📦 xzf-thu/audio-interaction

The First Always-On Audio Intelligence Model

Tired of audio models that force you to wait for a full clip before they speak? Meet AudioInteraction, the first unified model designed to break the offline bottleneck. Unlike standard tools that rely on isolated tasks, this system functions as an always-on engine that processes live streams in real time. It can handle transcription, translation, and interactive chatting simultaneously within one continuous session. By deciding for itself when to listen and when to respond, it offers proactive, seamless communication. Experience the next evolution in sound processing and see how this all-in-one model changes how we interact with live audio forever.

🆔 @hackernewsgithubprojects

📦 internscience/mlevolve

MLEvolve: The Autonomous Machine Learning Powerhouse

Imagine an AI that autonomously solves complex machine learning competitions with human-like precision. Meet MLEvolve, a groundbreaking agentic system that dominates the field by automating algorithm design and optimization. By utilizing advanced Monte Carlo Graph Search and multi-agent collaboration, it tackles intricate data science tasks with incredible speed. It even features a powerful experience-driven memory that learns from past attempts to avoid pitfalls and refine strategies. Whether it is crushing benchmarks or driving breakthroughs in mathematical optimization, this system sets a new standard for AI researchers. Check out this project to see the future of automated scientific discovery in action.

🆔 @hackernewsgithubprojects

📦 huawei-csl/kvarn

Boost Your LLM Context with KVarN

Struggling with limited context space in your AI agents? Meet KVarN, a powerful backend for vLLM that transforms your KV-cache efficiency. While typical quantization methods force you to sacrifice speed and accuracy, KVarN delivers three to five times more capacity while maintaining FP16-level precision and actually boosting your throughput. It uses clever variance normalization and Hadamard rotation to optimize memory usage without complex setup. Since it is calibration-free, you simply add one flag to your existing vLLM workflow to get started. Scale your long-context workloads today and unlock higher performance for your most demanding AI models with this efficient, plug-and-play solution.

📰 https://news.ycombinator.com/item?id=48399974

🆔 @hackernewsgithubprojects

📦 texttron/browsecomp-plus

Stop Guessing: How to Actually Benchmark Your AI Research Agents

Evaluating deep research agents is often messy because live web search results are unpredictable and impossible to replicate. That is where this repository comes in. It provides a specialized benchmark that swaps out live, opaque web searches for a fixed, curated library of roughly one hundred thousand human-verified documents. By controlling the data source, it allows you to finally isolate the performance of your retrievers from your language models, leading to fair and transparent comparisons. Whether you are testing new retrieval methods or fine-tuning agents, this tool makes your experiments consistent, reproducible, and truly meaningful.

📰 https://news.ycombinator.com/item?id=48407811

🆔 @hackernewsgithubprojects

📦 rashevskyv/dbipatcher

Localize Your Switch Tools with AI

Ever wish your favorite Nintendo Switch homebrew tool felt more like home? Meet DBI Patcher, an advanced engine that brings professional localization to the popular DBI tool. Using the power of AI, this project provides high-quality translations for over twenty-two languages, ensuring your interface looks perfect with smart alignment and strict validation checks. Whether you need Ukrainian, Japanese, or something else, this engine handles the heavy lifting to keep your UI consistent and readable. It is a brilliant way to make advanced software accessible to everyone. Give your Switch setup a personal touch and enjoy a truly global experience today.

🆔 @hackernewsgithubprojects

📦 denislupookov/altersend

Stop Using Cloud Storage for File Transfers

Tired of slow upload speeds and cloud storage limits when sending your own files? Meet AlterSend, a powerful open-source tool that lets you move data directly between devices without any servers involved. By using peer-to-peer technology, it creates a secure, end-to-end encrypted connection between your phone and computer. There are no accounts to sign up for and absolutely no file size restrictions, so you can transfer anything instantly. Whether you are on Windows, Mac, or mobile, your data stays private and stays yours. Ditch the middleman today and take complete control over how you share your files.

📰 https://news.ycombinator.com/item?id=48408647

🆔 @hackernewsgithubprojects

📦 pemartins1970/open-slap

Take Control with Open Slap: Your Private AI Desktop Assistant

Tired of cloud-only assistants? Meet Open Slap, a powerful desktop-local AI server that puts you in total control. It runs right on your machine, using a React frontend and a FastAPI backend to deliver a seamless experience. You get a full agentic team, multi-LLM support including local options, and robust memory powered by SQLite and RAG. With built-in connectors for your files and tools, plus strict permission settings for security, it intelligently automates your tasks while keeping your data private. It is time to upgrade your workflow with a truly personal and highly customizable AI assistant, so go check it out.

🆔 @hackernewsgithubprojects

📦 chintanpatel24/flint

Build Your Own Local AI Knowledge Base with Flint

Imagine having a digital brain right on your computer that keeps your notes private and powerful. Flint is a secure, local-first knowledge base that lets you organize ideas using Markdown and interactive graph visualizations. What makes it truly special is the built-in AI agent that runs locally on your machine, using tools like Ollama to read your notes and even browse Wikipedia for real-time information. Because everything stays on your device, you get all the benefits of smart, connected intelligence without compromising your privacy. Start building your own personal knowledge network today and take full control of your data.

🆔 @hackernewsgithubprojects

📦 geo-tp/esp32-bit-pirate

Turn Your ESP32 into a Powerful Hardware Hacking Tool

Ever wanted to hack hardware like a pro? Meet the ESP32 Bit Pirate, an open-source firmware that transforms your ESP32 into a versatile multi-protocol tool. Inspired by the legendary Bus Pirate, it lets you sniff, send, and interact with countless digital and radio protocols including I2C, SPI, UART, Bluetooth, Wi-Fi, and even Sub-GHz signals. Whether you are debugging electronics, analyzing radio frequencies, or cloning RFID tags, this tool provides a powerful web-based command-line interface to get the job done easily. Unlock your device's full potential and start exploring the hidden signals around you with this incredible hacking companion today.

📰 https://news.ycombinator.com/item?id=48409306

🆔 @hackernewsgithubprojects

📦 rustykuntz/clideck

Stop Juggling Terminals: Meet CliDeck

Are you tired of juggling multiple terminal windows for your AI coding agents? Stop the chaos with CliDeck, a local dashboard that brings all your agents into one organized browser interface. Instead of getting lost in a sea of panes, you get a clean chat-style layout with live status updates, session resume, and even an autopilot feature that routes work between specialists while you sleep. Everything runs locally on your machine for complete privacy. Whether you need to manage complex projects or control your agents remotely from your phone, CliDeck makes AI multitasking finally feel intuitive and effortless.

📰 https://news.ycombinator.com/item?id=47305108

🆔 @hackernewsgithubprojects

📦 goombalab/raven

Raven: The Future of Efficient Long-Context AI

Ever wonder how AI models can remember massive amounts of information without slowing down? Meet Raven, a powerful linear-time sequence model designed to handle long contexts with incredible precision. Unlike traditional models that update their memory in a dense, uniform way, Raven uses a clever routing memory mechanism. By employing a learned sparse router, it selectively updates only the most relevant memory slots for each token. This approach prevents interference, preserves long-range recall, and keeps your computations fast and efficient. If you are looking for a smarter way to manage memory in your next AI project, Raven is a game-changer.

🆔 @hackernewsgithubprojects

📦 maxmelichov/bluetts

BlueTTS: High-Performance AI Voice Synthesis

Want to generate lifelike speech with incredible speed? Check out BlueTTS, an impressive project designed for high-performance text-to-speech inference. It leverages the power of ONNX Runtime to deliver efficient results, with optional support for NVIDIA CUDA, Intel OpenVINO, or TensorRT acceleration to boost your processing power even further. Whether you are working with English, Hebrew, Spanish, Italian, or German, this tool makes high-quality synthesis accessible for your applications. It handles complex tasks like zero-shot voice conversion and offers both full-precision and quantized models for flexibility. Dive into this repository to start building your own high-speed voice projects today.

🆔 @hackernewsgithubprojects

📦 visual-ai/speed3r

Speed3R: Fast 3D Reconstruction Explained

Ever wonder how to make 3D reconstruction faster? Current models often struggle with slow processing because they rely on heavy, dense attention. Meet Speed3R, a breakthrough in feed-forward 3D reconstruction. By using a clever dual-branch attention mechanism, it mimics traditional keypoint matching to focus only on the most important parts of an image. This smart approach achieves a massive twelve-fold speedup on long sequences while keeping quality high. Whether you are working with large scenes or need efficient modeling, this tool delivers impressive results with a fraction of the usual computational cost. Experience the future of faster 3D modeling today.

🆔 @hackernewsgithubprojects

📦 panniantong/agent-reach

Give Your AI Agent Superpowers with Agent Reach

Tired of your AI agent struggling to access the real-time internet? Meet Agent Reach, the ultimate tool to give your AI eyes and ears across the web. Whether it is reading Twitter threads, summarizing YouTube videos, or navigating Reddit, this project handles the tedious configuration and API hurdles for you. By installing this toolkit, your AI gains the ability to seamlessly interact with platforms like GitHub, Bilibili, and more through simple commands. It is completely open-source, private, and works with popular agents like Claude Code or Cursor. Stop fighting with configurations and finally let your AI agent explore the internet today.

🆔 @hackernewsgithubprojects

📦 vllm-project/vllm-omni

Scale Your Multimodal AI with vLLM-Omni

Want to supercharge your AI model deployment? Meet vLLM-Omni, a powerful framework designed to make serving omni-modality models both fast and efficient. While traditional tools focus on text, this platform extends support to image, video, and audio processing, handling non-autoregressive architectures like Diffusion Transformers with ease. By using advanced memory management and pipeline stage execution, it delivers high-throughput performance across diverse hardware backends. With native support for popular models and an OpenAI-compatible API, it simplifies complex workflows for developers everywhere. Start using vLLM-Omni today to bring your multimodal AI projects to the next level with production-ready speed.

🆔 @hackernewsgithubprojects

📦 anthropics/anthropic-cli

Take Control of Claude Right from Your Terminal

Stop switching between your code and a web browser and start interacting with the Claude API directly from your terminal. The official Anthropic CLI tool simplifies your workflow by letting you execute commands against the Claude platform without writing extra wrapper code. You can easily send messages, create agents, and manage sessions using a clean resource-based structure. It even supports powerful features like passing local files directly into your requests or transforming output data using query syntax. Whether you are building automated scripts or exploring AI capabilities, this tool offers a faster, more professional way to develop.

🆔 @hackernewsgithubprojects

📦 mertcikla/tld

Visualize Your Code Architecture Like a Pro

Are you tired of struggling to visualize complex software architecture? Meet this powerful diagramming tool designed to make understanding your codebase effortless. It offers a standalone, dependency-free binary that combines a polished web interface with a command line tool, perfect for managing diagrams directly from your shell or CI pipeline. You can even use its intelligent agent to automatically generate detailed architecture diagrams from your source code. With features like bidirectional syncing and seamless integration with your editor, it treats your architecture as code, keeping your documentation perfectly in sync. Simplify your workflow and start mapping your systems today.

📰 https://news.ycombinator.com/item?id=48369739

🆔 @hackernewsgithubprojects

📦 lins-lab/masarena

Master Multi-Agent Systems with MASArena

Are you struggling to evaluate how well your AI agents perform? Meet MASArena, a powerful framework built to benchmark single and multi-agent systems with ease. This tool lets you swap agents, tools, and datasets seamlessly while providing visual debugging to inspect every interaction. It even features automated workflow optimization using evolutionary algorithms to boost your agent performance. Whether you are testing math, coding, or complex reasoning, this framework helps you identify exactly why agents fail and how to improve them. Start benchmarking today to unlock the full potential of your AI agents and take your multi-agent projects to the next level.

🆔 @hackernewsgithubprojects

📦 rdi-berkeley/agents-last-exam

Can Your AI Pass the Ultimate Professional Exam?

Are AI agents actually ready for the real world or are they just guessing? Agents Last Exam is a massive new benchmark designed to measure AI performance on long-horizon, economically valuable tasks. Co-led by Berkeley experts, it uses a massive collection of industry-specific workflows and verifiable outcomes to test how well agents handle complex, real-world work. The repository provides an orchestration toolkit called ale-run and includes over one hundred reference tasks across dozens of industries to keep things challenging. It is a vital step toward creating reliable, high-performing agents. Check it out to see if your favorite AI makes the grade.

🆔 @hackernewsgithubprojects

📦 oranger-l/mgsd

Master Visual Spatial Planning with MGSD

Ever wondered how AI models learn to navigate complex visual environments like mazes or behavior tasks? Meet MGSD, an innovative framework designed to help vision-language models master visual spatial planning. By using a clever two-step approach, it starts with perception training to help the model recognize states in images, followed by a technique called modality-gap-aware self-distillation. Here, a text-only teacher guides a visual student to bridge the gap between symbolic data and visual reality. This powerful pipeline allows models to solve tasks like pathfinding and object manipulation more effectively. It is a fantastic resource for advancing robotic intelligence and visual reasoning.

🆔 @hackernewsgithubprojects

📦 rccyx/asryx

Native Linux Voice-to-Text: Meet Asryx

Tired of bloated AI tools constantly running in the background? Discover Asryx, a lightweight, daemonless voice-to-text tool built specifically for Linux. By leveraging the power of whisper.cpp directly in your system, Asryx lets you record audio with a simple key press and instantly transcribes it to your clipboard. It requires no background services, cloud APIs, or complex runtimes to function. Whether you prefer PipeWire or ALSA, this native C++ binary handles everything locally and cleans up after itself automatically. If you want a fast, privacy-focused way to dictate text on your Linux desktop, this tool is the perfect solution.

📰 https://news.ycombinator.com/item?id=48399890

🆔 @hackernewsgithubprojects