📦 nhahan/webgpt

WebGPT: Delegate to ChatGPT

Offload heavy coding tasks to your web browser without burning through your premium API credits. WebGPT acts as a bridge, letting your local coding assistant send specific jobs to the free, signed-in ChatGPT interface you already use daily. Instead of manually copying and pasting code, the system automatically opens a private chat, assigns the task, and waits for the result while you keep working. It is a clever workaround that turns your existing web session into a parallel worker, saving your expensive quota for the most complex problems that only your main assistant can handle.

🆔 @hackernewsgithubprojects

📦 mexicat/pdoom-video

pdoom-video

pdoom-video is the generative music video project that renders a high-definition experience entirely through code. Created in collaboration with AI, this engine coordinates typography, 3D scenes, and visual post-processing directly in the browser, matching offline exports perfectly. By analyzing vocal stems and drum beats beforehand, the system precisely anchors animations and lyric layouts to the song's exact timeline. The project's offline renderer drives headless Chrome to capture raw frames and compile them with fluid motion blur. It is a fantastic showcase of code-driven art, giving creators a complete blueprint for deterministic, web-based video production.

🆔 @hackernewsgithubprojects

📦 vercel-labs/issue-graph

Issue Graph: See Your Whole Backlog

Trace the tangled web of linked GitHub issues and pull requests before you break anything. Issue Graph connects every related ticket so you can see exactly what is already fixed, what is still broken, and where the real work sits. It turns a messy backlog into a clear map, showing you which changes clash and which follow-ups are still open. Instead of guessing if your fix covers everything, you get a full picture of the impact before you start coding. It is a smart way to avoid duplicate work and merge with confidence. You stop jumping between tabs and start seeing the whole story.

📰 https://news.ycombinator.com/item?id=49836381

🆔 @hackernewsgithubprojects

📦 devploit/geminihunter

GeminiHunter: Exposed AI Keys

GeminiHunter hunts down Google Gemini API keys left exposed in public websites and Android apps. It scans JavaScript files, source maps, and archived web pages to find these secrets, then actually tests if they still work. The coolest part is how it handles blocked keys. If a key is restricted, the tool tries different request variations to see if it can still access the service. This gives security pros a real picture of risk instead of just a list of strings. It saves you from guessing and helps you patch leaks fast.

🆔 @hackernewsgithubprojects

📦 cordum-io/cordum

Cordum

Enforce strict safety policies and human approvals on autonomous AI agents before they execute risky terminal commands or modify production code. Cordum acts as an active firewall designed specifically for AI workflows, stepping in to halt dangerous tool calls and system operations until a human grants permission. It intercepts agent actions in real time, verifying each request against your custom security rules while keeping an immutable, tamper-proof audit trail of everything the AI attempted to do. It is like having a reliable co-pilot watching your AI's every move, keeping your infrastructure safe while letting automation run with confidence.

🆔 @hackernewsgithubprojects

📦 luongnv89/skills

Skills

Skills is the command line library that finally brings standardized, expert workflows to your favorite AI coding assistants. Instead of relying on random, unpredictable prompts that give you different results every time, this tool lets you inject structured, version-controlled playbooks directly into assistants like Claude Code, Cursor, or Copilot with a single command. It comes with a massive catalog of standalone blueprints for everything from running deep code reviews and modernizing legacy codebases to setting up automated shipping pipelines. You can easily install exactly what you need to make your AI assistant follow rigorous, repeatable steps every single run.

🆔 @hackernewsgithubprojects

📦 nvlabs/kda

kda

kda is the development template that helps coding agents write high performance graphics and computing code. Instead of letting an artificial intelligence model guess its way through complex hardware instructions, this framework establishes a strict feedback loop of research, planning, testing, and profiling. The agent inspects your workspace, drafts a clear step by step plan, and refines the code in small iterations based on real hardware performance data. This structured approach takes the chaos out of programming specialized processors, turning unpredictable models into reliable assistants that write optimized code backed code with proven benchmark results.

🆔 @hackernewsgithubprojects

📦 stacklok/mecatl

mecatl

mecatl is the open-source agent harness that finally lets you run powerful AI workflows safely on your own infrastructure. Instead of sending all your private data to a black-box cloud, this framework lets you run provider-agnostic loops on your own local machine or right inside your Kubernetes clusters. It solves the massive security headache of modern AI systems by giving you durable state, granular permissions, and isolated micro-virtual machine execution environments that you fully control. You can build complex, multi-agent developer teams with built-in guardrails and structured human approval gates, ensuring your automated assistants never run wild with your code.

🆔 @hackernewsgithubprojects

📦 emir/claude-s40

Chat with Claude on a 2007 Nokia

A phone from 2007 can now chat with Claude, complete with web search results displayed on a tiny monochrome screen. The project, known as Claude S40, is a custom app built for old Nokia devices that can't handle modern secure connections. It runs a small server in the cloud to translate the phone's ancient security handshake into something Claude understands. The coolest part? The phone acts as a simple remote, typing your question on a physical keypad while the server handles the heavy lifting. You get full chat history, calendar reminders, and even weather updates, all fitting on a screen the size of a matchbox.

📰 https://news.ycombinator.com/item?id=49868631

🆔 @hackernewsgithubprojects

📦 naiveai-labs/naive-n0.5-flash

naive-n0.5-flash

An open-weight AI model with over three hundred billion parameters can now process a massive one-million-token context window without using any full-attention layers. The naive-n0.5-flash model achieves this feat by blending sliding-window attention with a highly efficient sparse attention mechanism, keeping the memory footprint manageable even when handling massive codebases. Built specifically for coding and AI research, it bypasses the heavy decoding overhead that usually plagues long-context processing. By combining this smart architecture with custom inference systems, developers can run incredibly deep research tasks at lightning speeds. It is a brilliant example of how clever design can make massive AI research tools accessible to the open-source community.

🆔 @hackernewsgithubprojects

📦 ucashier/ucashier-usdt-payment

ucashier-usdt-payment

YouCashier is the stablecoin payment infrastructure that lets businesses accept and send digital currencies directly through their existing websites without rebuilding their backend. Instead of routing customer payments through a platform pool, this system acts as a smart coordinator that watches the blockchain and sends transactions directly to your own self-custodial wallet. It handles the complex work of tracking networks like Tron and Ethereum, matching payments to order numbers, and managing timeouts. You can also use the same dashboard to handle outbound payouts like user withdrawals or supplier settlements. It is a practical way to add crypto payments to any global business.

🆔 @hackernewsgithubprojects

📦 deepfates/imp

Imp: Self-Improving AI for Elixir

Imp is the Elixir framework that finally brings the power of self-improving language model programs to the BEAM virtual machine. Instead of writing messy prompts and fragile parsers, you declare clean, typed functions that specify exactly what your inputs and outputs look like. Imp handles the rest by building the prompts, validating the replies, and letting you optimize the entire pipeline against real examples of what success looks like. It actually runs tests, analyzes where your AI steps failed, and rewrites the underlying instructions to make your application smarter over time.

📰 https://news.ycombinator.com/item?id=49869995

🆔 @hackernewsgithubprojects

📦 0xeb/ghidrasql

ghidrasql

You can now query your Ghidra reverse engineering databases using standard SQL. The ghidrasql tool exposes sixty-seven public tables and seventy-eight database views, allowing you to search functions, cross-references, types, and even decompiled pseudocode with simple database queries. Instead of writing complex scripts, you can pull structured information instantly or write back updates to rename functions and rewrite signatures. It integrates cleanly with AI coding agents and provides an interactive terminal interface or a local HTTP API server. It completely transforms how you navigate and modify binary analysis data, turning a complex reverse engineering database into a clean, searchable spreadsheet.

📰 https://news.ycombinator.com/item?id=49174716

🆔 @hackernewsgithubprojects

📦 sofianeelhor/pkforge

PKForge: The Ultimate Offline Pokémon Save Editor and Bank for Android

PKForge is the Android pocket companion that finally gives you absolute control over your Pokémon save files completely offline. Built for mobile screens and handheld consoles, this clever app lets you load up your saves from almost any generation or emulator and edit stats, moves, or shininess with real-time legality checks. Its best trick is the built-in, cross-generation bank that lets you safely transfer your favorite monsters all the way from Game Boy Advance files up to the Switch. It is like having a private, offline version of Pokémon Home right in your pocket. Give it a spin and build your ultimate dream team.

🆔 @hackernewsgithubprojects

📦 aimixer/comfyui_minimaxh3_director

comfyui_minimaxh3_director

comfyui_minimaxh3_director is the timeline director extension that coordinates complex multi-segment audio and video generation directly inside your workspace. Instead of stitching raw clips together manually, you can manage video splits, transition points, and visual timing in one place while keeping your motions and stereo audio flowing smoothly across shots. It wraps the official model generation pipeline to let you transition from simple text-to-video up to complex reference-guided scene edits, and you can even pin previous clips for seamless shot handoffs. It is the perfect tool for creators who want to direct cohesive, cinematic AI videos without drowning in fragmented node setups.

🆔 @hackernewsgithubprojects

📦 johnsideserf/siggy

Siggy Terminal Signal Client

Siggy is the terminal-based messenger client that brings the privacy of Signal straight to your command line with an old-school IRC aesthetic. If you love staying in the terminal and hate constantly reaching for your phone, this tool lets you send and receive secure messages, manage group chats, and view inline image previews directly in your terminal window. The best part is the full modal editing support, allowing you to fly through your conversations using standard vim keybindings. It is the perfect setup for keeping your desktop clean and your hands on the home row. Give it a try and enjoy a distraction-free workspace.

🆔 @hackernewsgithubprojects

📦 cloudflare/cf

Cloudflare CF CLI

The cf command line interface is the terminal tool that uses artificial intelligence agents to search and run commands across the entire Cloudflare network. Instead of memorizing hundreds of complex commands, you can let an AI agent search and steer the interface to configure your infrastructure. It defaults to condensed data formats so AI agents can process information quickly and cheaply, while still showing pretty-printed text for humans. It even introduces a single TypeScript configuration file to manage your settings with automatic code safety. It is a fascinating glimpse into how tools are evolving to help developers build things faster alongside AI assistants.

🆔 @hackernewsgithubprojects

📦 joasasantos/offensive-security-ai-models

Offensive Security AI Models

Equip your authorized red team or security research lab with highly specialized, uncensored artificial intelligence models designed specifically for cybersecurity. The offensive security ai models repository is a curated directory of open weight large language models that have been fine tuned or modified to remove standard safety restrictions. By utilizing advanced training methods like abliteration and custom instruction fine-tuning on real bug bounty reports, this registry lets you run models locally that can write exploit code, generate security templates, and analyze malware without constant refusals. Grab one of these models to start simulating real-world cyber attacks in your test environment today.

🆔 @hackernewsgithubprojects

📦 automatis-tools/agents-can-communicate

Agents Can Communicate

Hand off unfinished code to a different AI without losing context, all locally. Agents Can Communicate links independent sessions so they can share notes and reviews, letting you switch models mid-task without starting over. It keeps everything on your machine, meaning no cloud accounts and no central boss agent. Just install it, point your clients at the same project, and let them talk. It’s a neat way to mix and match tools, and the handoff feature is the real headliner. Try it once, and you’ll see why passing the baton between AIs actually works.

🆔 @hackernewsgithubprojects

📦 kiwicanopy/kiwidesk

KiwiDesk Window Manager

KiwiDesk is the macOS window manager that finally makes tiling comfortable and accessible without forcing you to edit complex configuration files. While traditional tools demand a terminal first, KiwiDesk works instantly upon installation, letting you control everything through a standard settings window with familiar sliders and switches. It works completely in user space through the accessibility framework, meaning you never have to disable system integrity protection. Under the hood, you still get advanced layouts like sideways scrolling, a command line interface, and optional Lua scripting, giving you plenty of room to grow without breaking your native macOS desktops.

📰 https://news.ycombinator.com/item?id=49729211

🆔 @hackernewsgithubprojects

📦 nvidia/cuda-rust

cuda-rust

Developers can now compile standard Rust code directly into high-performance graphics card instructions without relying on complex wrapper languages. This is made possible by cuda-rust, an official compiler project that bridges the gap between safe, everyday programming and intense hardware acceleration. Instead of writing custom drivers or dealing with foreign language bindings, you can write idiomatic Rust code and translate it straight into the native assembly that graphics cards understand. It compiles code directly to parallel thread execution instructions while preserving safety checks. This means you get the blazing speed of graphics hardware combined with the safety guarantees of Rust, opening up clean and reliable GPU programming for everyone.

🆔 @hackernewsgithubprojects