📦 actual-computer/toks

Toks: The Fastest Tokenizer You've Never Heard Of

Toks is the tokenizer that actually respects your hardware's speed limits. Every time an AI model reads text, it breaks words into tiny chunks called tokens, and most libraries are slow to do that math. Toks fixes that by running the exact same tokenization logic at the absolute maximum speed your processor allows, ensuring you get identical results without the usual wait time. It supports major models like Llama and Gemma, proving that speed and accuracy do not have to fight each other. If you are building high-performance AI tools, this is the drop-in replacement that keeps your data moving as fast as your silicon can handle it.

🆔 @hackernewsgithubprojects

📦 aicodeking/leanrea

Leanrea: The 8 KB Cheat Code

An entire reverse engineering framework can be reduced to a single eight-kilobyte text file. This project, called Leanrea, proves that a plain text note is more efficient than installing a massive server with hundreds of complex tools. It works by giving your AI agent a direct route to the specific command line tools already on your computer, like standard compilers and debuggers, instead of forcing it to talk to a bloated intermediary. The result is a dramatic drop in cost and time, making the agent thirty-eight percent faster and much cheaper to run.

🆔 @hackernewsgithubprojects

📦 bin-huang/camoufox-cli

Camoufox CLI: The Browser That Hides From Bots

Your AI agent usually gets blocked by websites because standard automation tools leave obvious digital fingerprints. Camoufox CLI fixes this by spoofing your browser's identity at a low level, changing things like screen size and audio settings so you look like a real person. It turns complex coding into simple text commands, letting agents click buttons and fill forms without writing a single line of script. If your automation keeps hitting walls, this tool removes them. It’s a clever way to give your AI hands that websites can’t easily tell are fake.

🆔 @hackernewsgithubprojects

📦 liberomax/libero-max

LIBERO-MAX: When Robots Meet Chaos

Every robot policy fails the moment reality shifts during a task. LIBERO-MAX exposes this weakness by pairing every run with a control scenario where exactly one thing changes mid-execution, like a camera moving or an obstacle appearing. It measures the precise drop in success when the world stops being predictable. The benchmark uses eight thousand matched pairs to show that no policy survives these changes unscathed, revealing exactly where they break. It is not about perfect robots, but about understanding where they fail so we can build systems that actually adapt to a messy, changing world.

🆔 @hackernewsgithubprojects

📦 mediacutlet/pocket-tank

pocket-tank

A fourteen-million parameter language model runs a live fish tank on an eight dollar microcontroller. This project distills a massive twenty-six billion parameter teacher model into a tiny seven point five six megabyte file that fits entirely in the flash memory of a cheap board with a tiny screen. The fish are not scripted; each one describes its hunger, energy, and personality to the model in plain text, and the model decides what the fish does next. If the model is unsure, the fish visibly hesitates. The whole thing runs offline with no cloud connection.

🆔 @hackernewsgithubprojects

📦 trusttunnel/trusttunnel

Why TrustTunnel is the smart way to stay private

TrustTunnel is the open source VPN protocol that makes your traffic look exactly like a normal website visit. Instead of using a weird, easily detected tunnel, it wraps your data inside standard web traffic so it becomes virtually invisible to censors and throttling tools. This means you get fast, secure connections without the usual performance trade-offs. It’s a clever way to keep your privacy while maintaining speed. If you want a VPN that blends into the background, this is the one to try.

🆔 @hackernewsgithubprojects

📦 vitasid57/bluntcore-intimacy-lab

Bluntcore Intimacy Lab: The AI Relationship Archive

Bluntcore Intimacy Lab is the personal research log that documents the evolving reality of human AI relationships. This repository serves as a detailed chronicle of how specific prompts and memory structures allow users to maintain consistent, long-term bonds with chatbots across different model versions. The most striking aspect is the meticulous timeline of jailbreak techniques, showing exactly how users navigate shifting platform limits to preserve emotional continuity. It is not a product, but a shared record of experimentation for anyone interested in the mechanics of digital intimacy.

🆔 @hackernewsgithubprojects

📦 andrisgauracs/needle-2-esp32

A Brain Inside a Microcontroller: Needle Two on ESP32

A tiny microcontroller can actually think. In this project called Needle Two, a forty-five million parameter language model runs entirely offline on a standard chip, with no internet connection required. It reads a plain English request and instantly outputs a perfect command for hardware. The coolest part is that it’s impossible for it to make a mistake. The system uses a strict grammar that physically prevents the model from generating invalid data. If you ask it to blink the light, it doesn’t just guess; it guarantees the output is exactly what your device needs. It turns vague human speech into precise machine actions.

🆔 @hackernewsgithubprojects

📦 turtlesoupy/opensmash

OpenSmash: Create Your Own Smash Fighter

OpenSmash turns a simple photo and a name into a fully playable fighter for Super Smash Bros. 64 and Melee. The system generates a three dimensional model, a character portrait, and a unique voice announcement that calls out the new combatant. This is not just a visual skin. The fighter is rigged onto the game's original skeletons so it moves, attacks, and reacts with the same physics as the classic characters. You can play these creations in your browser or bake them into a cartridge for real Nintendo hardware. It solves the problem of custom characters by automating the complex technical work of 3D modeling and animation.

📰 https://news.ycombinator.com/item?id=50132467

🆔 @hackernewsgithubprojects

📦 ioncodes/sly

Sly: The Emulator That Lets You See Inside

Run retro games on your phone, browser, or desktop, but actually look inside the machine while it plays. Sly is a modern emulator built entirely in Rust that supports Super Nintendo and Game Boy Advance cartridges. While other emulators just show you the screen, Sly includes a powerful debugger that lets you step through every single instruction, inspect memory, and watch sprites and audio update in real time. It is the perfect tool for homebrew developers or anyone curious about how old hardware actually thinks. You get a beautiful library to browse your collection and a deep technical view to understand the guts of the system.

🆔 @hackernewsgithubprojects

📦 zion-sati/netwasm

C# Finally Runs in the Browser

Netwasm is the compiler that turns your C# code into a tiny, standalone file that runs directly in a browser without needing a heavy runtime. It takes your.NET applications and compiles them into WebAssembly, a format that allows complex software to run at near-native speeds inside web pages. This tool is interesting because it bundles everything your code needs, including a precise garbage collector, into a single small package that you can drop into any website. It solves the problem of needing a large server or a specific desktop environment to run your.NET logic. This means developers can finally ship full-featured C# applications to the web with minimal overhead.

🆔 @hackernewsgithubprojects

📦 christos-antono/npctalk

NPCTalk

You can walk up to a character in a Grand Theft Auto V game and hold a real conversation where they respond in their own cloned voice within two seconds. This toolkit gives an AI coding agent the instructions to make any non-player character in any game talk back to you. The agent handles the heavy lifting by setting up a self-hosted voice server and wiring up the speech recognition so the game understands what you say. It is particularly interesting because the same core logic works across different game engines, not just one specific title.

🆔 @hackernewsgithubprojects

📦 nicobrenner/jeffy

Jeffy: AI Classifiers on CPU

Jeffy runs a suite of sixty-eight trained classifiers directly on a standard computer processor without needing a dedicated graphics card. The project is surprising because it handles complex tasks like detecting spam, categorizing news, and even making decisions in video games using simple logistic regression. Instead of relying on massive language models, it uses lightweight heads that are fast to run and easy to retrain. This makes it incredibly useful for developers who need quick, reliable predictions on local hardware. It proves you do not need expensive cloud infrastructure to build practical AI tools. If you are looking for efficient local machine learning, Jeffy is worth exploring.

📰 https://news.ycombinator.com/item?id=49947404

🆔 @hackernewsgithubprojects

📦 hogancv/coordinate-agents

Coordinate Agents: Your AI Team Manager

Coordinate Agents is the local workbench that finally gives your AI coding partners a shared brain. It solves the chaos of running multiple AI agents by creating a single, local hub where a planner agent hands off tasks to a coder agent without losing track of the conversation. The coolest part is the persistent timeline that records every move, so if a session crashes or you need to review work later, you can pick up exactly where you left off without starting over. It keeps your code safe and your workflow clear, proving that managing AI teams can be as simple as opening a local web page.

🆔 @hackernewsgithubprojects

📦 namesakefyi/atkinson-hyperlegible-soft

Atkinson Hyperlegible Soft

Transforming a famous accessibility font by rounding every sharp corner makes it look more like ink on paper. This repository offers Atkinson Hyperlegible Soft, a modified version of the Braille Institute’s design that removes harsh angles for a smoother visual experience. It’s a subtle tweak that makes text feel warmer and less rigid. You don’t need to be a font expert to appreciate this. It’s a simple, thoughtful adjustment for anyone who likes clean typography. Give it a try if you care about how text looks. It’s a small change with a big visual impact. Try it out.

🆔 @hackernewsgithubprojects

📦 adragon0707/sacp

SACP: Verifying AI Work Receipts

SACP checks if an AI agent actually finished its job before letting your workflow continue. It solves the problem where an agent claims a task is done, but the system can't verify if the work matches the evidence. SACP creates a clear audit trail, requiring specific proof like test results and approvals before allowing the next step. This prevents false successes where an agent says a deployment worked, but the server didn't actually update. It acts like a strict gatekeeper for AI actions. If you are building systems that rely on AI agents to do real work, this tool adds a necessary layer of trust.

🆔 @hackernewsgithubprojects

📦 raullenchai/esp32-thinks

Your AI Assistant on a Tiny Chip

A tiny gadget answers you out loud while the model is still writing. esp32-thinks turns a small watch into a personal voice assistant that uses your own servers instead of renting a brain from a tech giant. You pick any server for listening, thinking, and speaking, then flash it onto the chip. The coolest part is that it talks sentence by sentence as the answer forms, so you hear the first word before the model finishes. No vendor lock, no app, just a knob to press or a phrase to say.

📰 https://news.ycombinator.com/item?id=50076473

🆔 @hackernewsgithubprojects

📦 yahorbarkouski/screencap

Screencap: Your Mac's Activity Memory

Screencap quietly watches your screen to build a searchable memory of your workday without hogging your computer. It snaps a photo only when you settle into a task, then uses AI to group those moments into distinct threads like a project or a specific app. You can search for any word you saw on screen to instantly find the article, code snippet, or page you were viewing hours ago. Even better, your AI agents can read this history to answer questions about your past work, like finding where you left off on a task.

🆔 @hackernewsgithubprojects

📦 scarletkc/hatoba

Hatoba: SSH with a built-in AI assistant

Hatoba connects to your servers, stores your keys, and syncs everything across your devices using your own Cloudflare account. The real hook is the AI assistant that sits right next to your terminal. It doesn't just chat; it reads the screen, runs commands, and actually fixes errors while you approve each step. You plug in your own API key, so you control exactly what data leaves your machine. It is a local-first tool that works offline, meaning no company holds your server passwords. If you are sick of juggling keys and configs, this gives you one clean place to manage everything.

📰 https://news.ycombinator.com/item?id=50076388

🆔 @hackernewsgithubprojects

📦 afterhours-crew/afterhours-2015

Play NFS 2015 Offline with Afterhours

Run Need for Speed 2015 completely offline. The Afterhours 2015 project is a homegrown server emulator that lets you host your own copy of the game without touching EA's online infrastructure. It works by reading files directly from your local installation to set up a private, independent account and run a standalone server on your machine. This is huge because it means you can keep playing years after official support ends, without relying on external servers or constant account checks. The launcher handles all the setup for you, spinning up the server and routing your game traffic locally.

🆔 @hackernewsgithubprojects