📦 kunchenguid/axi

AXI: A New Paradigm for Agent-Native CLI Tools

AXI is the agent-native interface that streamlines command-line tools for AI agents to achieve higher accuracy at a lower token cost. By applying ten design principles—such as token-efficient output formatting, pre-computed aggregates, and structured error handling—it creates a more ergonomic environment for agents compared to standard CLIs or structured protocols like MCP. The project includes a shared Node.js SDK and various reference implementations, such as tools for GitHub and browser automation. Adopting AXI-compliant designs helps you build apps that agents handle more effectively, ensuring your AI workflows remain fast, cost-efficient, and reliable every time.

🆔 @hackernewsgithubprojects

📦 kunchenguid/lavish-axi

Stop Screenshots: Use Lavish Editor for Agent Artifacts

Lavish is the local-first editor that transforms how you collaborate with AI on generated HTML artifacts. Instead of relying on static screenshots or long text descriptions, this tool opens agent-created HTML directly in your browser. You can pinpoint specific elements or text ranges to provide precise, visual feedback that the agent can immediately address. It functions as a seamless local server with zero cloud dependency, featuring layout audits that catch errors before you even review. Lavish bridges the gap between human oversight and agent output, turning your interactive projects into a true two-way conversation.

🆔 @hackernewsgithubprojects

📦 boltzmannentropy/mimikastudio

Local-First AI Voice Cloning for macOS

MimikaStudio is a local-first application built for macOS with Apple Silicon that enables powerful voice cloning and text-to-speech capabilities directly on your machine. By leveraging advanced engines like Qwen3-TTS and Kokoro, it allows you to clone any voice using as little as three seconds of audio or generate expressive speech from text. The software features an integrated audiobook creator with document support for formats like PDF and Markdown, all while maintaining a private, on-device workflow. With native Metal acceleration, it provides a fast, unified dashboard for managing voice presets and agentic voice cloning jobs, offering a complete professional-grade audio toolkit.

🆔 @hackernewsgithubprojects

📦 kunchenguid/firstmate

Stop Tab-Juggling: Meet Your Firstmate

Firstmate is the automation orchestrator that turns your coding agents into a scalable, coordinated crew. Instead of manually babysitting multiple AI sessions or copy-pasting context between windows, you talk to one single liaison that dispatches tasks in parallel. It handles the heavy lifting by spawning autonomous agents inside individual tmux windows, each running in its own isolated git worktree to prevent conflicts. Whether you need code fixes, project audits, or investigation reports, Firstmate supervises the process and delivers finished pull requests directly to you. It keeps your workspace clean, organized, and focused so you can simply lead the fleet.

🆔 @hackernewsgithubprojects

📦 elliotgao2/handsets

Blazing Fast Android Automation Without Root

Automate Android devices directly from your terminal with record-breaking speed. This high-performance command-line tool executes actions in just a few milliseconds, making it ideal for both human interaction and automated agents. It works by running a lightweight daemon on the device, allowing you to find elements by label, capture screenshots, and manipulate apps without needing root access or complex installations. Whether you are performing mobile testing or building custom automation workflows, the tool offers a seamless experience with an intuitive terminal inspector and robust scripting capabilities, giving you a professional-grade alternative for managing Android devices with ease.

📰 https://news.ycombinator.com/item?id=48219188

🆔 @hackernewsgithubprojects

📦 ivorpad/mercadona-cli

Automate Your Grocery Shopping with this CLI

Manage your grocery shopping directly from the terminal by using this unofficial command line interface for the Mercadona website. This tool lets you search the product catalog, check real-time prices, build a digital cart, and even complete the checkout process programmatically. It solves the need for manual navigation by offering structured data output that is perfect for agents or scripts to track inflation, compare value, or automate your weekly restocking. With built-in budget protections to prevent accidental overspending, this utility brings powerful, automated control to your routine errands while keeping your shopping experience efficient and completely hands-free.

🆔 @hackernewsgithubprojects

📦 sediment1024/physrag

PhysRAG: Physics-Aware Video Generation with RAG

PhysRAG is the generative framework that finally injects real-world physics awareness into AI video production. By using a retrieval-augmented generation approach, this project helps models like Wan2.2 better understand and simulate physical interactions, such as flowing liquids or bouncing objects. It works by retrieving cached physical reference features and injecting them into the denoising transformer process, allowing for more grounded motion sequences. Whether you are aiming to improve technical accuracy in your animations or just want more realistic physics in generated video content, PhysRAG provides the tools to bridge the gap between creative prompting and physical reality.

🆔 @hackernewsgithubprojects

📦 jia-lab-research/blockgtq

Optimize LLM Memory with Block-GTQ

Block-GTQ is the KV-cache quantization tool that optimizes large language model performance through intelligent RoPE-aware bit allocation. By recognizing that certain key-cache blocks carry significantly more signal than others, this project moves away from uniform quantization to strategically shift bit widths toward high-energy components. Using PyTorch and Triton, it calculates energy scores per block to prioritize essential data, resulting in substantial reductions in kernel errors across diverse models. This approach effectively balances memory usage with model accuracy, offering a robust method to streamline deployment while maintaining high performance for long-context tasks and large-scale language processing workflows.

🆔 @hackernewsgithubprojects

📦 yinjige/redi-match

REDI-Match: Fast and Robust Rotation-Equivariant Dense Matching

REDI-Match is the dense matching framework that effectively bridges the gap between vision foundation models and strict rotation equivariance. By distilling powerful semantics into a lightweight, rotation-aware encoder, it solves the challenge of maintaining accurate feature correspondence under severe in-plane rotations. The system uses an entropy-driven spatial alignment module to lock onto orientation, allowing for superior precision without massive data augmentation. It achieves significantly faster performance than current state-of-the-art models while using five times fewer parameters. This project demonstrates that high-performance vision tasks can remain both computationally efficient and geometrically consistent through clever architectural distillation.

🆔 @hackernewsgithubprojects

📦 nikitadoudikov/claude-pulse

Take Control of Claude Code with This Local Dashboard

Monitor your Claude Code sessions using a private, zero-dependency dashboard that runs locally on your machine. This project lets you track live token usage, recover lost sessions, and perform full-text searches across your entire history. You can even approve tool calls directly from your phone through secure notifications, ensuring your workflow remains uninterrupted by crashes or complex setups. By reading your existing session files, it provides an ambient office view and granular spending insights without ever sending data to the cloud. Install the necessary hooks today to gain total visibility and remote control over your coding assistant.

📰 https://news.ycombinator.com/item?id=48612844

🆔 @hackernewsgithubprojects

📦 hao-ai-lab/jetspec

JetSpec: Massive Speedups for LLM Inference

JetSpec accelerates large language model inference by up to nine times through a technique called causal parallel tree drafting. Traditional speculative decoding often struggles to balance drafting speed with accurate token conditioning, but this project solves that by using a specialized draft head that emits a full token tree in a single parallel pass. The target model then verifies this entire tree using a tree-causal attention mask, ensuring the output remains lossless. By combining paged memory management with custom Triton kernels, it provides a high-throughput engine that makes complex reasoning and coding tasks significantly faster.

🆔 @hackernewsgithubprojects

📦 looplj/axonhub

Simplify Your AI Development with AxonHub

Streamline your AI development by using this open-source gateway to access over one hundred different language models through any existing software development kit. It solves the headache of vendor lock-in and complex integrations by standardizing requests into a single, unified format, allowing you to swap between models like GPT, Claude, and Gemini without changing a single line of your code. You can also monitor your usage with real-time cost tracking, built-in failover for reliability, and comprehensive request tracing for easier debugging. Take full control of your infrastructure and optimize your development workflow today.

🆔 @hackernewsgithubprojects

📦 fossprime/steam-controller-auto-charge

Automate Your Controller Charging with Computer Vision

Triton Auto-Charge Vision Tracker is the browser-based tool that uses optical flow to steer your controller onto its charging puck. By leveraging a standard overhead webcam and OpenCV, the application tracks your Triton controller and guides it toward the magnetic dock using haptic pulses. It dynamically adjusts navigation as the controller nears the puck, ensuring a gentle and successful connection. The project even monitors battery levels and voltage in real time through WebHID telemetry, making your charging process completely autonomous. Give your desk setup a high-tech upgrade and let your hardware dock itself with ease.

📰 https://news.ycombinator.com/item?id=48693338

🆔 @hackernewsgithubprojects

📦 danielealbano/android-remote-control-mcp

Control Your Android Device with AI Using MCP

This Android application turns a phone into a remote Model Context Protocol server that allows AI models to interact directly with the device. By leveraging accessibility services and screen capture, the server enables AI to read user interface elements, perform taps, type text, and manage files. The implementation includes automated tunnelling options like Cloudflare and Ngrok to simplify remote connections while keeping token usage optimized. This project effectively bridges the gap between AI agents and physical device operations for research and automation purposes. Exploring this tool reveals how effectively external models can navigate and control complex mobile environments.

🆔 @hackernewsgithubprojects

📦 onemana-soft/onecamp-fe

Build Your Own All-in-One Private Workspace

Manage your entire workflow from a single, self-hosted platform designed to centralize team collaboration. This repository provides a modern Next.js frontend that powers an integrated environment for real-time chat, task management with Kanban boards, and collaborative document editing. By keeping everything on your own infrastructure, you maintain complete ownership of your data while utilizing professional tools like LiveKit for video meetings and Tiptap for rich-text documentation. It is a robust solution for teams looking to replace fragmented services with a unified, customizable, and private workspace. Explore the codebase to deploy your own private command center today.

🆔 @hackernewsgithubprojects

📦 tt-a1i/archify

Turn Natural Language into Beautiful Architecture Diagrams

Archify is the agent skill that finally turns your plain-English system descriptions into beautiful, professional architecture diagrams. Instead of manually drawing boxes and arrows, you simply describe your technical workflow, data flow, or system components to an AI, which then generates a self-contained HTML file. You get an interactive diagram with a built-in dark and light theme toggle and options to export crisp images in multiple formats or copy them directly to your clipboard. It is a fantastic tool for creating polished, high-resolution documentation for your projects, making it easier than ever to visualize complex processes effortlessly.

🆔 @hackernewsgithubprojects

📦 breaking-brake/cc-wf-studio

CC Workflow Studio: Build AI Agent Workflows Visually

CC Workflow Studio is the visual editor that finally lets you design complex AI agent workflows without writing code. This tool provides an intuitive drag-and-drop canvas for orchestrating agents, defining skills, and managing tool integrations, which it then exports into the clean Markdown format that modern AI agents natively understand. Whether you are using the VSCode extension for a seamless interface, the CLI for automation, or the MCP server for external agent connectivity, it solves the headache of manual prompt-guessing. You can design once and deploy across various agent platforms to ensure your AI automations are consistent, refined, and ready to run.

🆔 @hackernewsgithubprojects

📦 riddleling/ios-ocr-server

Turn Your iPhone Into a Private Local OCR Server

iOS OCR Server is the local utility that transforms your iPhone into a powerful, privacy-focused text recognition hub. By leveraging Apple's native Vision Framework, the project allows you to process images entirely on-device without any cloud dependencies or subscription fees. Once the app is running, it exposes a simple REST API that makes it easy to integrate text extraction into your own software or scripts. Whether you are building an automated document pipeline or repurposing an old device into a dedicated recognition node, this tool provides a secure, unlimited, and network-accessible way to handle all your OCR tasks.

🆔 @hackernewsgithubprojects

📦 rohitg00/pro-workflow

Stop Correcting Your AI: Self-Learning Workflows for Claude

Pro Workflow solves the recurring frustration of repeating corrections by embedding a persistent SQLite store directly into your Claude Code sessions. This system enables self-correcting memory, where every correction evolves into a durable rule that automatically loads at the start of new sessions. Beyond fixing mistakes, it integrates a searchable knowledge plane for research wikis and automated development workflows. With thirty-four specialized skills and twenty-two commands, it transforms your coding assistant into a structured engine that remembers your preferences, enforces quality gates, and grows smarter with every task. The takeaway is simple: your development tools should compound their intelligence over time.

🆔 @hackernewsgithubprojects

📦 pmxt-dev/pmxt

Trade Polymarket and Kalshi with One API

PMXT acts as a universal bridge for prediction markets by providing a unified API that supports platforms like Polymarket, Kalshi, and Limitless. It eliminates the frustration of managing disparate data formats and exchange conventions by consolidating everything into a single, consistent interface. You can use it as a hosted service to handle custody and order signing or self-host it to maintain full control over your keys and credentials. Because it is language-agnostic and features ready-to-use SDKs for Python and TypeScript, it allows you to build sophisticated trading tools easily. Ultimately, this project simplifies cross-platform interaction for any prediction market strategy.

🆔 @hackernewsgithubprojects