📦 shawn0728/opensearch-vl

Build Your Own Multimodal Search Agent with OpenSearch-VL

Ever wondered how to build your own advanced AI research agent that can see, search, and reason? Meet OpenSearch-VL, a completely open-source toolkit designed to train multimodal agents that operate in a closed loop. Unlike standard models, these agents inspect images, use web search tools, and gather evidence before answering. This repository provides the full recipe, including specialized training datasets and a fatal-aware reinforcement learning algorithm to handle tool errors effectively. Whether you are interested in agentic supervised fine-tuning or scaling up with deep reinforcement learning, this project gives you the complete blueprint to create your own powerful search agent.

🆔 @hackernewsgithubprojects

📦 andyhuo520/aetherviz-master

Turn Any Subject Into an Interactive 3D Lesson

Ever wish complex topics were easier to visualize? AetherViz Master is an innovative educational tool that transforms abstract concepts into immersive 3D interactive webpages. Whether you are studying physics, chemistry, or mathematics, this project uses Three.js and SVG rendering to create dynamic visual simulations based on your input. It automatically identifies your chosen subject and generates high-quality, responsive 3D environments with embedded UI controls for a hands-on learning experience. By bridging the gap between static text and interactive models, it makes learning intuitive and engaging. Dive into the repository to start building your own custom visualizations today.

🆔 @hackernewsgithubprojects

📦 strands-agents/shell

Give Your AI Agent a Shell Without Risking Your Machine

Want to give your AI agents the power of a shell without compromising your system security? Meet Strands Shell, a lightning-fast, Bourne-compatible shell built to run entirely in-process. It solves the risk of giving agents machine access by providing an isolated, sandbox-ready environment. With over 50 built-in commands like grep, curl, and jq, it executes instantly without the overhead of containers or system calls. You define exactly what files and URLs your agent can reach, ensuring total control and security. Start sandboxing your agent loops today and keep your machine keys safely tucked away for complete peace of mind.

📰 https://news.ycombinator.com/item?id=48572485

🆔 @hackernewsgithubprojects

📦 yongaifadian1/nvmos

Is Your AI Speech Sounding Natural? Meet NVMOS

Have you ever wondered if your synthetic speech sounds truly human? Evaluating non-verbal vocalizations like coughs, laughs, or sighs is notoriously difficult for standard AI. That is where NVMOS comes in. It provides an automated way to predict the perceptual quality of these specific vocal events in your audio files. By using advanced cross-attention modeling, this tool assigns a quality score from zero to five, helping you identify if your AI-generated sounds feel natural or robotic. It is a powerful way to refine your audio projects, ensuring your synthetic voices sound more lifelike and authentic than ever before.

🆔 @hackernewsgithubprojects

📦 knostic/openant

Automate Security with OpenAnt

Tired of dealing with endless false positives in your security scans? Meet OpenAnt, an open source, LLM-based vulnerability discovery tool designed to help developers find and verify real security flaws. By using a clever two-stage pipeline, it first detects potential issues and then launches automated attacks to confirm which vulnerabilities are actually exploitable. This approach helps you focus on what truly matters by separating real risks from noise. Whether you use Go, Python, JavaScript, or C, OpenAnt provides a configurable way to secure your code. Take control of your security pipeline and start proactively defending your projects today.

🆔 @hackernewsgithubprojects

📦 zlab-princeton/ceobench-src

Can AI Agents Run a Startup for 500 Days?

Ever wonder if an AI can master the long game of running a business? CEO-Bench is a challenging simulation that puts AI agents in the driver seat of a startup for 500 days. The agents must navigate a complex, evolving market, manage company resources, and handle competitors to maximize their ending cash. It provides a realistic environment where agents use a programmable interface to make strategic decisions, providing a unique look at how AI handles long-term planning and business logic. It is a fascinating test of true agent autonomy and strategy. See if your favorite model can survive the business grind.

🆔 @hackernewsgithubprojects

📦 zleap-ai/sag-benchmark

Reproduce SAG Benchmark Results with Ease

Want to see how your retrieval systems truly stack up against the latest paper benchmarks? This repository provides the official reproduction code for the SAG benchmark, allowing you to replicate performance scores across datasets like HotpotQA and MuSiQue. It streamlines the evaluation process by providing tools for dataset uploads, index initialization, and running multi-strategy searches, including atomic entity-first expansion. By using containerized services for data management and tracking, it ensures your testing remains consistent and professional. Check out this project to validate your own RAG performance and push your retrieval accuracy to the next level.

🆔 @hackernewsgithubprojects

📦 epicgames/lore

Lore: The Next-Gen Version Control System Explained

Ever wonder how to handle massive projects that combine complex code with huge binary assets without slowing down? Meet Lore, an open-source, next-generation revision control system built by Epic Games. Designed for incredible scale, Lore uses a centralized, content-addressed architecture powered by Merkle trees and immutable revision chains. It optimizes performance for large teams by utilizing chunked storage and on-demand data hydration, meaning you only download what you actually need. Whether you are a developer or an artist, Lore provides a reliable, high-performance source of truth for your creative work. Dive into this growing ecosystem and start versioning smarter.

🆔 @hackernewsgithubprojects

📦 nashsu/llm_wiki

Build Your Own Automated Personal Knowledge Wiki

Imagine having a personal knowledge base that builds and organizes itself automatically. This desktop application turns your documents into an interlinked wiki, moving beyond basic search by using intelligent agents to process your files into a persistent, searchable network. It features a powerful knowledge graph, automated community detection, and two-step reasoning to ensure your information stays connected and current. Whether you are managing research or personal notes, it keeps your data structured without the constant manual effort of traditional systems. Start building a smarter digital brain today and watch your collective knowledge grow as your library expands.

🆔 @hackernewsgithubprojects

📦 thetom/llama-cpp-turboquant

Supercharge Your LLM Inference with TurboQuant

Want to run large language models more efficiently on your own hardware? This repository is a specialized fork of the popular engine for C and C plus plus, designed to push the boundaries of inference performance. It integrates an advanced codec stack that uses clever techniques like rotated polar quantization to significantly compress your KV cache and model weights. Whether you are using Apple Silicon, NVIDIA, or AMD hardware, this tool helps you reclaim VRAM and speed up your workflow. It is a powerful way to handle long contexts and complex models without sacrificing the quality of your output.

🆔 @hackernewsgithubprojects

📦 botchetdig/workout-gate

Stop Procrastinating: The AI Coding Tool That Forces You To Work Out

Are you tired of sitting at your desk for hours while your AI does all the heavy lifting? Meet Workout Gate, a clever tool for Claude Code that literally blocks your prompts until you complete a physical challenge. Using your webcam and pose detection technology, it forces you to perform push-ups or squats before your code will execute. You cannot simply skip it, as session-persistent debt ensures your reps are tracked until you finish. With customizable modes, streak tracking, and a built-in dashboard, it is time to make sure that you are working just as hard as your AI.

📰 https://news.ycombinator.com/item?id=48558998

🆔 @hackernewsgithubprojects

📦 mrgoonie/claudekit-skills

Supercharge Your AI Agents with ClaudeKit Skills

Stop struggling with generic AI responses and start building professional-grade automation with ClaudeKit Skills. This repository offers a massive collection of specialized agent workflows designed to help Claude handle complex, multi-step tasks with total precision. Whether you are performing advanced backend development, mastering context engineering for your agent architectures, or processing multimodal media, these organized skills provide the necessary guardrails and technical references to get the job done right. You can even use simple commands to add these plugins directly to your development environment. Unlock your AI's true potential and streamline your entire coding workflow today.

🆔 @hackernewsgithubprojects

📦 haowjy/creative-writing-skills

Turn AI into Your Personal Creative Writing Partner

Struggling to finish that novel or stuck in a creative rut? This repository offers a powerful collection of specialized AI agents and craft skills designed to transform your writing process. By integrating these tools into your workflow, you can brainstorm plot ideas, maintain character continuity, and generate prose that perfectly matches your unique voice. The system even simulates reader reactions and provides structured critiques to refine your work from the first idea to the final draft. Stop staring at a blank page and let this comprehensive AI toolkit help you write, revise, and polish your stories with professional precision.

🆔 @hackernewsgithubprojects

📦 tongxuluo/gamecraft-bench

Can AI Build Your Next Favorite Video Game?

Ever wondered if AI could actually build a playable video game from scratch? Meet GameCraft-Bench, an innovative research project designed to test if coding agents can transform simple text instructions into fully functional, interactive games. Unlike basic coding challenges, this benchmark requires agents to generate complete Godot projects featuring working mechanics, scenes, and assets that stand up to real-time player input. With 140 unique tasks across 15 different game genres, it reveals that while modern AI is getting smarter, creating a polished, end-to-end game remains a massive hurdle. It is the ultimate test for the future of automated game design.

🆔 @hackernewsgithubprojects

📦 weichenfan/spectral_forcing

Stop Wasting Compute: Boost Your Diffusion Models with Spectral Forcing

Are your diffusion models wasting precious compute trying to learn from pure noise? Researchers have discovered that standard pixel-space denoisers often struggle to separate signal from noise, leading to inefficient capacity allocation. To solve this, they introduced Spectral Forcing, a clever, parameter-free operator that applies a time-conditional low-pass mask to the noisy input before it hits the model. By filtering out high-frequency noise early in the generation process and gradually revealing the full spectrum, this approach forces the model to focus only on essential data. It is a simple, plug-and-play adapter that makes your diffusion models smarter and more efficient.

🆔 @hackernewsgithubprojects

📦 mizhenxing/one4d

One4D: Revolutionary Unified 4D Generation and Reconstruction

Ever wondered how to bring static images or sparse frames to life in stunning four dimensions? Meet One4D, a powerful unified framework designed to tackle both generation and reconstruction tasks with incredible flexibility. Whether you are starting from a single image, a full video, or just a simple text prompt, this tool uses unified masked conditioning to produce high-quality results. By employing decoupled LoRA control, it separately processes RGB frames and pointmaps to ensure pixel-level consistency without interference. It is a game-changer for anyone looking to create realistic 4D scenes efficiently. Dive into this technology and start transforming your creative vision today.

🆔 @hackernewsgithubprojects

📦 ruc-datalab/coda-bench

Can AI Agents Really Handle Big Data? Meet CoDA-Bench!

Ever wonder if your favorite AI code agent can actually handle real world data tasks? Meet CoDA-Bench, a powerful benchmark designed to test AI performance on data-intensive challenges. Unlike simple coding tests, this tool forces agents to navigate complex Linux file systems, discover relevant data across hundreds of files, and write code to solve messy, real-world analytical problems. With over one thousand tasks sourced from Kaggle, it provides a rigorous sandbox for evaluating discovery and execution accuracy. If you are building the next generation of intelligent agents, this is the ultimate test to see if they are ready for production.

🆔 @hackernewsgithubprojects

📦 blockrunai/awesome-finance-mcp

Supercharge Your AI Agents with Finance MCP Servers

Are you building AI agents that need to handle real-world finance or trading tasks? This incredible curated collection offers a massive library of Model Context Protocol servers designed specifically for financial operations. Whether you need to pull live stock market data, perform technical analysis, manage crypto portfolios, or automate banking workflows, this list has you covered. It solves the problem of finding reliable integrations by organizing everything from DeFi swaps to tax preparation tools in one place. Explore these professional-grade servers today to give your AI the specialized financial capabilities it needs to succeed in the market.

🆔 @hackernewsgithubprojects

📦 yu-lin-li/dycon

Stop Overthinking: Optimize AI Reasoning with DyCon

Ever wonder if your AI is spending too much time overthinking simple problems? Meet DyCon, a clever new approach that brings dynamic control to complex reasoning models. Instead of using static estimates, DyCon monitors the model's internal state in real-time, allowing it to wrap up easier tasks early while giving tougher ones the space they need. By fitting a lightweight regressor to hidden step embeddings, it achieves training-free efficiency without sacrificing accuracy. It is a smart way to cut down on redundant tokens and streamline your inference process. Give your models the flexibility they deserve with DyCon.

🆔 @hackernewsgithubprojects

📦 aws-devtools-labs/aws-blocks

Build Full-Stack AWS Apps Faster with AWS Blocks

Struggling to manage complex AWS infrastructure while keeping your application code in sync? AWS Blocks is a powerful backend toolkit that simplifies full-stack development by offering self-contained building blocks for your projects. Instead of wrestling with configurations, you can pick specific components like databases, authentication, or AI agents and compose them directly into your workflow. The best part is that you can run and test everything locally without needing an AWS account, then deploy the exact same code to production. It even provides native client support for Kotlin, Swift, and Dart, ensuring end-to-end type safety for your entire application.

📰 https://news.ycombinator.com/item?id=48567837

🆔 @hackernewsgithubprojects