📦 victortaelin/uniichat

UniiChat: The Chat That Never Ends

UniiChat is the single-thread AI assistant that finally solves the problem of context rot. Instead of starting fresh every time, it stores every message you send into a permanent memory tree that compresses older details into broad summaries while keeping recent thoughts sharp. This means the agent can remember specific things you asked about years ago without losing its place in the current conversation. The system runs on your own hardware using standard model providers, letting you build a personal AI that grows with you over time.

🆔 @hackernewsgithubprojects

📦 hybirdss/hycli

Hycli: Turn Websites into AI Tools

Hycli is the website tool that makes your entire web presence available to your AI without risking your account. It works by quietly inspecting a site’s structure and your actual browser login to build a set of safe, typed commands for your assistant. The coolest part is how it handles writes. It figures out how to create or edit things on a site but refuses to actually do it until you explicitly say yes. This means your AI can search, read, and plan changes using your real data, while keeping your login secrets locked inside your computer so they never leak to the model.

🆔 @hackernewsgithubprojects

📦 ely-open-source/elyoffice

Ely Office: The Offline Desktop Suite

A full office suite written in Rust runs entirely on your computer without a single ping to a server. That is exactly what Ely Office provides. It is a collection of desktop applications for documents, spreadsheets, mail, and presentations, all built to work offline. The team chose Rust for speed and safety, pairing it with a custom graphics engine to make the apps feel native and fast. This means your work stays on your machine, protected from cloud outages and privacy concerns. It is a local-first alternative to the heavy web-based tools we all use every day.

🆔 @hackernewsgithubprojects

📦 hugo-dz/spritefusion-sprite-normals

Give Flat Sprites Real Depth with Sprite Fusion Sprite Normals

Take any flat pixel art sprite and instantly give it the illusion of three-dimensional depth. This tool, called Sprite Fusion Sprite Normals, solves a specific problem for game developers: it turns simple, two-dimensional images into normal maps that allow them to react realistically to dynamic lighting. The surprising part is how it works; the software analyzes the shape and painted shading of your artwork to calculate a fake surface relief, meaning your sprites can cast shadows and catch light just like physical objects do. It handles single images, sprite sheets, and animated sequences, keeping the original pixel resolution and transparency intact.

🆔 @hackernewsgithubprojects

📦 mnemon-dev/mnemon

Mnemon: Memory for AI Agents

Your coding bot forgets everything the second you close the tab. That is the biggest annoyance in modern development. Mnemon fixes this by giving your artificial intelligence a permanent brain that actually thinks. Instead of shoving every sentence into a massive, slow database, it stores facts as a connected web of ideas. When your agent works, it asks what it already knows before guessing. This means it remembers your project structure and past decisions across different tools. It is a single, local file with no cloud subscriptions or extra keys. It simply plugs into the bot you already use. The result is an assistant that actually knows your codebase.

🆔 @hackernewsgithubprojects

📦 miaai-lab/deepseek-v4.1-flash-1017-2x-dgx-sparks

DeepSeek on Two Sparks

DeepSeek V4.1 Flash is the dual chip setup that turns two small desktop units into a massive local server. Instead of buying one giant expensive machine, you link two compact devices with a direct high speed cable to share the heavy lifting. The coolest part is that you can run a full one million token window on this modest hardware. That means you can feed in huge documents or long conversations without the system forgetting what you said at the start. It handles images and complex tools right out of the box. You just run one simple script to start everything up.

🆔 @hackernewsgithubprojects

📦 deepclause/deepclause-sdk

DeepClause SDK: The Self-Modifying Agent

Build an artificial intelligence agent that writes its own source code. DeepClause SDK is a TypeScript library that runs an interactive terminal where you speak plain English to an AI. Instead of returning a static answer, the agent edits a live Prolog program that serves as the agent's own brain. This program defines every tool, memory, and logic rule the agent uses. Because the agent modifies its own underlying code, it can instantly learn new skills and adapt to your specific task. You can even watch it think, test its changes, and roll back mistakes with a single command.

🆔 @hackernewsgithubprojects

📦 da7-tech/mind

Mind: The Brain for Your AI Coder

Give your AI coding assistant a brain that actually remembers what you told it last week. Mind is a single Python file that acts as persistent memory for your agents, solving the annoying problem of them forgetting context or hallucinating old details. It automatically captures decisions, links related concepts, and even runs a nightly dream cycle to consolidate what matters while pruning the noise. The coolest part is that it runs entirely offline with zero dependencies, meaning your code and context never leave your machine. It is simple, private, and genuinely useful.

🆔 @hackernewsgithubprojects

📦 heychengc/wallstreet-explainer

Wallstreet Explainer

Turn any financial question into a polished, cinematic video. This project provides a complete toolkit for creating those high-end, Wall Street-style explainers, handling everything from sourcing real news footage to rendering the final frame. The coolest part? It uses code to build a three-dimensional visual stage where giant numbers and real people interact, making complex data actually make sense. You get scripts for finding assets, cutting out specific speakers from interviews, and mixing audio, all designed to keep the pacing tight and professional. It’s less about writing code and more about giving your AI assistant a strict creative brief.

🆔 @hackernewsgithubprojects

📦 ip-please/ip-please.github.io

IP, Please: The Pixel Game That Is Actually A Joke

IP, Please is the browser game that makes your internet address look like a passport. It is a pixel art spinoff of the famous border control game where you play as a reviewer at a fictional station. Your only job is to spot travelers who are lying about their region. The best part is that everything is fake, from the stamps to the sounds, which are all made by code. It is a short, funny story about how we decide who belongs where based on a simple string of numbers. It is a clever little satire that you can play right in your browser without downloading anything.

🆔 @hackernewsgithubprojects

📦 google-deepmind/superhuman

AI Solves Open Math Problems

Google DeepMind released Superhuman, a collection proving that artificial intelligence can now solve long-standing math problems that have stumped humans for decades. The most striking result is an AI agent proving the Erdős-Hajnal conjecture for finite paths, a famous unsolved problem in combinatorics. This project is not just about answering test questions; it contains machine-checked proofs of complex theorems, such as a fifty-year-old question regarding Hilbert’s tenth problem. The repository serves as a transparent archive of these achievements, allowing researchers to inspect the logic and verify the work. It represents a significant step toward machines contributing original mathematical knowledge.

🆔 @hackernewsgithubprojects

📦 xwtk/hmg_nemesis

Hack Your Car's Infotainment System

Take complete command over your car's infotainment screen with Nemesis. This tool exploits a hidden maintenance path in Hyundai vehicles to install itself permanently, granting you full administrative control over the system. Once active, you can run custom scripts, install new apps, and access the root shell directly from your dashboard. It turns the car's head unit into a customizable interface without needing a computer. If you want to own your vehicle's software experience, this is how you get started.

🆔 @hackernewsgithubprojects

📦 joasasantos/rea-prompts

How REA Prompts Lets You Reverse Engineer Anything

REA Prompts is the instruction library that turns your AI coding agent into a reverse engineering expert. This collection solves the problem of not knowing how to ask the right questions when you want to understand how software actually works. Instead of guessing, you paste these ready-made prompts to trace features, clone behavior, or analyze native binaries. The most interesting part is how it handles clean-room cloning, where the agent maps exactly how a program behaves without copying a single line of code. You get a full spec of inputs and outputs to build your own version from scratch.

🆔 @hackernewsgithubprojects

📦 deepseek-ai/deepinfer

DeepInfer: The Engine Behind DeepSeek

DeepInfer is the open-source engine that actually runs DeepSeek’s models in real-world production. Instead of just loading a model, it orchestrates every step of generating text with extreme precision, ensuring that the output stays mathematically identical no matter how many requests are running at the same time. This is the surprising part: it uses a technique called batch-invariant inference so that the AI does not change its mind based on the traffic around it. You get consistent, reliable answers even under heavy load, which is critical for serious applications.

📰 https://news.ycombinator.com/item?id=50163277

🆔 @hackernewsgithubprojects

📦 elpapi42/pi-observational-memory

Pi Observational Memory: Endless Sessions

An AI coding agent can stay sharp for weeks without forgetting a single decision. Pi Observational Memory fixes the silent amnesia that happens when long coding sessions hit their context limit. Instead of summarizing your entire history in one slow, lossy burst, it quietly watches your work in real time, picking out key decisions and turning them into durable facts before you even need them. This means when the system finally has to compress your session, it already has a clean, accurate snapshot ready to go, so the handoff is instant. You stop losing the thread of your logic after a few hours of debugging.

🆔 @hackernewsgithubprojects

📦 coollabsio/openclaw

OpenClaw: The AI Agent That Actually Works

OpenClaw is the self-hosted Docker image that gives your AI agent a real, persistent body. It solves the frustration of running complex AI tools in containers by baking in a complete, automated setup that survives restarts. You get a full browser environment with a visible desktop, allowing the AI to log into sites, handle two-factor authentication, and maintain active sessions just like you do. The entire system, from chat history to configuration, lives on a single persistent volume so your data never vanishes when the container is recreated.

🆔 @hackernewsgithubprojects

📦 stackedlol/ascii-sharks

Ascii Sharks: A Living ASCII Ocean

Sharks swim across your screen rendered entirely in text characters. This project, called Ascii Sharks, builds a living ocean simulation using only a code grid and zero external libraries. It features eight distinct species, from classic great whites to rare thresher sharks, each with unique shapes and movements. The simulation uses mathematical currents to guide the fish, creating a natural, flowing experience. It also respects accessibility settings, offering a static view for users who prefer reduced motion. This is a fascinating look at how creative coding can turn simple characters into a complex, animated world. The takeaway: even basic tools can create surprisingly rich digital art.

🆔 @hackernewsgithubprojects

📦 k0valik/pi-blackhole

pi-blackhole: Free AI Memory

Compress long AI conversations without losing crucial details, thanks to pi-blackhole. This tool solves the frustrating problem where artificial intelligence forgets important decisions after summarizing a lengthy chat. Instead of asking a model to write a summary, which is slow and costly, this project uses a mathematical method to shrink the text instantly for free. While compressing, three background helpers quietly read the conversation to remember key facts. They save these notes so your agent stays smart and consistent, even after hours of work. It turns a chaotic, forgetful chat into a reliable, long-term memory system. Give your AI a true long-term memory by installing pi-blackhole today.

🆔 @hackernewsgithubprojects

📦 reticlehq/reticle

Reticle: The AI QA That Catches Lies

Your AI coding agent just told you the fix is done, but it has never actually opened your app to see if it works. Reticle solves this by giving your AI the ability to look at the live software, click buttons, and check the network traffic behind the scenes. Instead of just taking its word for it, the tool drives your real running application and returns a clear verdict: pass, fail, or cannot tell. This is crucial because it catches silent errors like a broken button that still loads the page but does nothing when clicked.

📰 https://news.ycombinator.com/item?id=50163896

🆔 @hackernewsgithubprojects

📦 memtensor/memt-1.0

MemT-1.0: The AI That Actually Remembers

Give your AI a permanent memory that doesn't bloat your hard drive. MemT-1.0 is a new family of foundation models that keeps a constant-sized memory state no matter how long your conversation gets. Instead of storing every single word you ever typed, it compresses your chat history into a tiny, evolving snapshot. This means you can talk to it for hours, save that memory to a file, and pick up exactly where you left off later without reloading the entire transcript. It even remembers how to code by keeping past experiences in its head rather than pasting them into the prompt.

🆔 @hackernewsgithubprojects