📦 shift-labs-ai/pi-peer

Pi-Peer: Let AI Sessions Talk to Each Other

Pi-peer is the messaging layer that lets your AI coding sessions find and chat with each other. Instead of copy-pasting notes between terminals, you simply tell one session to message another, and it delivers plain text right into the other’s workflow. It works locally on your machine and scales across your team using a shared server, so you can alert a colleague that a branch moved or ask if anyone is mid-task. The design is intentionally simple: it sends only text, keeps a persistent mailbox for offline messages, and strictly prevents one session from controlling another. It turns isolated AI agents into a cooperative team without adding complexity.

🆔 @hackernewsgithubprojects

📦 mateusdcc/pi-gpt-search

Pi-Gpt-Search: Zero Cost Web Search for Pi

Pi Gpt Search lets any coding model grab fresh web results without burning a single token of your model’s quota. Instead of making the AI guess or wait, it quietly taps into OpenAI’s own search backend to pull live snippets and links directly. Your main model stays in charge of thinking, while this tool handles the fetching in the background. It even supports deep multi-step research that follows links and checks page content automatically. You get current facts without paying extra for inference or reading stale data. It works with any model you choose and keeps your privacy intact by only sending the search query.

📰 https://news.ycombinator.com/item?id=49255837

🆔 @hackernewsgithubprojects

📦 jiashunwang/hybrid-motion-imitation

hybrid-motion-imitation

hybrid-motion-imitation lets robots handle both walking and climbing by mixing two distinct movement styles into one smooth system. Instead of learning each action separately it trains a single agent that switches between locomotion and whole-body tasks based on terrain. You can use real motion capture data to teach the robot how to climb boxes or grab objects then watch it adapt in simulation or on actual hardware. It builds on proven research but makes it easier to run with standard simulators. This approach means your robot does not just shuffle around but actually interacts with the world around it making complex tasks feel natural.

🆔 @hackernewsgithubprojects

📦 schildep/verified-3d-mesh-intersection

Verified 3D Mesh Intersection

Compute 3D mesh intersections without trusting AI bugs, by reading just 93 lines of formal math specs in Lean. This repository proves a mesh intersection algorithm correct with machine-checked logic, letting you skip reviewing the thousand lines of generated implementation code entirely. You get guaranteed watertight results where the output solid perfectly matches the mathematical intersection of the inputs, handling every weird edge case automatically. It trades raw speed for absolute correctness, taking twenty seconds on heavy models instead of milliseconds, but that slowness is the price for zero trust assumptions.

📰 https://news.ycombinator.com/item?id=49083239

🆔 @hackernewsgithubprojects

📦 bgreenwell/xleak

View Excel Files in Your Terminal

Open an Excel spreadsheet right inside your terminal without needing Microsoft Excel or a web browser. This project, called Xleak, turns your command line into a fast, interactive spreadsheet viewer. The real magic happens with the interactive mode, where you can hunt through massive files using a full-text search that scans every single cell. You can also peek at the actual Excel formulas behind the numbers and copy data directly to your clipboard, even over SSH. It handles huge files by only loading what you see, so it stays snappy.

🆔 @hackernewsgithubprojects

📦 kylin010/tcpfit

Tcpfit: The Machine-Specific TCP Tuner That Finds Hidden Bottlenecks

Tcpfit is the Linux tuning tool that finally stops guessing and starts measuring by calculating custom network parameters for your specific hardware instead of applying generic defaults. It solves the problem of suboptimal internet speeds by running a smart sweep that identifies exactly where your machine’s limits hit, automatically adjusting thirty-two system settings and buffer sizes to match real-world conditions rather than theoretical ideals. This one vivid approach means you get optimized throughput tailored to your exact bandwidth and bottleneck, ensuring your server performs at its peak without manual tweaking.

🆔 @hackernewsgithubprojects

📦 aphid91/fluoddity

Fluoddity: Evolving Digital Lava Lamps

Watch thousands of tiny particles wander across your screen, leaving behind glowing trails that act like a digital nervous system. Fluoddity lets you play god with these blobs by tweaking sliders that change how they sense their environment. The real magic happens because each particle has its own tiny, randomized brain made of math rules. When you click on a particle you love, you can clone it, mutate its brain slightly, and watch the new generation evolve into something completely different. It feels like watching an interactive lava lamp that actually thinks, creating swirling rivers, branching lightning, or strange microscopic life forms as you guide their evolution.

🆔 @hackernewsgithubprojects

📦 watzon/pindrop

Pindrop: Offline AI Dictation for Mac

Pindrop lets you speak and instantly see your words typed out on your Mac, all without sending a single byte of audio to the cloud. It runs completely offline using local speech-to-text models, so your conversations stay private and your microphone stays silent to the rest of the internet. The app sits quietly in your menu bar, waiting for you to hit a shortcut, and it even pauses your music so your dictation stays clean. You can pick different local engines based on your Mac’s speed, and every transcript lands right at your cursor or copies to your clipboard.

🆔 @hackernewsgithubprojects

📦 iannuttall/ralph

Ralph: The AI That Codes Itself

Ralph turns your computer into an autonomous coding assistant that builds projects by itself. It starts with a simple text idea, writes a technical plan, and then enters a continuous loop of writing code, testing it, and committing changes to git. What makes Ralph fascinating is that it treats your project files as its memory instead of relying on a temporary chat window. It remembers exactly what it did last time, fixes its own mistakes, and keeps working until the feature is finished. You can even swap in different AI models to run the tasks.

🆔 @hackernewsgithubprojects

📦 shape-machine/tusk-macos

Tusk: The Minimal Native macOS Database Tool

Tusk is a lightweight, native database manager built specifically for macOS, designed to let developers interact with PostgreSQL without the bloat of web-based wrappers. Instead of relying on heavy Electron frameworks, the project uses SwiftUI to create a fast, responsive interface that feels like it belongs on the system. It handles the full workflow from managing secure connections to exploring complex table structures and running live queries. You can browse your database schema visually, inspect relationships with interactive graphs, and execute SQL commands with instant syntax highlighting. The tool keeps things simple by storing sensitive credentials safely in your local Keychain and offering a clean, distraction-free workspace.

📰 https://news.ycombinator.com/item?id=47515590

🆔 @hackernewsgithubprojects

📦 crazygirl437/polymarket-5min-bot

Polymarket 5-Min Arbitrage Bot

This Rust-based bot automatically hunts for pricing errors in Polymarket’s five-minute crypto prediction markets. It monitors the combined cost of buying both "yes" and "no" tokens and triggers a trade whenever that total dips below one dollar. Once both tokens are purchased, the code automatically merges them back into stablecoins, capturing a small but consistent profit from the pricing gap. It also handles risk management by balancing positions and cleaning up trades right before the timer runs out, offering a complete automated loop for short-term market inefficiencies.

🆔 @hackernewsgithubprojects

📦 botcoder254/limboo

Limboo: The OS for AI Coding Agents

Watch Limboo give your coding agent a real workspace where it can do actual engineering. Instead of just chatting in a blank window, the app manages your repositories, watches files, runs terminals, and holds project memory so the AI can focus entirely on writing code. The coolest part is that it lets multiple agents like Cursor and Claude work together inside one session, keeping everything organized in durable workspaces with full git control. It is a local-first desktop tool that turns a simple chat into a complete development environment, giving developers the structure they need to build complex software without losing track of their progress.

🆔 @hackernewsgithubprojects

📦 perplexityai/perplexity-cli

Perplexity CLI

Run web searches directly from your terminal and get clean JSON results that code can actually use. This tool lets you query the internet without opening a browser, returning snippets and URLs in a format that coding agents or your own scripts can easily parse. It also extracts relevant text from multiple web pages in one go, giving you exactly the info you need from a bunch of links instantly. It works on macOS and Linux, installs with a single line, and plays nice with your existing command line workflow. It is perfect if you want to automate research or feed search results into AI tools without the clutter.

🆔 @hackernewsgithubprojects

📦 ryanfitzpatrickio/threejs-playground

Threejs Playground: Procedural City & Parkour

Threejs Playground is an experimental browser 3D playground that lets you explore an endless procedurally generated city with parkour and vehicle mechanics. Built using Vite, Solid, Three.js r185 WebGPU, and Rapier physics, it serves as a prototype workspace rather than a finished game. The project shines in its ability to stream infinite city blocks with seeded randomness while keeping collision data perfectly synchronized with rendering. You can experience freerun traversal, grapple swinging, and wingsuit flying through dynamically generated streets and rooftops. It also features enterable interiors using Wave Function Collapse for layout generation, along with wet road materials and volumetric sky effects.

🆔 @hackernewsgithubprojects

📦 arrow-air/project-caribou

Project Caribou Drone Design

Project Caribou is the hexacopter project that finally treats every motor arm as its own independent power station. Instead of wiring one giant battery to six motors like a traditional drone, this design gives each of the six arms its own dedicated smart battery pack and custom safety board. The clever part is the hardware kill switch: a physical cutoff that cuts power at the battery stage before it even touches the electronics, ensuring the flight computer can never accidentally cause a short circuit. This repository collects all the engineering blueprints, custom circuit board files, and 3D models that make this safe, scalable drone architecture possible.

🆔 @hackernewsgithubprojects

📦 sdk445/hmon

Hmon: Real-Time Linux Monitoring

Hmon is a fast, lightweight Linux system monitor built in C and ncurses. It provides real-time insights into your computer's health, tracking everything from CPU and GPU usage to memory, disk, and network activity. What makes it particularly interesting is its plugin architecture, which allows you to extend its capabilities by adding new collectors for specific metrics like Docker containers, database connections, or even cron jobs. The tool offers a clean, terminal-based interface with three distinct viewing modes: a full dashboard, a condensed Zen mode, and a dense Pro mode for detailed kernel stats.

🆔 @hackernewsgithubprojects

📦 telivityai/careerops

CareerOps: Your Personal Career Operating System

CareerOps replaces your scattered job search tools with one open-source app that remembers exactly what you do. You capture daily wins into a secure memory bank, then promote those real accomplishments into tailored resumes and interview prep without AI inventing metrics. It tracks applications, compares offers, and keeps your career data local so you own it completely. Instead of starting from a blank page every time you apply, you build a durable history of your work that actually makes the next job search easier. It is built for engineers and anyone who wants a serious record of their career.

🆔 @hackernewsgithubprojects

📦 nvidia/xr-ai

XR AI: NVIDIA's Open-Source Agent Stack for Mixed Reality

XR AI is the NVIDIA toolkit that lets developers build AI agents capable of seeing, hearing, and interacting with real-time virtual environments. Instead of just chatting, these agents can process live video, transcribe your voice, and control 3D objects using natural language commands. It connects your favorite headsets and web browsers to powerful local or cloud models, giving you a practical way to prototype spatial computing experiences. Whether you are building simple voice assistants or complex mixed reality apps, this stack handles the heavy lifting of audio, video, and reasoning so you can focus on the creative part of the experience.

🆔 @hackernewsgithubprojects

📦 infatoshi/craftax.cu

Craftax.cu: The 7x Faster Gaming AI Trainer

Craftax cu is the ultra-fast reinforcement learning toolkit that runs complex game simulations seven times faster than standard methods. It takes a classic puzzle-adventure game and moves every calculation onto the graphics card, allowing thousands of parallel worlds to exist and evolve simultaneously without slowing down the computer. Instead of training slowly on a traditional processor, this tool uses specialized graphics hardware to simulate millions of game steps in seconds, letting artificial intelligence learn strategies at incredible speed. It keeps every result perfectly accurate by double-checking its own work, ensuring that the fast training does not sacrifice quality.

🆔 @hackernewsgithubprojects

📦 hebing-sjtu/surveyor

Surveyor: Map a Research Field in Minutes

Surveyor is a research library that builds a smart map of any academic field from survey papers. Instead of drowning in endless PDFs, it treats a review article as a seed to automatically extract the entire field’s structure and core reading list. You simply paste a survey link, and the tool reads the paper’s own sections to build a taxonomy and ranks the most important references by how often they are discussed. Import a few surveys, and it calculates the true core literature that all experts agree on, letting you ask natural questions in Chinese or English.

🆔 @hackernewsgithubprojects