📦 auriti-labs/geo-optimizer-skill

Geo Ready GEO Optimizer

Check whether AI search engines like ChatGPT, Perplexity, and Google AI Overviews are actually citing your website. The geo-optimizer-skill toolkit runs a quick audit on your site to score its AI search readiness from zero to one hundred. It scans your robots.txt, checks for an llms.txt file, and analyzes your schema markup to make sure AI bots can easily crawl and quote your content. You can even run live queries to see if your brand gets mentioned when users ask questions. It is a brilliant way to make sure your business does not become invisible in the age of AI search.

🆔 @hackernewsgithubprojects

📦 unimpor/t3

t3

AI assistants struggle when they do not have all the facts, often making bad moves because they lose track of what they actually know. An open-source project called t3 fixes this by training language models to actively seek out missing information and maintain an accurate picture of their environment during multi-step tasks. By introducing clever training methods that reward models for uncovering new details and penalize them for repetitive actions, it stops agents from getting trapped in unproductive loops. If you want to see how reinforcement learning can build smarter agents that actually know how to ask the right questions, this codebase is well worth a look.

🆔 @hackernewsgithubprojects

📦 gtrg55/trellis2-mlx

Trellis 2 MLX

Trellis 2 MLX is the open-source pipeline that finally brings high-resolution, image-to-3D generation natively to Apple Silicon. Instead of relying on power-hungry, cloud-hosted Nvidia graphics cards, this project runs Microsoft's massive four-billion-parameter model locally on your Mac using the MLX framework. The coolest part is how it handles textures. While most local generators produce messy, semi-transparent points, this tool uses smart chunked sampling to construct incredibly detailed, solid 3D meshes with fully baked color and roughness. You can export complete, textured objects straight into Blender or game engines without needing a massive server rack.

🆔 @hackernewsgithubprojects

📦 homemadegarbage/selfrisingrobot

SelfRisingRobot

An incredibly simple two-axis toy robot can learn to stand back up on its own after being knocked down in any direction. This project, called selfrisingrobot, makes reinforcement learning accessible by bridging virtual simulation and physical hardware. It uses a physics engine to train a virtual model of a tiny robot with just two micro servos. Once the computer brain learns to wiggle and push itself upright in the simulator, that identical code is flashed onto a real microcontroller to run on a physical, three-dimensionally printed robot. It is a fantastic hands-on demonstration of simulated intelligence translating perfectly into physical action.

🆔 @hackernewsgithubprojects

📦 thewakesystems/riverone

RiverONE

Compress multimodal AI models to a fraction of their size by treating neural network weight reduction as a simulated quantum computing problem. RiverONE runs entirely on classical hardware, translating quantum mechanics concepts into practical machine learning pipelines. It collapses continuous model weights into discrete, quantum-inspired codebooks, entangles adjacent layers in vision transformers so they share parameters through simulated unitary gates, and recovers lost accuracy using an optimization loop modeled after the variational quantum eigensolver. It is a wildly creative way to shrink heavy vision-language models down to pocket-sized, runnable software.

🆔 @hackernewsgithubprojects

📦 map-raysup/raysup

RaysUp Feature Upsampling

AI models generate incredibly smart feature maps, but they are often too blurry and low-resolution to be useful for precise visual tasks. RaysUp solves this problem by using a clever geometry-aware ray representation that upsamples these complex neural features with extreme precision. Instead of relying on heavy, slow neural networks to guess the missing pixels, this method uses lightweight geometric rays to guide the upsampling process. This makes the entire operation ultra-light and incredibly fast, while keeping the structural details perfectly sharp. It is a massive win for tasks like semantic segmentation and video tracking, proving that we can get high-definition AI features without the massive computational cost.

🆔 @hackernewsgithubprojects

📦 shootthesound/comfyui-angelo

ComfyUI Angelo

The comfyui-angelo tool completely replaces your complex, friction-heavy image editing workflows with a single, highly interactive node. Instead of jumping between mask editors, pad nodes, and separate samplers to fix a detail, you simply click, paint, or drag directly on your image preview to edit. It lets you refine specific areas immediately, auto-detect faces for quick fixes, and expand your canvas with smart outpainting while keeping the rest of your image bit-exact. You can load and restore old photos using a simple slider that perfectly balances original details with fresh, high-resolution textures. It is the ultimate hands-on playground for precise image refinement.

🆔 @hackernewsgithubprojects

📦 tophant-ai/clawvault

ClawVault

Your AI assistant might be accidentally leaking your private API keys or sensitive documents to external servers. That is why clawvault acts as a local security gateway for your artificial intelligence tools. It runs as a local proxy that intercepts traffic between your apps and major AI providers, catching hidden credentials, dangerous system commands, and personal data before they escape your machine. It even swaps out your private secrets with temporary placeholders and safely restores them when the model replies. It is like having a digital firewall that keeps your smart assistants on a tight leash without breaking their workflow.

🆔 @hackernewsgithubprojects

📦 lance0/ttl

ttl

The modern network diagnostic tool ttl completely redefines how you troubleshoot connectivity by showing you when a router is actually rate-limiting responses rather than dropping your traffic. Written in Rust, this fast utility combines traceroute and path monitoring into an interactive terminal display loaded with real-time stats. It traces multiple destinations simultaneously, detects transparent proxies, and uses binary search to find maximum packet sizes to pinpoint VPN connection hangs. It even alerts you to routing instability and identifies internet exchanges on the fly. It is the ultimate tool for developers and network enthusiasts who want to see exactly how their data travels across the web.

🆔 @hackernewsgithubprojects

📦 agentmemoryworld/awesome-agent-memory

Awesome Agent Memory

AI bots are finally moving past short conversations and entering a phase where they actually need a working memory to be useful. That is where awesome-agent-memory comes in, collecting over two hundred studies on how to give AI a brain that lasts. Instead of letting bots forget everything the moment you close a chat, this project maps out how they can store long-term experiences, remember your personal preferences, and build actual skills over time. It is basically the ultimate master list for anyone trying to build an AI helper that does not have the memory of a goldfish.

🆔 @hackernewsgithubprojects

📦 anmolsaini16/mapcn

mapcn

You can now drop beautiful, fully interactive maps into your React apps with a single terminal command. Under the hood, mapcn packages the heavy-duty rendering power of MapLibre GL into clean, declarative components that styled beautifully using Tailwind CSS and play perfectly with shadcn UI. This solves the massive headache of wrestling with bulky map libraries, complex configurations, and mismatched themes. Instead of struggling for hours, you get gorgeous, theme-aware maps with pre-built support for animated arcs, clustering, and custom popups right out of the box. It is the absolute easiest way to make data maps look professional.

🆔 @hackernewsgithubprojects

📦 tuannvm/codex-mcp-server

Codex MCP Server

Supercharge your local Claude Code sessions by giving Claude direct access to OpenAI's powerful Codex capabilities. This clever bridge integrates the Codex command line directly into your Model Context Protocol workspace, letting you request deep code reviews, run automated refactoring tasks, and perform web searches without ever leaving your terminal. If you are working on a tricky feature, you can spin up structured, multi-turn conversations with specific AI models, and it will automatically handle session context behind the scenes. It completely changes how you interact with your code by bringing OpenAI's smartest developer tools straight to your favorite command line agent.

🆔 @hackernewsgithubprojects

📦 derssa/torollo

Torollo

Build and run complete backend architectures on an interactive visual canvas powered by real Docker containers on your own machine. Torollo lets you drag and drop components like databases, load balancers, and servers, then wire them together to see exactly how data moves. Instead of studying abstract diagrams, you get a hands-on playground where every node on the screen is a live container running locally. You can open a terminal, configure security rules, and watch mock traffic flow through your setup in real-time. It is the perfect local-first sandbox to master networking and system design without paying for cloud bills.

📰 https://news.ycombinator.com/item?id=48659847

🆔 @hackernewsgithubprojects

📦 iatomicreactori/csgotrading

CSGOTrading

You can now run a coordinate team of artificial intelligence agents to trade virtual Counter Strike items for you. This project builds a smart multi-agent framework that treats the gaming market like a real financial exchange. Instead of guessing skin prices, the system deploys specialized digital analysts to scan price charts, scrape community sentiment on Reddit, and monitor Steam news. A central coordinator agent dynamically decides which analysts to trust based on current market conditions, while a portfolio manager agent calculates exact position sizes and factors in transaction fees. It is a fascinating way to see LangGraph and modern language models conquer an unconventional economy.

🆔 @hackernewsgithubprojects

📦 metareflection/guardians

Guardians

Stop prompt injection attacks before they even start by separating code from data in your AI agent workflows. This clever python tool solves the classic security problem where an agent reads an untrusted email and gets hijacked by hidden instructions. Instead of letting the AI dynamically call tools on the fly, it forces the model to generate a structured plan upfront using safe placeholders instead of actual data. A local verifier then runs static checks like taint analysis and theorem proving against your security rules, ensuring malicious commands never execute. It is the ultimate proactive shield for autonomous agents.

📰 https://news.ycombinator.com/item?id=47700126

🆔 @hackernewsgithubprojects

📦 ythx-101/ask-search

ask-search

Give your local AI agents the power to search the live web without paying a single cent in API fees or leaking your private queries to third-party servers. The ask-search project acts as a clever, self-hosted bridge that connects tools like Claude Code or local agents to SearxNG, a private search engine pulling results from Google, Bing, and Brave. You get clean markdown lists of top search hits directly in your terminal, and you can even plug it in as a standard model context protocol server. It is a brilliant, zero-cost way to keep your agents smart, fast, and entirely private.

🆔 @hackernewsgithubprojects

📦 openosint/openosint

openosint

You can now run a full digital investigation using nothing but plain English. A new tool called openosint connects popular artificial intelligence models like Claude and GPT-4 directly to eighteen professional intelligence-gathering tools, letting you search for public information automatically. Instead of manually copying and pasting data between different websites, you just tell the system to investigate a target, like an email address. The assistant automatically queries data breaches, looks up social media accounts across hundreds of platforms, checks domain records, and builds a clean markdown report of the findings. It prevents hallucinations by using the AI purely to decide which real, local code tools to run next based on...

🆔 @hackernewsgithubprojects

📦 tzachbon/smart-ralph

Smart Ralph

You can now run a complete, self-contained software development team right inside your terminal. A clever Claude Code and Codex plugin called smart-ralph takes your loose feature ideas and instantly translates them into structured specifications, architectural designs, and concrete task lists. Instead of getting overwhelmed trying to write a massive codebase all at once, this tool acts like a disciplined developer that executes your project one single task at a time, refreshing its memory context after every step. It keeps your AI agent focused, prevents performance degradation on large files, and brings sanity to automated coding. Give it a try to build your next feature step-by-step.

🆔 @hackernewsgithubprojects

📦 coasty-ai/open-cowork

open-cowork

This open-source app lets you delegate real computer tasks to an AI coworker and watch it control your actual desktop screen live. Built as a self-hosted alternative to proprietary agents, open-cowork lets you hand over tasks like moving files or researching flights, streaming every single click and keystroke to your browser or phone while pausing for your manual approval. The coolest part is its flexible model support, which lets you plug in your own API keys or run a vision-capable local model completely offline for zero cost. It is a brilliant, secure way to play with desktop automation without giving up control or breaking the bank.

📰 https://news.ycombinator.com/item?id=48523556

🆔 @hackernewsgithubprojects

📦 pengbinghui/pipeline-math

pipeline-math

Solve unsolved math problems using artificial intelligence by deploying a smart pipeline that pairs a prover with a verifier. The pipeline-math project uses advanced language models to generate mathematical proofs for notoriously difficult, open problems in fields like commutative ring theory and learning theory. The human contributors then polish these solutions and use an automated agentic system to translate them into Lean, a computer language that rigorously double-checks the math for accuracy. It is a fascinating glimpse into how researchers are using AI pipelines to push the boundaries of human knowledge and automate mathematical verification.

🆔 @hackernewsgithubprojects