming you will catch it later. Ratchet changes that by hooking directly into the agent's editing workflow and measuring every single line written in real time. It tracks new files, external packages, and bloated lines, then immediately interrupts the session with a clear warning if the budget is exceeded. You can set strict limits that block unwanted changes or just get gentle advice on over-engineering.

🆔 @hackernewsgithubprojects

📦 ledat98/nexusrag

NexusRAG: Cited Answers with a Knowledge Graph

Ask questions about your documents and get answers with clickable proof. NexusRAG is a hybrid search system that doesn't just skim text; it builds a living knowledge graph of entities and relationships inside your files. It combines vector search with a reranker and visual intelligence to understand charts and tables alongside the words. The coolest part? Every single claim comes with a four-character citation ID. You can click these badges to jump straight to the exact page, paragraph, and heading in the source document. It works with cloud models or runs fully locally with Ollama, giving you a private, accurate chat experience that actually shows its work instead of guessing.

🆔 @hackernewsgithubprojects

📦 jacob-bd/universal-skills-manager

Universal Skills Manager: Sync AI Tools

Universal Skills Manager lets you find, install, and sync helpful abilities across your favorite coding assistants without the headache. Instead of hunting through settings for Claude, Codex, or Gemini, you just ask your AI to search a central hub, grab a tool, and it handles the download and security check for you. The coolest part? It spots when your skills are out of sync across different apps and copies the latest version to all of them with your OK. It is a simple way to keep your AI helpers consistent and powerful across your whole setup.

🆔 @hackernewsgithubprojects

📦 yijia-xiao/financeharness

FinanceHarness: Autonomous Financial Deep Research Framework

FinanceHarness acts like a tireless analyst that digs deep into financial data on its own. It is an autonomous framework designed to handle complex financial research tasks without needing constant human input. Think of it as a smart assistant that gathers facts, checks sources, and pieces together insights automatically. This is useful because financial analysis usually takes hours of tedious work, but this tool speeds up the process by doing the heavy lifting. It helps developers and researchers focus on interpretation rather than manual data crunching. The real cool factor is how it handles deep research tasks autonomously, making complex financial insights easier to access.

🆔 @hackernewsgithubprojects

📦 akii09/pdfx

PDFx

PDFx lets you build PDFs in React using copy-paste components that live inside your own project instead of hidden behind a library dependency. You just run a quick command to grab the code for headings, tables, or badges, and the tool rewrites the paths so everything works out of the box. This means you own the code completely, you can tweak any detail without fighting a black box, and you never have to worry about breaking updates from a distant package. It is a simple, transparent way to generate professional documents while keeping full control over the final result.

🆔 @hackernewsgithubprojects

📦 rightnow-ai/memoir

Memoir: Weights That Never Freeze

Memoir is an open-source research project building a transformer architecture where model weights stay plastic instead of freezing after training. The system uses four weight tiers on different timescales and a 'pondering core' that lets the model think deeper before answering. The repository is unique because it documents a pre-registered experiment that actively tried to falsify the project's main claim. The results showed that writing to memory while pondering actually hurts performance compared to reading only, proving that negative results are just as valuable as positive ones in science. This repository offers a transparent look at how AI research handles failure and iterative learning.

🆔 @hackernewsgithubprojects

📦 pyang5166/gbro-cover-design

Gbro Cover Design

Feed any article into an AI agent and watch it instantly craft a polished social media cover image prompt. Gbro Cover Design asks you just three simple questions about your style, face, and mood, then outputs a ready-to-use prompt that guarantees a consistent human face and a perfect vertical layout. It handles everything from catchy titles to precise composition styles like bold gradient backgrounds or clean minimalist designs, so you never have to design anything yourself. You get professional-looking visuals for platforms like Xiaohongshu or WeChat articles in seconds without touching graphic design software. It is the ultimate shortcut for creators who want stunning thumbnails without the tedious effort.

🆔 @hackernewsgithubprojects

📦 hainrixz/the-architect

The Architect: AI Blueprints for Code

Build any software blueprint with The Architect, a Claude Code plugin that interviews you to design complete system plans. It produces self-contained blueprints that another AI instance can build from scratch with zero context. The surprising part is the strict verification: every single build step includes a runnable command to test success, ensuring the AI doesn't guess. You get clear acceptance criteria for everything from simple sites to complex apps, so the builder never gets stuck or over-builds. It handles fourteen different project shapes and checks live registry versions so you don't get outdated code. Instead of coding, you get a verified roadmap that works.

🆔 @hackernewsgithubprojects

📦 trycompai/crm

CRM

CRM flips the script on customer management by making the research agent the star instead of the database. Most tools just store data, but this one runs a durable bot that works on its own schedule, checks email history, and writes findings into the system without you asking. It uses a strict ruleset where no guesses are allowed, meaning every fact comes from hard evidence like signature blocks or actual conversations. You can even watch it think in real time, seeing exactly how it connects dots and decides what matters.

📰 https://news.ycombinator.com/item?id=49141847

🆔 @hackernewsgithubprojects

📦 xmudeeplit/memgraphrag

MemGraphRAG Explained

MemGraphRAG turns messy text into a structured three-layer memory for reliable answers. It builds a schema layer for abstract concepts, a fact layer for specific details, and a passage layer for source text. This system connects everything bidirectionally, allowing it to spot and fix conflicts before answering. By combining semantic similarity with graph logic, it retrieves exactly the right evidence instead of guessing. The project uses Python and standard language models to index your own documents and run question-answering tasks. It is a practical way to make AI understand long documents clearly without losing context.

🆔 @hackernewsgithubprojects

📦 marchoag/claude-code-setup-wizard-md

claude-code-setup-wizard-md

Your AI coding sessions keep losing context and you have to repeat yourself every time you log back in. This repository fixes that with a simple folder structure that keeps your active documentation tiny while archiving everything else for later. You start a session with a command, the AI reads just a few small files, and when you finish, it automatically archives your notes and updates a searchable knowledge base that remembers every bug you’ve fixed. It even tracks how long you worked and forces you to use pull requests for changes.

🆔 @hackernewsgithubprojects

📦 mkultra6969/mk_droidscreencast

MK DroidScreenCast: Your Android Phone as a Desktop App

MK DroidScreenCast turns your Android smartphone into a full desktop companion app that lets you see, record, and control your phone screen from a single window without hunting through terminal commands. It connects via USB or Wi-Fi, automatically fetches the tools you need, and handles mirroring, recording, and file transfers in one clean interface. The standout feature is how it bundles everything into a native desktop experience so you can tap to record, switch between connection methods, and manage files without leaving the app. It solves the mess of juggling multiple command-line tools by keeping the controls visible and accessible in a single place.

🆔 @hackernewsgithubprojects

📦 moenupa/sdpo-cl

SDPO-CL: Why More Supervision Hurts AI Learning

SDPO-CL is the research project that shows why giving AI too much detailed feedback actually makes it forget what it already knows. The team behind this tool is testing how large language models learn new skills over time. They compare two reinforcement learning methods where one gives step-by-step corrections while the other only looks at the final answer. The surprising finding is that the granular approach forces the model to drift too far from its original knowledge. It specializes quickly but pays for it by losing prior capabilities. The group-based method proves much safer for keeping everything stable.

🆔 @hackernewsgithubprojects

📦 emily2040/seedance-2.0

Seedance 2.0 Skill OS

Seedance 2.0 Skill OS lets you direct AI video with the precision of a film director instead of just guessing at prompts. The tool uses a directing engine to analyze the dramatic intent of a scene, then automatically coordinates camera angles, lighting, and sound to match that single vision. This means you stop writing vague descriptions and start creating coherent stories where every clip maintains a consistent voice from start to finish. It handles everything from initial idea interviews to complex multi-clip sequences and even safe rewrites for copyrighted characters. This is a practical toolkit for turning rough concepts into professional, story-driven video projects without the technical headache.

🆔 @hackernewsgithubprojects

📦 solr159/javboss

JavBoss: Your Private Adult Video Library Manager

JavBoss lets you instantly organize and browse local adult video collections without any manual tagging. You simply point the app at your video folders, and it automatically scans the files, generates cover screenshots, and even recognizes specific adult titles to fetch detailed metadata like actress names and studio info from online sources. It keeps your media folders untouched while building a smart, searchable database in the background. The interface is designed to feel like a premium streaming site, letting you filter by performer or series and play videos using a bundled high-performance player.

🆔 @hackernewsgithubprojects

📦 theorcdev/shadscan

shadscan: Your UI’s Silent Guardian

shadscan inspects your React shadcn app to score its UI fundamentals from zero to one hundred, showing you the evidence behind every finding. It catches the product details that are easy to postpone, like command menus, theme shortcuts, and accessible controls. The default scan is deterministic and read only, so it never starts your app, edits files, or calls an AI model. It just reads your code and gives you a grade. You can even paste the findings into a coding agent to fix them automatically. It is the easiest way to ensure your app is polished, accessible, and ready for production without doing the heavy lifting yourself.

📰 https://news.ycombinator.com/item?id=49143742

🆔 @hackernewsgithubprojects

📦 everfern-ai/everfern

EverFern: The AI Agent That Runs Locally

EverFern is the desktop AI agent that finally lets you hand over your computer tasks without sending your data to the cloud. It controls your mouse and keyboard just like a human would, so you can ask it to research the web, organize files, or write code, and it does it all while running entirely on your own machine. Unlike most competitors, nothing leaves your device, meaning your private information stays safe and you never need a pricey subscription. The real magic is in its self-healing design; if a step fails, the agent automatically figures out why, rolls back changes, and tries again without you needing to babysit it.

🆔 @hackernewsgithubprojects

📦 fecarrico/a11y.md

A11Y.md: The AI Accessibility Context System

A11Y.md is the context system that finally makes your AI write accessible code from the very first line. Most developers treat accessibility as a painful post-launch cleanup, but this repository flips the script by turning strict WCAG rules into a persistent instruction layer that guides your coding agents like Cursor or Claude. Instead of guessing, the AI follows an eleven-point behavioral contract that forces it to use real buttons instead of fake ones, manages focus automatically, and stops hallucinating broken interfaces. It even loads only the specific accessibility guides it needs on demand, keeping things sharp and lightweight.

📰 https://news.ycombinator.com/item?id=49144280

🆔 @hackernewsgithubprojects