📦 rociiu/talorys

Talorys: Your Personal Cloud AI

Talorys is the open-source AI agent that lives exclusively inside your own Cloudflare account. Instead of renting a chatbot from a third party, you deploy your own private assistant with a single command, keeping your chats, tasks, and memories entirely under your control. The standout feature is its durable memory; it remembers your preferences and past conversations so you never have to repeat yourself, and even if the AI service goes down, your to-do list keeps working. It is free, private, and ready to go. You get a smart, personal digital secretary that actually belongs to you, with no data sent to strangers.

📰 https://news.ycombinator.com/item?id=50031614

🆔 @hackernewsgithubprojects

📦 feizhululu/deepseek-bot

DS Bot: Your AI Team on DeepSeek

DS Bot is the plugin for DeepSeek Harness that finally gives you a real team of AI workers. Most tools hand you one assistant, but this project lets you create a roster of distinct bots that actually talk to each other. The coolest part is the group chat mode. You throw a problem into a shared space, and the bots take turns, delegate tasks to specialists, and coordinate the answer right in front of you. It feels less like using a tool and more like managing a small office.

🆔 @hackernewsgithubprojects

📦 parcha-ai/agentrun

Agentrun: The Workflow Language for Your AI Agents

Your AI agents are powerful, but they often improvise their way through tasks, which makes their results hard to trust. Agentrun solves this by giving you a specific language to write out exactly what your agents should do, step by step. Think of it like a recipe: you define the steps, the checks, and the fallbacks, and the agent follows that script while keeping its own tools and permissions. The most surprising part is that you can actually inspect and test these workflows like regular code, so you know exactly why an agent gave a certain answer.

📰 https://news.ycombinator.com/item?id=49821438

🆔 @hackernewsgithubprojects

📦 leeguooooo/iphone-use

The AI That Drives Your Real iPhone

iphone-use is the open-source tool that lets your AI coding agent physically operate your actual iPhone. While most automation tools struggle with apps that have no public API, this project reads the screen as text and executes taps, swipes, and typing directly on the hardware. It works with banking, health, and chat apps, bypassing the need for complex code integrations. You can even record a task once and replay it instantly without burning through tokens. It’s a clever way to give your digital assistants a pair of hands, turning your phone into a fully automated, reliable tool that works exactly like a human would.

🆔 @hackernewsgithubprojects

📦 pama-lee/macminimode

Mac mini Mode

Your MacBook can run as a headless server with the lid closed, right now, without plugging in a dummy HDMI adapter. A project called Mac mini Mode handles this automatically, keeping your computer awake and working only when it is plugged into a wall charger. The moment you unplug it to move, the system instantly returns to normal sleep mode so you never carry a hot, running laptop in your bag. It also includes a simple health check to make sure your Mac is ready for remote access.

🆔 @hackernewsgithubprojects

📦 naklecha/simple-llm

SimpleLLM: A Faster, Leaner LLM Engine

Run OpenAI’s GPT Open Source models on a single high-end graphics card using just one thousand three hundred fifty one lines of code. Simple L M is a minimalist inference engine built from scratch that skips the heavy infrastructure of standard tools. The standout feature is its raw speed; it decodes text eleven to fifteen percent faster than the industry standard, all while matching official accuracy scores perfectly. By writing custom math routines directly for the hardware, this project proves that you don’t need a massive, complex framework to get high performance. It loads models without reprocessing their data and runs smoothly on one machine.

🆔 @hackernewsgithubprojects

📦 mindroom-ai/mindroom

Mindroom: Your Personal AI Team

Mindroom is the open source chat app that turns AI models into a team that actually knows you. Most assistants are generic, but this builds a private room where your agents learn your specific work habits and context. You can self-host the whole thing, meaning your data stays on your machines and you can pick any model you like. It’s less like a simple chatbot and more like a dedicated staff member that lives inside your own infrastructure. The best part is that it’s built to be flexible, so you are never locked into a single provider or a rigid workflow.

🆔 @hackernewsgithubprojects

📦 amazon-far/php_parkour

php parkour

Humanoid robots can now autonomously run through complex obstacle courses using only a standard depth camera. This repository, called php parkour, solves the problem of teaching machines to chain dynamic human skills like climbing and stepping into long, continuous movements. Instead of relying on perfect sensors, it uses motion matching to build trajectories from human motion data and then distills that knowledge into a policy that runs on simple vision. It is interesting because it bridges the gap between simulated motion databases and real world autonomy. The takeaway is that robots are getting closer to handling unpredictable terrain without expensive lidar.

🆔 @hackernewsgithubprojects

📦 data-goblin/claude-code-filetree

Claude Code Filetree: Watch Your Code Live

Visualize every move your AI coding assistant makes in real time with this open-source extension. It turns a blank terminal into a live dashboard that shimmers with color as your agent reads, writes, or commits files, giving you instant clarity on exactly where in your project the intelligence is working. It also overlays your git status directly onto the file tree, so you see changes before they hit your branch. This tool transforms a black box into a transparent workspace, letting you supervise your digital pair programmer with total confidence and zero guesswork.

🆔 @hackernewsgithubprojects

📦 polymarket/py-sdk

Python Meets Prediction Markets

Automate your entire trading workflow on Polymarket using the official Python SDK. This library gives you a single, coherent way to handle everything from checking public market data to placing authenticated orders and managing your crypto wallet. The coolest part is that it hides all the messy, complicated backend plumbing behind one simple, friendly interface, so you don't have to worry about low-level connection details or complex authentication steps. You can even grab real-time data streams and batch your position updates with just a few lines of code. If you are building bots or analyzing prediction markets, this tool makes your life significantly easier.

🆔 @hackernewsgithubprojects

📦 oxynote/oxynote

Oxynote: Your System With Subtitles

Place your live system metrics directly next to the text explaining them. Oxynote solves the headache of keeping dashboards and documentation separate by putting a live chart and its written context on the same page. You can even review changes to both the code and the chart together, just like a software update. This keeps your team’s notes and your actual server health in perfect sync, so you always know what a number means. Stop guessing what your data says and start documenting your system clearly.

🆔 @hackernewsgithubprojects

📦 explorium-ai/vibeprospecting-mcp

Vibe Prospecting: Your AI Sales Researcher

Vibe Prospecting is the remote plug-in that turns your AI assistant into a sharp sales researcher without you installing a single thing. Most data tools make you wrestle with confusing code or forgetful browser tabs, but this connects directly to your favorite chat bot to pull real company details and contact info right into your conversation. You simply ask for a list of potential clients in your industry, and it instantly shows a quick sample so you can check the quality before committing. It is the perfect way to test the waters on lead generation without burning through your entire budget on expensive full data exports.

📰 https://news.ycombinator.com/item?id=50041433

🆔 @hackernewsgithubprojects

📦 lithos-ai/lithos-metal

Lithos Metal: Faster AI on Your Mac

Lithos Metal is an open-source engine that makes large language models run significantly faster directly on your Mac’s built-in graphics chip. The biggest surprise here is how it squeezes speed out of the hardware by combining multiple math steps into single, massive instructions, letting the processor handle several words at once instead of just one. This means local AI responses appear much quicker, which is great for developers using coding assistants without sending data to the cloud. The project focuses on low latency for specific Apple chips, proving you can run heavy models locally with impressive speed.

🆔 @hackernewsgithubprojects

📦 a-funk/open-appleremote

Turn Your Old Apple TV Remote Into a Linux Keyboard

Resurrect that first-gen Apple Siri Remote lying in a drawer by turning it into a functional Linux input device. Open Apple Remote is a pure Python project that unlocks the hardware, decodes its raw touch and button data, and translates those signals into usable keyboard events or smooth, Apple TV-style gestures. It works directly on the Linux kernel without any external libraries, letting you control media players, browse the web, or build custom hardware interfaces using just standard tools. You get a live web lab to test your swipes and a configurable key map to assign any key to the remote's buttons.

🆔 @hackernewsgithubprojects

📦 nandhakishorm/vegaml

Vegaml: Decisions via a Tiny Physics Engine

Vegaml treats answering a question like dropping a ball into a bowl to see where it settles. Instead of generating text, this tiny model uses a small physics engine to roll a particle toward the most likely answer. It gives you a clear choice with a confidence score, and it tells you when it is unsure. It runs on your own hardware and even reads images. You get calibrated probabilities without the fluff. It is a smart, honest way to get decisions from AI. It is not a chatbot. It is a calculator for certainty. Check out Vegaml for reliable, grounded answers.

🆔 @hackernewsgithubprojects

📦 humanlayer/fold

Fold: The Coding Agent That Switches Models on the Fly

Fold is the coding agent framework that lets you swap AI providers without rewriting your code. It gives you a single, consistent interface to talk to models from Anthropic, OpenAI, or even subscription services like Codex, so you are never locked into one vendor. The standout feature is its automatic tool selection; it detects exactly which model you are using and hands it the correct file editing tools, whether that is a standard write command or a specialized patch format. This means you can change your brain on the fly without breaking your workflow, making it an incredibly flexible foundation for building your own intelligent coding assistants that work everywhere.

🆔 @hackernewsgithubprojects

📦 aipmandy/how-to-survive-unemployment

How to Survive Unemployment

Protect your savings by navigating the messy aftermath of a job loss with the How to Survive Unemployment guide. This open-source manual cuts through the confusion of severance negotiations and government benefits to give you a clear, step-by-step roadmap. Instead of vague advice, it provides fifty specific action tasks and six detailed playbooks that tell you exactly what to do, from checking your medical insurance status to calculating your cash flow buffer. It even includes real policy details for Beijing and Shanghai so you can verify the rules yourself. You get a practical toolkit to manage your money and your job search without guessing.

🆔 @hackernewsgithubprojects

📦 fei2411/deepseek-bot

DeepSeek Bot

DeepSeek Bot is the bot plugin that turns a single AI assistant into a collaborative team. Instead of talking to one generic chatbot, you can spawn specific workers with distinct roles, memories, and personalities. The coolest part is the group chat feature, where you drop multiple bots into a single room and they actually coordinate. They pass work to each other, share context, and even talk to you as a unit. It feels less like using a tool and more like managing a small digital office. Since it runs locally on your machine through DeepSeek Harness, your data stays private.

🆔 @hackernewsgithubprojects

📦 chenwei810/semiconductor_manufacturing_introduction

Introduction to Semiconductor Manufacturing

Grab this open book to finally understand how a plain piece of silicon becomes the tiny chip inside your phone. It skips the heavy physics and instead follows the real journey from raw sand to a finished device, explaining the tricky steps in plain English. You get a clear map of the factory floor, the right words to use, and the practical know-how that engineers rely on every day. It is the perfect starting point if you are new to the industry or just curious about how your favorite gadgets are actually made. Stop guessing and start reading the real deal.

🆔 @hackernewsgithubprojects

📦 slickamogus/silent-hill-decomp

Silent Hill 1 Decomp

A team used AI to decompile the original PlayStation code for Silent Hill 1 and rebuild it into a fully playable port for modern computers. The project is called Silent Hill Decomp, and it is still in active beta. The most surprising part is how much of the game’s inner structure has been recovered, including every major level, the specific monsters that roam them, and even the underlying collision logic that makes the world feel solid.

🆔 @hackernewsgithubprojects