📦 cognicore-dev/cognicore-my-openenv

Give AI Agents a Brain with CogniCore

CogniCore gives your AI agents a persistent, searchable memory and a built-in immune system that blocks prompt injections. Unlike standard models that forget everything between runs, this Python framework lets agents recall past experiences by meaning, not just keywords. It even features a time-travel engine to replay decisions or branch into new paths. The best part? It uses nearly ninety percent fewer tokens than alternatives and needs no API keys to start. Check it out if you want your code to actually remember what it learned.

🆔 @hackernewsgithubprojects

📦 ozeranskii/httptap

See Exactly Where HTTP Requests Wait with Httptap

httptap is a Python tool that dissects every HTTP request into its individual phases, showing you exactly where time is lost during DNS resolution, connection, TLS handshake, and data transfer. Instead of just seeing a total load time, you get a clear timeline of each step, making it easy to spot bottlenecks in your API calls. It supports rich visual outputs for interactive troubleshooting, compact summaries for logs, and even automated performance checks to ensure your services meet speed requirements. This is a great way to debug slow network requests without needing complex system tools or deep networking expertise.

🆔 @hackernewsgithubprojects

📦 aml-memory/agent-memory-leaderboard

Agent Memory Leaderboard

Agent Memory Leaderboard is the standardized testing ground that finally makes it possible to compare how well different AI assistants remember your long-term conversations and history. Right now, everyone is building memory systems but using totally different tests, so a high score here means nothing to a competitor over there. This project fixes that chaos by providing a unified protocol where you simply plug in two functions: one for saving new info and another for retrieving old stuff. The platform then runs everything else, from generating answers to grading them, ensuring the only variable changing is the memory system itself.

📰 https://news.ycombinator.com/item?id=49145408

🆔 @hackernewsgithubprojects

📦 mte90/linus-torvalds-skill

Linus Torvalds Skill

This project distills Linus Torvalds’ code review methodology from thirty-eight thousand of his actual emails into a reusable set of rules and a personality profile. It turns thousands of specific Linux kernel complaints into a language-agnostic guide that tells an AI exactly how to prioritize correctness, protect existing users, and reject unnecessary complexity. You get a skill file for strict logic and a soul file that captures his blunt, no-nonsense tone without the swearing unless it is warranted. It is a clever way to capture the mindset of the Linux kernel maintainer for any coding project.

🆔 @hackernewsgithubprojects

📦 fraunhofermevis/radyolo

RadYOLO

RadYOLO lets you find and outline objects in MRI and CT scans in just seconds. While standard models take minutes or even hours to analyze a single 3D volume, this tool runs up to a hundred times faster on a standard computer. It works by treating medical imaging like a game of spot-the-object, using a clever twist on the famous YOLO architecture to spot lesions and organs instantly. This speed is huge for researchers who need to process hundreds of scans without waiting around all day. You get accurate results that match top-tier competitors but with a fraction of the compute power.

🆔 @hackernewsgithubprojects

📦 mumucc01/mojito

MOJITO: Autonomous Driving with Diffusion

MOJITO takes autonomous driving to the next level by using a diffusion model that learns from images, lidar scans, and past trajectories all at once. Instead of just reacting to what is in front of the car, it actually imagines multiple possible futures and picks the smoothest path. It builds on DiffusionDrive and runs on the standard nuPlan dataset, making it a serious contender for end-to-end self-driving control. The code is open source, so you can train it yourself. This is a fascinating look into how generative ai is reshaping how cars navigate the real world. Check it out and see the future of driving.

🆔 @hackernewsgithubprojects

📦 fabioyanezromero/knowledge-graph-builder

Knowledge Graph Builder

Turn messy text into a connected web of knowledge with the Knowledge Graph Builder. This project takes your raw documents and uses artificial intelligence to pull out key facts, linking people, places, and concepts together like a giant map. It even spots when pieces of information are missing and automatically invents the missing links to tie everything together, making sure your data is fully connected. You can watch it work in real time with interactive visuals that let you click and drag the nodes around. It supports both local and cloud tools, so you can run it privately or use the cloud.

🆔 @hackernewsgithubprojects

📦 kaushalt2004/cognicore-my-openenv

CogniCore: Persistent Memory for AI Agents

CogniCore gives AI agents the ability to remember past conversations, so they do not forget everything between sessions. This Python framework works with any model to store experiences and retrieve them by meaning rather than just keywords. It also includes a built-in safety shield that blocks prompt injections and an autonomous coding agent that can fix bugs and open pull requests on its own. The most interesting part is how it connects scattered facts from different sessions to solve complex tasks. This means your code gets smarter over time without needing constant retraining.

🆔 @hackernewsgithubprojects

📦 iamcorey/birth

Birth: The macOS Startup Manager That Sees Everything

Stop guessing why your Mac feels slow and start taking control with Birth, a free open-source tool for macOS that finally unifies your scattered startup items into one clear, readable window. Instead of hiding background processes behind vague system settings, this app gathers every login app, daemon, and background agent into a single view so you can see exactly who installed them, verify their developer identity, and check if they are actually running right now.

🆔 @hackernewsgithubprojects

📦 feigecode/navop

Navop: The All-in-One Developer Workspace

Navop lets you manage every connection and tool you use daily inside one single desktop app. You can query databases, SSH into servers, transfer files, and even control remote desktops without juggling half a dozen different programs. It is built with Rust for speed and includes a smart AI assistant that helps with code and terminal tasks right where you work. Instead of switching between tabs and apps, you keep your entire workflow in one native window that syncs securely. It is the cleanest way to handle complex server work without the chaos.

🆔 @hackernewsgithubprojects

📦 pgsty/silo

Silo: The Open-Source Storage Server That Won't Go Dark

Silo is the community-maintained object storage server that keeps your data safe when the original developers stop supporting it. Think of it as the MinIO server you know, but rebuilt by a group that refuses to let your backups disappear. The creators simply renamed the delivery tools, so while the executable files change, the protocol and your actual data stay exactly the same. It is a fork designed to preserve functionality and security long after the upstream project goes quiet. If you need reliable storage that will not vanish overnight, this is the project to watch. It proves that open-source tools can survive without corporate backing.

📰 https://news.ycombinator.com/item?id=49229591

🆔 @hackernewsgithubprojects

📦 mininglamp-oss/octo-web

Octo Web: AI-Powered Workplace Chat

Build a unified workplace experience by shipping a single codebase to both browsers and desktops while natively integrating AI agents into daily chats. Octo Web lets developers create a modern communication hub that feels just like your favorite messaging app but with built-in artificial intelligence help. It solves the messy problem of maintaining separate apps for different screens by using React and TypeScript once, so teams get a consistent look and feel whether someone is on a laptop or in a web browser. The most interesting part is how it handles AI, letting smart bots jump right into conversations to summarize threads or help with tasks without breaking the flow.

🆔 @hackernewsgithubprojects

📦 wassermanproductions/slate

Slate: The Prompt Studio for AI Filmmaking

Slate is the prompt studio that finally gives you a real pre-production layer for AI filmmaking. Most tools just guess at prompts, but Slate treats them like a professional film set. You build a project bible with your characters and locations, then direct individual shots using specific camera angles, lighting, and movement controls. The app’s local AI brain helps you refine these prompts while keeping strict continuity across your entire story. It even spots inconsistencies in wardrobe or lighting before you generate anything. Instead of generic text boxes, Slate compiles precise instructions tailored for specific generators like Sora or Midjourney.

🆔 @hackernewsgithubprojects

📦 kunagent/kun

Kun: The Local AI Agent Workspace

Kun is the local-first AI agent workspace that finally unifies coding, writing, and design into one desktop application. Most tools force you to juggle separate apps for different tasks, but Kun keeps everything running on your own machine. The single coolest feature is how it handles complex workflows through a visual graph mode. You can literally map out a process with connected nodes for code, text, and automation, watching the AI execute each step in order. It feels less like chatting with a bot and more like directing a tiny, highly capable team.

🆔 @hackernewsgithubprojects

📦 wan-video/wan-animate-2

Wan-Animate-2: Real-Time Character Animation

Get a still photo to dance by feeding it a reference video and letting Wan-Animate-2 do the heavy lifting. This project skips the usual complicated steps of extracting motion data. Instead, it takes your picture and a driving clip straight into a smart neural network that figures out the movement on the fly. The coolest part is the speed. You get a special lightweight version that runs animation in real time, which is pretty wild for generating video. It also lets you tweak the camera angle independently, so the character moves without the background going crazy.

🆔 @hackernewsgithubprojects

📦 praetorian-inc/julius

Julius

Discover exactly which AI server software runs on any network endpoint in seconds. Julius acts as a digital detective, sending harmless HTTP probes to identify services like Ollama, vLLM, or LiteLLM without guessing. Instead of manually checking ports, it matches specific response signatures to pinpoint the infrastructure. This saves security teams hours of tedious testing during assessments. It is a fast, offline tool that turns vague IP addresses into clear service names, making reconnaissance simple and accurate for anyone protecting AI systems today.

🆔 @hackernewsgithubprojects

📦 joelbqz/writer-computer

Writer: The Local Markdown Editor That Respects Your Gitignore

Writer is a fast, local-first desktop markdown editor built with Tauri and React that keeps your documents directly on disk while intelligently respecting your workspace Gitignore rules so hidden files stay hidden. It turns your local folder into a powerful writing space with features like multi-window support, extended markdown rendering for tables and diagrams, and a clean sidebar that handles drag-and-drop organization. Unlike cloud editors, it works offline and syncs with your existing file system, making it perfect for Obsidian vaults or personal wikis. The app is lightweight, signed for macOS, and focuses on keeping your workflow simple and local.

📰 https://news.ycombinator.com/item?id=49231821

🆔 @hackernewsgithubprojects

📦 patilshreyas/debroid

Debroid: AI Android Debugging

Debroid lets an AI coding assistant debug a live Android app just like a human developer using Android Studio, but without needing a screen or mouse. Traditional AI tools are blind to runtime details because they cannot click through a graphical interface to inspect memory or pause code. This project bridges that gap by acting as a headless translator between the AI’s text terminal and the Android virtual machine. It allows an agent to autonomously set breakpoints, catch exceptions, and even mutate variable values in real time.

📰 https://news.ycombinator.com/item?id=49197738

🆔 @hackernewsgithubprojects

📦 yegors/degdid

How to Delete Microsoft's Hidden Device ID

The repository 'degdid' reveals that Windows assigns a persistent, server-generated device identifier even when you only use a local account. This tool deletes all existing copies of that identifier from your system and blocks the network path that Microsoft uses to mint new ones. It creates a hard gate preventing the computer from acquiring a unique global device fingerprint. The project focuses entirely on this specific identity mechanism rather than general telemetry. It is designed for personal, unmanaged Windows installations. The script clears registry entries and blocks key Microsoft login servers to ensure the identifier does not reappear after reboot.

🆔 @hackernewsgithubprojects

📦 rocm/spur

Spur: The Modern Slurm Alternative

Spur is a job scheduler written entirely in Rust that lets you run your existing Slurm scripts without changing a single line of code. It replaces the complex legacy architecture with a single controller that handles both scheduling and accounting, using a built-in Raft protocol to keep your data safe even if the server crashes. Instead of relying on fragile TCP networks, it creates a secure WireGuard mesh between all your machines, making cluster communication fast and private. It is designed to be GPU-first, meaning it understands how to manage graphics cards intelligently rather than just treating them as generic resources.

📰 https://news.ycombinator.com/item?id=47927837

🆔 @hackernewsgithubprojects

📦 alibaba/aacr-bench

AACR-Bench: The Code Review Scorecard

AACR-Bench is the multilingual benchmark that finally tests whether AI code reviewers actually understand your entire project structure. Most tools just glance at changed files in isolation, but this dataset forces models to read the whole repository, tracking how a single edit ripples through fifty different open-source projects across ten languages like Python, Go, and Rust. It captures real pull requests where eighty senior engineers worked alongside artificial intelligence to label bugs and security flaws with pinpoint precision. The result is a rigorous scorecard measuring how accurately an AI can locate specific code lines and spot context-heavy issues that a human would catch.

🆔 @hackernewsgithubprojects