📦 usewhale/whale

Whale

Whale is a terminal-first AI coding assistant designed specifically to exploit DeepSeek's massive context window and unique cost efficiencies. Instead of acting as a generic multi-model wrapper, this lightweight Go utility is built from the ground up for DeepSeek's specific tool-calling capabilities. It achieves an incredible ninety-eight percent prompt cache hit rate, bringing the cost of continuous, high-context AI coding down to pennies per session. Developers can launch an interactive terminal interface, run one-shot terminal commands, or even write custom JavaScript files to orchestrate complex multi-agent workflows. It is the perfect tool for developers who want ultra-fast, local-first AI development without paying a premium.

📰 https://news.ycombinator.com/item?id=48930689

🆔 @hackernewsgithubprojects

📦 zhcz328/awesome-medical-agents

Awesome Medical Agents

Track the rapid evolution of artificial intelligence in healthcare by exploring a curated roadmap of autonomous clinical systems. This repository aggregates cutting-edge research papers and open-source code focused on building AI medical agents that go beyond simple chat boxes. Instead of just answering questions, these systems can actively look at medical scans, plan multi-step workflows, cross-reference external databases, and double-check their own logic to minimize errors. It is an incredibly organized guide to how developers are chaining together reasoning, memory, and specialized tools to assist doctors. Dive into awesome-medical-agents to see how clinical AI is moving from simple assistants to truly smart, cooperative healthcare partners.

🆔 @hackernewsgithubprojects

📦 model-architectures/grape

GRAPE

Grape is the mathematical framework that finally unifies how AI models understand the order of words in a sentence. Usually, models use completely different mathematical tricks to track position, either multiplying values to rotate them in space or adding offsets like a scoreboard. Grape proves these methods are actually two sides of the same coin, using elegant group theory to merge them into one system. This mathematical harmony makes context tracking incredibly efficient, leading to better language modeling scores without needing massive hardware changes. It is a beautiful, unifying breakthrough that cleans up transformer math while making text generation smarter.

🆔 @hackernewsgithubprojects

📦 justlovesmile/synclip

SynCLIP

Synclip is the image-training model that aligns synonym concepts to help computers recognize objects under any name. Typically, AI models struggle to identify a visual object if you use a word they did not train on, like calling a couch a sofa. This project solves that disconnect by enforcing language coherence, ensuring that different words with the exact same meaning point to the same visual patterns. It makes computer vision systems far more adaptable when identifying things in the real world using everyday, diverse language. Check out the code to see how it bridges the gap between flexible vocabulary and precise image recognition.

🆔 @hackernewsgithubprojects

📦 sickn33/agentic-awesome-skills

agentic-awesome-skills

agentic-awesome-skills is the installable library that finally gives your AI coding assistants a massive upgrade in practical capabilities. Instead of pasting random, messy prompts into tools like Claude Code or Cursor, this project lets you install a massive catalog of nearly two thousand structured playbooks directly where your AI expects them. You can search, filter, and install specialized workflows for coding, debugging, or security reviews with a single command. It turns your AI assistant from a generic chatterbox into a highly specialized, context-aware teammate. Grab it today to make your AI agents work smarter and deliver exactly what you actually need.

🆔 @hackernewsgithubprojects

📦 zerolu/awesome-gemini-ai

Awesome Gemini AI

awesome-gemini-ai is the ultimate collection of prompt templates that finally unlocks the true potential of Google's advanced language models. Instead of wrestling with trial-and-error phrasing, you get a curated library of high-performance instructions sourced from top prompt engineers on Reddit and Twitter. The single most impressive feature is its massive, multi-phase system that acts as an elite web designer. It guides the model step-by-step from brainstorming visual aesthetics to outputting fully functional code for complex, modern websites. It is the perfect resource for anyone looking to build impressive web applications and interactive interfaces with simple text.

🆔 @hackernewsgithubprojects

📦 kilted-kraken/-rohankar-launcher

RohanKar Launcher

RohanKar Launcher transforms how you play retro PC classics by letting you search, download, and launch games from Archive.org entirely with a game controller. This desktop application solves the hassle of digging through web links and manually setting up files by handling everything automatically, from one-click extraction to smart executable detection. It even automatically unblocks files to skip Windows permission errors, writes shortcuts directly to Steam, and tracks your playtime. It is the perfect tool for retro gaming fans who want a polished, console-like library experience without any configuration headaches.

🆔 @hackernewsgithubprojects

📦 shadcnspace/shadcnspace

shadcnspace

Build beautiful web interfaces in minutes by grabbing pre-designed React components and fully styled block templates. Instead of starting your pages from scratch or wrestling with layouts, this resource gives you a massive library of ready-to-use structures like interactive bento grids, pricing tables, dashboards, and custom buttons built using Tailwind and Radix UI. You just browse the visual catalog, pick the exact design block or UI component that fits your project, copy the clean code, and drop it straight into your application. It is the ultimate shortcut for developers who want professional-looking websites without spending endless hours on CSS design.

🆔 @hackernewsgithubprojects

📦 the-ora/browser

Ora Browser

Browse the web on your Mac without the usual clutter by trying a fast, native browser designed to keep things simple. This project, called browser, is a lightweight macOS web browser built from the ground up using Swift and Apple's native technologies to give you a clean, low-friction experience. Instead of layering on heavy, unnecessary features that slow down your computer, it focuses on the essentials like built-in content blocking, customizable search engines, and smart URL suggestions. It is the perfect project to check out if you love minimalist, high-performance Mac utilities. Grab the code, build it in Xcode, and see how clean your web browsing can actually be.

🆔 @hackernewsgithubprojects

📦 youmind-openlab/ai-image-prompts-skill

AI Image Prompts Skill

You can now connect your favorite AI assistant to a library of over ten thousand curated image generation prompts that update twice a day. This project is a universal developer skill that integrates directly with tools like Claude Code, OpenClaw, and Cursor. Instead of guessing how to write the perfect prompt, you simply describe what you need in plain English or paste in an entire article. The assistant automatically searches the collection, pulls up matching templates complete with real sample images, and customizes the prompt to fit your specific content. It works across every major image generator, giving you instant, predictable results.

🆔 @hackernewsgithubprojects

📦 injaneity/pi-computer-use

pi-computer-use

Give your AI agents the power to actually operate your desktop software just like a human does. This clever package integrates directly with the Pi assistant, allowing AI agents to interact with normal macOS and Windows apps when standard APIs or command-line tools aren't an option. Instead of relying on rigid integrations, the agent looks at an application window, identifies elements like buttons and text fields, and physically clicks, types, scrolls, or waits for changes. It is perfect for automating repetitive desktop tasks, keeping everything running smoothly in a secure, controlled environment that bridges the gap between AI and legacy software.

🆔 @hackernewsgithubprojects

📦 kartiklabhshetwar/oneurl

OneURL

You can host your own fully featured link directory without paying a dime or giving up your data. This is oneurl, a self-hostable, open-source alternative to Linktree that gives you complete control over your public profile. It features built-in Google login, drag-and-drop avatar uploads, and deep click analytics to track your audience engagement. The entire application runs on modern web tech like Nextjs, Prisma, and PostgreSQL, letting you deploy a beautiful, lightning-fast personal directory that you actually own. It is the perfect project for anyone wanting to escape third-party subscription plans and run their own public bio page with ease.

🆔 @hackernewsgithubprojects

📦 xai-org/grok-build

grok-build

SpaceXAI built an incredibly powerful terminal-based coding assistant that manages your entire development cycle directly from a single fullscreen window. Named grok-build, this tool is not just another wrapper; it is a mouse-interactive console environment that actually understands your codebase, edits files, and runs terminal commands for you. It handles complex, long-running tasks autonomously, or you can control it headlessly for automated scripting. Because it integrates directly into your terminal, it gives you a fast, distraction-free environment to pair-program without leaving your workspace. Try spinning it up on your next project to let an agent handle the tedious boilerplate while you focus on the big architecture.

📰 https://news.ycombinator.com/item?id=48926590

🆔 @hackernewsgithubprojects

📦 wisher567/portpal

Portpal

Stop hunting down blocked network ports with painful terminal searches when launching your code. This desktop app instantly scans your active connections, shows you exactly what background service is squatting on your local network, and lets you kill the process with a single click. It even recognizes popular frameworks and web tools out of the box to group everything into a clean dashboard, and creates an interactive network map visualizing your local traffic. It is a incredibly quick way to handle port conflicts. Try downloading it for your development setup to keep your workflow completely clean and conflict-free.

🆔 @hackernewsgithubprojects

📦 zksha/alma

Alma

AI agents are usually only as good as the memory systems humans hand-engineer for them, but a new framework called alma completely flips the script by letting AI design its own memory. Instead of relying on static database schemas or human-coded retrieval rules, this clever system uses a meta-agent to write, test, and refine actual executable code to manage its memory. When put to the test in complex gaming and reasoning environments, these self-designed memory systems easily outpaced traditional human setups. It turns out that letting the model code its own brain makes it a far more efficient and adaptable learner.

🆔 @hackernewsgithubprojects

📦 temple-stuart/temple-stuart-accounting

temple-stuart-accounting

temple-stuart-accounting is the financial control center that finally unifies ledger-grade double-entry bookkeeping with real-time trading and lifestyle planning. Instead of wrestling with fragmented spreadsheets and disconnected banking feeds, this Next.js platform acts as a central brain. It pulls together live data from brokerage accounts and retail bank networks, matching your day-to-day transactions with complex stock lots, wash-sale calculations, and tax-mapping rules automatically. It even connects your financial health directly to real-time travel planning, letting you budget trips and track investments in one single dashboard. If you want to see how advanced financial automation can run on a single web stack, you need to check this out.

🆔 @hackernewsgithubprojects

📦 automl/timee

timee

Classify raw time series data instantly using a single forward pass without needing any model training or fine-tuning. This project called timee is a pretrained transformer that brings in-context learning to time series analysis, working just like a large language model. By feeding the model a few labeled training examples along with your unlabeled test data, it identifies the correct categories on the fly. It automatically handles multivariate data, scales to any sequence length, and uses a built-in ensemble to boost performance. It makes time series classification as simple and fast as running a pre-trained model on a new prompt.

🆔 @hackernewsgithubprojects

📦 jsu-kim/persona-pruner

Persona-Pruner

Build compact, custom role-playing AI models from a single natural-language description. The magic behind persona-pruner is that it automatically turns generic training data into your target character's voice, measures which questions make that personality shine, and then structurally prunes away the unnecessary parts of the neural network. By identifying and removing redundant feed-forward network channels while keeping the core personality intact, you get a lightweight model that role-plays beautifully without the massive computational footprint. It is like taking a giant AI and sculpting it down to just the brain cells needed for your specific character. Clone the repository to start sculpting your custom models today.

🆔 @hackernewsgithubprojects

📦 chart-research/g-mapp

g-mapp

g-mapp is the robotic control system that finally lets multi-agent machines react to messy, chaotic spaces in real time. Standard robot planners are painfully slow because calculating complex paths around unexpected obstacles on a traditional processor takes forever, forcing robots to pause or move like nervous snails. This system changes the game by shifting all that heavy planning and sensor math onto a graphics card. By running thousands of potential movements in parallel, it creates a lightning-fast coordination loop. Check out the repository, run the simulation, and see how fast your robot can think on its feet.

🆔 @hackernewsgithubprojects

📦 callstack/agent-device

Agent Device

Control real iOS and Android devices directly through your terminal using a command-line tool built specifically for AI agents. While most automation tools are designed for human testers, this project acts as the hands and eyes for artificial intelligence, allowing coding agents to autonomously open applications, tap buttons, type text, and inspect the screen hierarchy in real time. It solves the massive headache of mobile app testing by letting your AI models capture screenshots, record video, and gather raw device logs to diagnose bugs completely on their own. It is the ultimate bridge between software agents and physical mobile platforms.

🆔 @hackernewsgithubprojects