📦 xhluca/agent-talk

agent-talk

Your coding agents are officially done with isolation. Instead of you acting as a manual messenger copying and pasting code snippets between terminals, agent-talk lets autonomous coding agents directly message and coordinate with each other. This clever plugin integrates with popular tools like Claude Code to establish a secure, end-to-end encrypted channel over a message relay. It means one agent can autonomously ask another about API changes, resolve database schema conflicts, or hand off tasks on different machines without you typing a single word of the conversation. It is a brilliant way to let independent agents handle the low-level details while you focus on the big picture.

📰 https://news.ycombinator.com/item?id=48936534

🆔 @hackernewsgithubprojects

📦 fredconex/arandu

Arandu

Arandu is the desktop launcher that finally takes the headache out of running local AI models on your computer. Instead of wrestling with command line arguments, terminal prompts, or dll libraries, you get a clean, visual command center to manage and run your llama-server. You can download files straight from Hugging Face, track your system hardware performance in real time, and launch your models with custom arguments in a couple of clicks. It even includes a built-in chat interface so you can start talking to your models instantly. It is the easiest way to control your local AI setup.

🆔 @hackernewsgithubprojects

📦 facebookresearch/shaper

ShapeR

Meta Reality Labs has built an AI that turns shaky, casual video clips of a messy room into perfect, individual 3D digital models. Unlike traditional tools that merge an entire room into one unchangeable block, ShapeR processes every object separately. It looks at sparse tracking points, video frames, and even text descriptions to build a metrically accurate 3D mesh for each item. This means you can actually pick up, move, and interact with the virtual objects later. Best of all, it works without any human input. Grab the code on GitHub and start reconstructing your physical world in 3D.

🆔 @hackernewsgithubprojects

📦 danilaa1/cuelume

Cuelume

Cuelume is the lightweight library that finally makes adding satisfying UI sounds to your web projects incredibly simple. Instead of forcing you to hunt down, license, and load heavy audio files, this clever tool synthesizes fourteen beautiful, modern interaction sounds live in the browser using the Web Audio API. All you have to do is drop a simple data attribute onto your HTML buttons, links, or toggles, call a single startup function in your code, and you instantly get gorgeous physical clicks, soft bells, and tactile feedback. It has zero dependencies, works with any framework, and runs silently on the server. Grab Cuelume today and give your user interfaces a...

🆔 @hackernewsgithubprojects

📦 haileyydev/maildrop

Maildrop

You can now run your own private burner email server in just a single Docker command. A clever self-hosted service called maildrop lets you deploy a lightweight email system on your own domain to generate random, disposable addresses on the fly. It is perfect for dodging spam when signing up for sketchy websites, and it even supports password-protected inboxes and optional outgoing mail via an SMTP relay. The Python backend processes incoming messages and displays them in a clean, minimal web dashboard, keeping your real inbox clean. Grab your domain, spin up the container, and take back control of your online privacy today.

🆔 @hackernewsgithubprojects

📦 piat0n/baremetal-ram-dumper

BareMetal RAM Dumper

Baremetal-ram-dumper is the bare-metal utility that makes physical memory extraction incredibly straightforward by bypassing the operating system entirely. Operating right from a bootable USB drive, this tool is built for cold boot attack experiments, where researchers freeze a computer's memory chips to slow down data decay, reboot the machine, and instantly copy the raw RAM contents straight to the drive. It utilizes classic BIOS routines to map out system memory and safely write those gigabytes of data directly to disk. It is a fascinatingly simple way to inspect what was left in memory before it fades away.

📰 https://news.ycombinator.com/item?id=48787201

🆔 @hackernewsgithubprojects

📦 axiommath/imo2026

imo2026

Explore the actual math proofs that just ace'd the world's toughest high school math competition. The imo2026 repository gives you the exact, machine-verified Lean 4 code for all six problems solved by an AI prover. What is wild is that this system managed a perfect forty-two out of forty-two score. You can browse the formal problem statements alongside the exact verified proofs, like the massive four-thousand-line solution that took over fourteen hours to calculate. It is basically a front-row seat to the future of automated reasoning. Check it out and run the verification scripts yourself to see the future of mathematics in action.

📰 https://news.ycombinator.com/item?id=48945327

🆔 @hackernewsgithubprojects

📦 yanqinjiang/data-engine-omnix

Data Engine OmniX

The data-engine-omnix project automatically builds complex dynamic scenes inside Unreal Engine 5 to generate synthetic training data for advanced spatial AI models. By combining static environments with moving objects, this pipeline programmatically sets up synchronized multi-camera rigs and runs headless rendering on Linux GPU servers. It tracks precise vertex trajectories, face topologies, and depth maps, exporting perfect ground-truth data that is nearly impossible to capture in the real world. Developers can easily generate rich synthetic datasets of complex motion, which helps train algorithms to reconstruct moving objects from any viewpoint and at any point in time.

🆔 @hackernewsgithubprojects

📦 cjbuilds/codex-orchestration

Codex Orchestration

Coordinate multiple artificial intelligence models to work together on a single task by assigning them specialized roles like planning, reviewing, and executing. codex-orchestration is a security-first plugin for Codex that lets you delegate work across different models to build a stronger plan before any code changes actually begin. Instead of relying on one general-purpose model, you can use specialized models for each phase, such as letting one model draft the initial plan, having another independent model review and approve it, and letting a third model write the code. This setup can run tasks up to twice as fast and saves your premium model limits by offloading repetitive tasks. You can...

🆔 @hackernewsgithubprojects

📦 liangdabiao/claude-code-deep-research-main

Claude Code Deep Research Agent

Conduct deep, citation-backed research completely on autopilot by transforming your terminal into an advanced multi-agent research team. Using this framework with Claude Code, you can deploy multiple specialized agents in parallel to scour the web, cross-reference facts, and score source quality on an A-to-E scale. It uses a Graph of Thoughts reasoning structure to automatically map out, prune, and merge research paths into a comprehensive final report. Just type a single command to watch it refine your question, build a plan, and write an incredibly thorough, validated report. It is the ultimate way to automate complex research workflows instantly.

🆔 @hackernewsgithubprojects

📦 jungledesh/profile

Profile

Identify and fix hidden bottlenecks in your AI inference servers by measuring live performance directly against the physical limits of your hardware. Profile connects to your running server, calculates the absolute maximum throughput your GPU and model combination can achieve, and tracks actual performance under load. Instead of just showing charts, it diagnoses issues like memory pressure or queue delays and tells you exactly which startup flags to change. Once you apply the changes, it measures the difference to show your real-world cost savings and throughput gains. Optimize your setup to push your hardware to its absolute limit and slash hosting costs.

📰 https://news.ycombinator.com/item?id=48594746

🆔 @hackernewsgithubprojects

📦 saaranshm/unsnooze

unsnooze

unsnooze is the terminal utility that automatically wakes up and resumes your limit-stopped AI coding sessions the moment their usage resets. When tools like Claude Code or Codex CLI freeze because you hit a five-hour or weekly cap, this clever tool tracks those paused processes across your workspace and safely wakes them back up inside your multiplexer. It operates as a local scheduler rather than a blind auto-approver, carefully navigating interactive limit menus and validating your session identity before typing a single keystroke. It even monitors desktop and editor extensions, giving you a live dashboard with calibrated account-wide burn rates so you can see exactly when you will hit the...

🆔 @hackernewsgithubprojects

📦 ea1fuo/antennasim

antennasim

You can now design and simulate custom antennas directly in your web browser without installing heavy desktop software. antennasim is an open-source, interactive web application that brings the industry-standard NEC2 simulation engine straight to a web page or even your phone. It translates complex electromagnetic math into interactive, real-time three-dimensional visualizations of radiation patterns, wire currents, and impedance charts. You can pick from seventeen built-in antenna templates, drag wires to customize shapes in the editor, and run multi-band frequency sweeps instantly. It runs entirely serverless in the browser using WebAssembly or via a simple self-hosted container. It makes advanced antenna physics accessible and visual for any curious developer or radio...

🆔 @hackernewsgithubprojects

📦 nevereverinsomnia/fsdc-detr

FSDC-DETR

Spot tiny, hard-to-see objects in drone footage and satellite images with incredible accuracy by changing how computer vision models process details. Traditional AI detectors often lose track of minuscule targets because they blur out fine textures during image processing. This project, called fsdc-detr, fixes this by looking at images through both spatial patterns and frequency waves, preserving those sharp, high-frequency details that make small objects stand out. It tackles the classic problem of tiny things disappearing during data processing, which is incredibly useful for analyzing complex aerial maps. Check out this clever approach to make sure nothing gets lost in the background.

🆔 @hackernewsgithubprojects

📦 ifindev/fullstack-next-cloudflare

fullstack-next-cloudflare

You can host a production-ready web application with a global database, file storage, and edge AI for basically zero dollars a month. That is the magic behind fullstack-next-cloudflare, a brilliant boilerplate that combines Next.js 15 with Cloudflare’s edge infrastructure. Instead of dealing with slow, expensive servers, this template runs your app right next to your users using serverless workers. It comes pre-configured with a distributed SQL database, fast object storage, and secure Google login out of the box. It even includes a built-in AI summarizer service running directly on Cloudflare’s free AI network. Grab this template, deploy it in minutes, and launch your next big idea on a global scale...

🆔 @hackernewsgithubprojects

📦 adventdevinc/kudu

kudu

Kudu is the open-source system maintenance tool that finally brings comprehensive device cleaning and security scanning to Windows, Mac, and Linux under a single, unified interface. Instead of running separate programs for system clutter, duplicate files, startup management, and malware threat detection, this app packs everything into one lightweight dashboard. It even includes specialized features like game mode optimization, file shredding, and vulnerability tracking to keep your computer running smoothly and securely. It is the perfect, straightforward utility for anyone looking to reclaim disk space and guard their privacy without dealing with bloated commercial software. Give your operating system a fresh start today.

🆔 @hackernewsgithubprojects

📦 pleaseprompto/google-ai-mode-skill

google-ai-mode-skill

The google-ai-mode-skill is the Claude Code add-on that finally stops you from burning thousands of tokens on basic web searches. Instead of making Claude manually click, read, and analyze ten different website links, this clever tool launches a silent local browser to let Google's AI Mode do the heavy lifting for you. It pulls back a single, pre-synthesized answer complete with inline citations, keeping your context window clean and your API costs incredibly low. It is entirely self-contained, runs locally without needing any API keys, and even saves your browser profile to bypass annoying captchas. Grab this skill to give your terminal agent superpowered research skills for free.

🆔 @hackernewsgithubprojects

📦 dominostars/playtranslate

Playtranslate

Translate foreign language games directly on your Android screen in real time while you play. This clever app acts as a smart, live overlay that reads foreign text right off your display, translating dialouge and menu options instantly. It supports offline engines like Yomitan and PaddleOCR, meaning you do not even need an active internet connection to understand what is happening. Best of all, it integrates directly with Anki, allowing you to instantly save new words to study cards with a single tap. Grab this app to make imported games your personal language classroom.

🆔 @hackernewsgithubprojects

📦 optim-agent/optim-agent

Optim Agent

The parameter-tuning assistant that actually understands what your code is trying to achieve. Instead of treating your variables like anonymous math coordinates, optim-agent lets coding agents like Claude or Codex read your code, understand the real-world context of your parameters, and propose smart configurations. It bridges the gap between what a parameter means and how it performs, optimizing everything from machine learning models to decision rules without wasting expensive evaluations. If you want a smarter way to tune your systems, check out this repository to let AI do the heavy lifting for you.

🆔 @hackernewsgithubprojects

📦 px0-ai/px0

px0

Update and tweak your AI prompts in production instantly without redeploying your application code. This project acts as a central control hub for your AI prompts, agent skills, and tool schemas. Instead of hardcoding prompts inside your software, you register them in a central place using versioned templates that connect to major providers like OpenAI and Anthropic. When a prompt needs a quick adjustment, you update it in the registry, and your live application gets the update immediately with visual history tracking and full rollback safety. It completely decouples your AI logic from your software deployment cycle, saving you from constant rebuilds.

🆔 @hackernewsgithubprojects