📦 sqliteai/warp

Warp runs massive AI models on normal computers

Warp is the inference engine that runs massive AI models on normal computers by streaming data directly from your SSD. Most people think you need thousands of dollars of expensive graphics cards to run huge language models, but Warp changes that by keeping only the essential parts of the model in your RAM while reading the rest from your fast storage. It uses a smart trick called a mixture of experts where it only loads the tiny fraction of code needed for each specific question instead of loading everything at once.

🆔 @hackernewsgithubprojects

📦 ombharatiya/ai-system-design-guide

AI System Design Guide

Design production-grade artificial intelligence networks and ace senior engineering interviews using a massive, open-source knowledge base. The ai-system-design-guide repository acts as a living, practical reference that solves the problem of outdated textbooks by continuously updating its material for modern architectures. It bypasses abstract theory to focus on real-world engineering challenges, like building multi-tenant security layers, optimizing token economics, and deploying self-driving agent loops. The single most impressive resource inside is a complete collection of real-world case studies with full architectural diagrams, showing you exactly how to handle complex issues like real-time search, voice latency, and autonomous coding. It is the ultimate roadmap to mastering modern AI engineering.

🆔 @hackernewsgithubprojects

📦 productdevbook/port-killer

Port Killer: Free Up Ports Instantly

Port Killer lets developers instantly find and kill stubborn processes hogging network ports on macOS and Windows. Instead of guessing which background app is blocking your work, the tool automatically scans your system to show exactly what is listening. You can then stop that specific process with a single click, clearing the way for your own server to start. It even handles complex tasks like managing local Kubernetes connections and Cloudflare tunnels. This saves hours of tedious command line hunting by turning a confusing technical problem into a simple, one-click solution that keeps your development environment running smoothly.

🆔 @hackernewsgithubprojects

📦 yohey-w/multi-agent-shogun

Multi Agent Shogun

An open source project called multi-agent-shogun coordinates teams of local AI agents using a traditional Japanese military hierarchy, running them simultaneously inside a terminal multiplexer to tackle complex software projects. By separating responsibilities into clear roles, the Shogun acts as the user-facing general, the Karo plans the operation and splits tasks, and the Ashigaru complete the actual coding. This structure prevents parallel agents from overwriting each other's work or wasting API credits, using an event-driven file inbox system for flawless communication. It is a brilliant way to watch a coordinated AI army systematically refactor codebases and run unit tests in parallel.

🆔 @hackernewsgithubprojects

📦 yeachan-heo/gajae-code

Gajae Code

Build and run autonomous coding agents that can search, edit, and patch codebases entirely on their own. This repository provides a complete MVP of an agentic coding assistant, combining a TypeScript orchestrator with native Rust-powered components for speed. It features a rich set of pre-configured tools for terminal access, file editing, AST-grep searching, and even a terminal-based user interface. By executing targeted actions through a well-defined RPC architecture and custom plugin system, it lets developers automate repetitive development tasks safely. It is an incredibly detailed blueprint for anyone building the next generation of AI-driven developer tools.

🆔 @hackernewsgithubprojects

📦 robbyant/lingbot-depth

LingBot-Depth

Transform sketchy, incomplete, and noisy depth sensor data into razor-sharp, metric-accurate 3D measurements instantly. The lingbot-depth repository solves a massive headache in robotics and computer vision by taking raw, glitchy depth sensor feeds and filling in the gaps with incredible precision. It aligns standard camera images with depth geometry in a unified space, creating a highly accurate spatial understanding model trained on over three million real and simulated samples. Whether a robot is trying to grasp a delicate glass cup or track dynamic movements in 3D space, this project makes it possible. Check it out and give your robot eyes.

🆔 @hackernewsgithubprojects

📦 asdfghj1237890/webvideo2nas

WebVideo2NAS

Send streaming videos directly from your web browser to your home server with a single click. This project links a clever Chrome extension to a self-hosted backend, making it incredibly easy to save streaming media. Instead of downloading giant video files to your computer and manually uploading them to your network storage, this tool handles the heavy lifting in two smart ways. It can either grab the video link and download it directly on your server, or safely fetch and stitch the video segments right inside your browser session to bypass strict website blocks. It is the perfect, hassle-free pipeline for building your personal media library.

🆔 @hackernewsgithubprojects

📦 nari-labs/dia2

Real-Time Conversational AI with Dia2

Ever wanted a text-to-speech model that actually feels like a live conversation? Meet Dia2, an innovative streaming dialogue model that breaks the mold by generating audio instantly as you type. Unlike standard systems that wait for a full script, Dia2 starts speaking immediately, allowing for fluid, natural interactions. By conditioning the output on audio prefixes, you can create realistic, context-aware speech perfect for your own speech-to-speech projects. Whether you are using the command line or the provided interface, Dia2 brings high-fidelity, responsive voice technology directly to your fingertips. Dive in and start building more human-like experiences today.

🆔 @hackernewsgithubprojects

📦 paper-design/paper-mono

Paper Mono

Paper Mono is a clean, modern monospace typeface designed by Paper for both coding and creative design. The repository makes it easy to grab the final font files directly from the releases page, so you can install them on your machine without any hassle. If you are curious about how the fonts were built, the project handles all the heavy lifting using automated quality checks and testing pipelines. You can see the results of those tests in plain HTML reports that show how the letters look at different sizes.

🆔 @hackernewsgithubprojects

📦 zjunlp/skillnet

SkillNet: Teach AI to Perform Real Tasks

SkillNet turns chaotic AI instructions into reusable, modular skills. Instead of asking a chatbot to do everything from scratch every time, this project lets you build specific capabilities like finding objects, cleaning up rooms, or even navigating online stores. It organizes these tasks into clear, reusable blocks that an AI can combine to solve complex problems. Think of it like teaching a robot not just to wave, but to cook dinner by teaching it separate steps like chopping, heating, and plating. The repository provides the actual code and structure to create, test, and connect these skills across different environments.

🆔 @hackernewsgithubprojects

📦 hcavarsan/pipedash

Pipedash

Monitor all your software builds in one beautiful, unified screen instead of jumping between ten different browser tabs. If your team runs some projects on GitHub, others on GitLab or Buildkite, and maybe even has a legacy Jenkins server hiding in the closet, keeping track of everything is a headache. This clever tool pulls data from all your different pipeline providers and displays them together in real time. You can view build histories, cancel running tasks, and even trigger new workflows directly from the app. It runs completely locally with top-tier encryption, so your API keys stay safe. It is the ultimate command center for keeping your sanity during deployments.

🆔 @hackernewsgithubprojects

📦 unclecheng-li/vulnclaw

Vulnclaw

Automate security testing from start to finish by simply telling an AI what target to analyze. Vulnclaw coordinates smart agents, protocol tools, and deep security playbooks to automatically handle reconnaissance, find weaknesses, execute safe proofs, and write up structured reports. Think of it as giving a highly experienced digital companion a list of tasks, which it runs through step-by-step while tracking its history so you can pause, resume, or undo changes at any point. It is an incredibly clever way to see how modern language models can safely navigate complex, multi-step diagnostic workflows without getting lost in endless loops.

🆔 @hackernewsgithubprojects

📦 jianshuo/ccglass

Watch AI Coding Agents Think Live

ccglass lets you peek inside opaque coding agents like Claude Code or Codex by running a lightweight local proxy that captures every request your AI sends to the model. Instead of guessing what prompt was sent or why a tool failed, you get a live web dashboard showing the exact system instructions, raw message history, and real-time token costs. It even highlights exactly which context blocks were added between turns so you can spot wasted API calls instantly. The tool handles complex setups by intercepting plain HTTP traffic, so you do not need to fiddle with network certificates or deal with broken proxy patches.

🆔 @hackernewsgithubprojects

📦 pantheon-security/medusa

Medusa AI Security Scanner

Vetting remote code repositories before you clone them is the best way to keep your system safe, and this project makes it incredibly simple. It lets you scan any GitHub repository for malicious AI supply chain attacks with a single command before bringing it onto your machine. It targets a massive modern security headache, specifically sniffing out weaponized AI editor configs, poisoned hooks, and compromised permissions hidden in settings files for tools like Claude Code and Cursor. It even scans your local AI chat and shell histories to find and safely redact leaked API keys. If you want to safely navigate the wild west of modern developer tools, this is...

🆔 @hackernewsgithubprojects

📦 sv-number/mcp-server

MCP Server

An AI agent can rent a real phone number in over two hundred countries just to catch a one-time verification code. The server handles the whole chore: it finds a working number for a specific website, waits for the text message to arrive, extracts the code, and hands it back. It even checks how many other users are online to guess reliability before you commit. You do not need a developer to manage these temporary lines, the tool does all the heavy lifting. This makes account creation for AI workflows actually possible without dealing with real personal numbers.

🆔 @hackernewsgithubprojects

📦 repoprompt/repoprompt-ce

RepoPrompt CE

repoprompt-ce is the native macOS context engineering app that helps AI coding agents fully understand your codebase before they write any code. Instead of feeding messy, bloated files into LLMs, this tool assembles highly focused, reviewable context from file trees, Git diffs, line slices, and CodeMaps while respecting your token budget. It features an integrated Model Context Protocol server, a debug CLI, and an orchestrator to run agents directly in your workspace. Built in Swift for macOS 26, it is a clean, local-first workspace that lets you inspect, refine, and verify exactly what an AI model sees before it acts.

🆔 @hackernewsgithubprojects

📦 xweiba/location-spoofer

Location Spoofer: iOS Location Research Tool

Location Spoofer is the iOS research tool that simulates Apple location responses for safe app testing. It lets developers pick any spot on a built-in map, then intercepts and swaps the real GPS data with your chosen coordinates using a local proxy or third-party client. This lets you test how location-dependent apps behave anywhere without actually moving, all while keeping your personal data private. It is designed strictly for developers and researchers, offering clear diagnostics and dual coordinate support to help you verify app reliability. It is a smart, secure way to debug location features without compromising your privacy or security.

🆔 @hackernewsgithubprojects

📦 isaac-mason/crashcat

Crashcat Physics Engine

Crashcat is a JavaScript physics engine designed to bring realistic rigid-body dynamics to games and simulations without bogging down performance. It handles complex collisions between shapes like boxes, spheres, and custom convex hulls while keeping the code lean enough to tree-shake for smaller bundles. The standout feature is its persistent body-pair caching system, which remembers what touched last frame to skip expensive calculations for bodies that haven't moved much. This warm-starting approach dramatically reduces the computational load during dense simulations like stacking boxes or jiggly ragdolls. It also supports continuous collision detection to catch fast projectiles that might otherwise tunnel through walls.

🆔 @hackernewsgithubprojects

📦 evianis/travel-photo-abstraction

Travel Photo Abstraction: Distilling Photographs into Sparse Editorial Abstractions

The Travel Photo Abstraction skill turns ordinary travel snapshots into sparse editorial art by mapping their visual relationships to minimal abstract marks. It treats a photo as a visual system, extracting shapes, counts, and colors to create a nonliteral lower panel while leaving the original image completely untouched above. Instead of generic filters, it deconstructs the scene, distills its essence, and reconstructs the relationships into a quiet composition with generous negative space. This approach preserves the photograph’s unique rhythm and asymmetry, offering a fresh, abstract interpretation that respects the source material without adding invented content.

🆔 @hackernewsgithubprojects

📦 iyadhkhalfallah/clauditor

Clauditor: Stop Burning Through Your API Quota

Stop Claude Code from burning through your quota in twenty minutes. Clauditor monitors your session size and automatically blocks the tool when you are wasting tokens, saving your progress so you can start fresh without losing context. Every turn in a standard session resends your entire conversation history, meaning a long chat burns exponentially more quota than necessary. Clauditor detects when your session gets too bloated and forces a rotation, then injects a structured summary of what you were doing into the new chat so you can pick up exactly where you left off. It also tracks failed commands to prevent you from making the same mistake twice.

🆔 @hackernewsgithubprojects