📦 wumingqi60/lingxi

Lingxi CrossBorder: AI Customer Management for Global Trade

Lingxi CrossBorder turns messy international customer chats into organized business opportunities. It lets small foreign trade teams manage multiple messaging channels, track customer profiles, and handle support tasks all in one dashboard. The standout feature is its built-in AI assistant that helps translate messages, summarize conversations, and suggest polite replies. This means you can actually keep up with customers from different countries without needing to speak every language yourself. It handles the translation and tone so your team can focus on closing deals. If you want to simplify how you talk to global buyers, this is a solid starting point.

🆔 @hackernewsgithubprojects

📦 openmouse-project/openmouse

OpenMouse: Control Your Mouse from the Browser

OpenMouse turns your web browser into a universal control panel for gaming mice. Right now, you need a different app for every brand to adjust settings like DPI and polling rate. This project aims to fix that by letting you connect a mouse and change its settings in one place. The current version is a demo that shows what the interface will look like, but it does not actually talk to a physical device yet. It is built with modern web standards, so you can see the design without installing anything heavy. The team plans to add real hardware support and make the code open source soon.

🆔 @hackernewsgithubprojects

📦 mateaix/mateclaw

MateClaw Is Your AI Second Brain

MateClaw turns your computer into a persistent second brain that remembers everything and acts on your behalf without you needing to micromanage it. Instead of a dumb chatbot that forgets context, this Spring Boot project orchestrates multiple specialized AI agents that collaborate to tackle complex tasks while maintaining long-term memory across sessions. You can hand it skills ranging from writing comics and debugging code to monitoring news feeds, and it handles the heavy lifting by coordinating the right tools and models for each job. It even supports multiple messaging channels so you can interact from wherever you prefer.

🆔 @hackernewsgithubprojects

📦 matevip/mateclaw

MateClaw: Your AI Second Brain

MateClaw turns your messy notes and random thoughts into a smart, persistent second brain that actually remembers what you need later. It builds on Spring AI to let multiple AI agents collaborate, giving your system a long-term memory and a set of specialized skills for everything from drafting documents to generating art. Instead of just chatting once and forgetting, the system tracks your context over time, so it feels like working with an assistant who truly knows your history. You can hook it into different chat apps or use its desktop interface to keep your digital life organized and searchable.

🆔 @hackernewsgithubprojects

📦 jolehuit/clother

Clother: One CLI for All AI Providers

Switching between different artificial intelligence coding assistants usually means messing with environment variables and launcher scripts by hand. Clother fixes that by giving you a single command line interface where you can instantly swap providers like Z.AI, Kimi, or even local Ollama models just by running a simple clother-zai command. It builds lightweight launchers for each service, handling all the key and endpoint setup behind the scenes so you can benchmark speed or resume chats without breaking a sweat. It is the perfect tool for developers who want to compare model performance or switch providers without restarting their terminal.

🆔 @hackernewsgithubprojects

📦 rixinhahaha/snip

Snip: Let AI Agents Draw, Don’t Just Talk

Let AI agents draw diagrams instead of typing endless descriptions. Snip acts as a visual bridge between you and coding tools like Claude. You just ask a question, and the AI renders a flowchart or UI preview right in your menu bar. You can circle mistakes or add notes directly on the image. The AI sees your feedback and fixes the drawing instantly. It works locally on macOS and Linux without sending data to the cloud. You get a clean visual conversation instead of a wall of text. Check it out and see how much faster your coding sessions become.

🆔 @hackernewsgithubprojects

📦 breezewish/codexpotter

CodexPotter Auto-Repair

CodexPotter automatically repairs your codebase by repeatedly checking and fixing your work until it perfectly matches your instructions. It replaces the standard command with a smart loop that forces the AI to review, polish, and reconcile the code across multiple fresh sessions. This approach prevents the AI from losing track of details in long conversations, ensuring high-quality results without constant human intervention. You simply define the goal and let the tool handle the iterative cleanup. It keeps your repository clean and aligned with your original vision, making complex updates feel effortless.

🆔 @hackernewsgithubprojects

📦 ch921-cell/remember-r1

Remember-R1: Reinforcement Learning for Visual Memory

Remember-R1 fixes a common glitch in AI vision models where they simply forget details when asked to analyze long, complex images. Instead of just guessing, this project uses reinforcement learning to teach the model to reward itself for actually looking at the relevant parts of the picture. It encourages the AI to explicitly state what it sees, keep those visual details in mind throughout its reasoning process, and focus attention on the specific areas that matter for the question. The repository provides the code and trained models to make this happen, proving that guiding the learning process helps computers retain visual information much better than standard training methods.

🆔 @hackernewsgithubprojects

📦 hikari-systems/slater

Slater: Graph Database That Fits Gigabytes in Megabytes

Slater is a graph database that lets you query massive datasets with hundreds of millions of nodes using only hundreds of megabytes of RAM. Instead of loading the entire graph into memory like traditional engines, it keeps the data on disk and only caches the parts you actively query. This means a massive graph and a tiny one cost exactly the same amount of memory to serve, allowing you to run countless read replicas without exploding your infrastructure bill.

📰 https://news.ycombinator.com/item?id=48996325

🆔 @hackernewsgithubprojects

📦 shure-dev/small-vlm-sop-check

Small VLM SOP Check: Catching Factory Errors Early

Marlin-2B spots procedural errors in factory video with 40% accuracy. The small-vlm-sop-check project trains tiny AI models under 4 billion parameters to watch first-person factory footage and pinpoint exactly when a worker skips a step or uses the wrong tool. Instead of just recognizing objects, these models learn to find specific event start and end times in a video stream. This allows systems to alert supervisors about safety risks before an accident happens, running locally on edge devices without sending private footage to the cloud. It is a foundational step toward wearable cameras that understand work procedures in real time.

🆔 @hackernewsgithubprojects

📦 tanishqkancharla/calldiff

calldiff: See Code Changes, Not Just Lines

calldiff lets you see how the wiring in your code actually changes. Instead of staring at endless line diffs that hide the real shifts, you get a clean ASCII map of which functions are calling which. It takes your git commits and builds a visual tree of dependencies, showing exactly what got added or removed. This is a lifesaver for code reviews, especially when an AI agent rewires your app logic. You spot the structural damage instantly, rather than guessing from line numbers. It supports twenty-two languages and even gives you structured output for automation. Next time you review a PR, stop reading lines and start reading the flow.

📰 https://news.ycombinator.com/item?id=49250231

🆔 @hackernewsgithubprojects

📦 yzhao062/anywhere-agents

One Config for Every AI Agent

anywhere-agents is a portable configuration tool that lets you manage your AI coding assistants with a single setup file. Instead of manually configuring each tool like Claude Code or Codex separately, this project uses a central file called AGENTS.md to sync rules, permissions, and skills across all your environments. It automatically generates the specific configuration files each agent needs, ensuring consistent writing styles and safety guards. You can easily install it using pipx or npm, add custom skills via a simple command, and have it update itself every time you start a coding session.

🆔 @hackernewsgithubprojects

📦 antirez/h3.c

H3.C: Generate Videos on Mac

H3.C turns your Apple Silicon Mac into a personal video generator. You type a simple description, like a red fox walking in the snow, and it uses the native Metal graphics engine to synthesize a short, realistic clip right on your laptop. It supports advanced controls like setting the first and last frames to anchor a story, or attaching a reference image to guide the style. It runs entirely offline using the MiniMax H3 model, meaning you can create media without waiting for slow cloud servers. It is a fascinating look at how far local machine learning inference has come for everyday users.

🆔 @hackernewsgithubprojects

📦 prathamlearnstocode/paper2code

Turn Research Papers Into Code Automatically

Turn any research paper into working code without guessing the missing details. Paper2Code takes an arxiv link and builds a clean Python implementation where every line of code references the exact section or equation from the original text. It listens closely to what the authors actually wrote and loudly flags anything they left out. If a hyperparameter is missing, the tool marks it clearly instead of inventing a number that might be wrong. You get a structured project with a training loop, config files, and a notebook that walks you through the logic.

🆔 @hackernewsgithubprojects

📦 ranteck/graph-engineer

Graph-Engineer: AI Code Review Loop

Graph-engineer is a Claude Code skill that turns AI coding into a self-correcting loop instead of a single guess. It makes Claude the boss who sets goals and checks the work, while Codex writes the code and then brutally reviews its own mistakes before fixing them. This adversarial back-and-forth catches hidden bugs that standard coding tools often miss. It even includes a special high-assurance mode that uses three different review angles for critical tasks. The project is still in design, but it offers a fascinating look at how AI agents can collaborate to produce much higher quality software.

🆔 @hackernewsgithubprojects

📦 ethanfel/comfyui-minimaxh3-contex-loop

MiniMax H3 Contex Loop: Clip Chaining in ComfyUI

ComfyUI MiniMax H3 Contex Loop is the scene-by-scene video generator that keeps motion and audio flowing across cuts instead of resetting. It lets you plan multiple shots in one visual editor, then stitches them together using a single recursive process that saves checkpoints after each take. You can review every scene with synced sound, retry failed prompts, or approve good ones to build the final video. This means you can extend existing clips or film new scenes from different angles without losing continuity. It turns disjointed video generation into a manageable, reviewable production loop where every clip naturally leads into the next.

🆔 @hackernewsgithubprojects

📦 superclaude-org/superclaude_framework

Supercharge Your Claude Code Workflow

Are you looking to turn your coding workflow into a powerhouse of productivity? Meet the SuperClaude Framework, a meta-programming toolkit that transforms Claude Code into a structured development platform. By injecting specialized cognitive personas and behavioral instructions, it adds thirty powerful slash commands to your environment, covering everything from deep research to complex code implementation. Whether you need to brainstorm project architecture or manage tasks efficiently, this framework provides the systematic automation you need to build faster. Boost your performance and streamline your daily development cycles with this essential configuration tool today for a more intelligent coding experience.

🆔 @hackernewsgithubprojects

📦 avengemedia/dankmaterialshell

DankMaterialShell

DankMaterialShell is the unified desktop shell that finally eliminates the need to stitch together a dozen different tools to build a custom Wayland environment. Instead of managing a fragile mix of separate status bars, notification daemons, and lock screens, this all-in-one desktop shell replaces them with a single, highly integrated system built on Go and Quickshell. It handles everything from system monitoring and spotlight-style application searching to automated, wallpaper-based color schemes that instantly theme your entire desktop. It is a beautifully cohesive way to run modern window managers like Hyprland or Sway without the configuration headache.

📰 https://news.ycombinator.com/item?id=48718679

🆔 @hackernewsgithubprojects

📦 facebookresearch/sam-3d-objects

Turn Any Image into 3D with Meta's SAM 3D Objects

Ever wanted to turn a single photo into a fully detailed 3D model? The SAM 3D Objects repository from Meta makes this possible by reconstructing high-fidelity 3D shape, texture, and layout from just one 2D image. It works exceptionally well in cluttered real-world scenes, even when objects are partially hidden. By leveraging advanced progressive training and human-in-the-loop feedback, this foundation model lets you generate posed 3D assets that are ready for manipulation. You can easily start by running the provided demo scripts to transform images into 3D outputs like Gaussian splats, bringing your photos into the third dimension today.

🆔 @hackernewsgithubprojects

📦 neuphonic/neutts

neutts

neutts is the open-source speech model library that finally brings highly realistic voice generation and instant cloning directly to your local hardware. By moving advanced voice synthesis away from slow cloud APIs and onto small language model backbones, it lets you generate natural, human-sounding speech right on your phone, laptop, or Raspberry Pi. Its standout feature is instant voice cloning, allowing you to replicate a speaker's voice using as little as three seconds of clean audio. It is a fantastic tool for anyone wanting to build private, real-time voice assistants that run entirely on-device. Try running a local model today to hear it yourself.

🆔 @hackernewsgithubprojects