📦 davide-desio-eleva/kirograph

Supercharge Your AI Coding with KiroGraph

Are your AI coding tools wasting time and context on redundant searches? Meet KiroGraph, a powerful semantic knowledge graph built specifically for Kiro. By using Tree-sitter to parse your codebase into a local SQLite database, it gives your AI instant access to symbol relationships, call graphs, and type hierarchies without constant file scanning. It solves the problem of bloated tool calls, keeping everything 100% local and private. With built-in features like impact analysis, security vulnerability tracking, and architecture insights, you get faster, more accurate responses on complex tasks. Install it today to see how much more efficient your development workflow becomes.

📰 https://news.ycombinator.com/item?id=48220335

🆔 @hackernewsgithubprojects

📦 libertywing/flashmemory-deepseek-v4

Slash Your LLM Memory Costs with FlashMemory

Struggling with the massive memory demands of long-context AI models? Meet the FlashMemory Retriever, a lightweight tool designed to sparsify the DeepSeek-V4 compressed KV-cache. This project solves the problem of heavy GPU memory usage by predicting exactly which cache chunks are necessary for upcoming tokens. By keeping only the most relevant ten to fifteen percent of data on-device and offloading the rest, it maintains high-performance reasoning while drastically reducing memory overhead. It is a brilliant way to optimize resource efficiency without sacrificing accuracy, giving you a smarter, leaner approach to handling ultra-long sequences in your own AI workflows.

🆔 @hackernewsgithubprojects

📦 genalyu/3spo

3SPO: A Breakthrough for Fine-Grained AI Training

Are you struggling with sparse rewards when training large language models? Meet 3SPO, a powerful framework designed to bridge that gap. By converting simple trajectory outcomes into detailed, step-level supervision, 3SPO uses a dynamic state-score mechanism to prioritize challenging states and guide model learning more effectively. It helps you avoid wasting resources on overly difficult tasks while maximizing the impact of every training step. With built-in support for adaptive rollouts and compatibility with popular engines like vLLM and SGLang, 3SPO gives you the tools to refine complex model performance. Try it out to take your reinforcement learning experiments to the next level.

🆔 @hackernewsgithubprojects

📦 qinghonglin/data2story-skill

Turn Any Dataset Into a Verifiable Story with This AI Agent

Ever wanted to turn raw data into a compelling, evidence-backed story without doing all the heavy lifting? Meet this clever data journalist agent that acts like a newsroom in a box. By orchestrating a pipeline of seven specialized roles, including a detective, analyst, designer, and programmer, it researches context, profiles data, and builds a fully traceable multimodal article. Every sentence produced is verifiable, meaning you can trace findings directly back to their source. Whether you are creating charts, videos, or music, this tool helps you build reproducible, professional-grade stories automatically. Take your data storytelling to the next level.

🆔 @hackernewsgithubprojects

📦 haodong-lei-ray/pathspec

Speed Up Text-to-Image Generation with PathRelax

Struggling with slow auto-regressive text-to-image generation? Meet PathRelax, a powerful repository focused on accelerating this process through Parallel-Path Relaxed Speculative Jacobi Decoding. This project provides researchers and developers with a suite of benchmark scripts to evaluate performance across popular datasets like MSCOCO and PartiPrompts. By leveraging specific techniques like static tree structures for draft token generation, it helps optimize efficiency in complex generative tasks. If you are looking to push the boundaries of model inference speed, these evaluation tools offer a practical way to test and refine your configurations for high-quality image output.

🆔 @hackernewsgithubprojects

📦 alibaba-flyai/flyai-skill

Plan Your Next Trip Directly From Your Terminal

Stop switching between travel sites and your coding environment. FlyAI is a powerful skill that brings real-time travel search directly into AI agents like Claude Code. Whether you need to compare flight prices, find hotels near specific attractions, or search for train tickets and concerts, this tool handles it all using natural language. It connects your terminal to the vast travel inventory of Fliggy, providing structured data you can easily filter or analyze. Stop juggling browser tabs for your next vacation plan and let your AI assistant handle the logistics for you starting today.

🆔 @hackernewsgithubprojects

📦 zebbern/claude-code-guide

Master Claude Code with 100+ Specialized Agents

Stop struggling with generic AI prompts and start working like a pro using the Claude Code Guide. This comprehensive resource offers a massive library of over one hundred specialized workspace agents designed to handle specific technical tasks. Whether you need a backend developer, a security auditor, or a test architect, these pre-configured agents provide the expertise you need directly within your coding environment. By leveraging structured workflows and expert skills for everything from database optimization to accessibility testing, you can significantly level up your development speed and code quality. Dive into this guide to unlock your true power-user potential.

🆔 @hackernewsgithubprojects

📦 george-ling3/blue

Stop Wasting Compute: Smarter AI Driving with BLUE

Are your vision-language models for autonomous driving working too hard? Often, these models generate unnecessary language reasoning that slows down performance and can even cause errors. The BLUE repository tackles this by introducing a compact zero-point-one-one million parameter gate that intelligently decides frame by frame when language generation is actually useful. By predicting driving actions only when needed, this approach boosts performance while significantly reducing inference latency. If you are building autonomous systems and want more efficiency without sacrificing accuracy, this project offers a smarter way to handle complex reasoning tasks in real time.

🆔 @hackernewsgithubprojects

📦 sumanth077/ai-engineering-toolkit

The Ultimate Toolkit for AI Engineers

Are you struggling to navigate the overwhelming landscape of AI development tools? This repository is a powerhouse for AI engineers, acting as a massive curated directory of over one hundred essential libraries and frameworks designed for building Large Language Model applications. Whether you need vector databases like Pinecone for data retrieval, agent frameworks such as AutoGen for autonomous workflows, or fine-tuning tools like Unsloth to optimize performance, this collection has it all. It saves you valuable time by centralizing production-ready resources in one place, helping you build, deploy, and scale smarter AI systems today.

🆔 @hackernewsgithubprojects

📦 noopapp/noop

Take Full Control of Your WHOOP Data Offline

Tired of locked-in health data and subscription fatigue? Meet NOOP, a powerful, local-first companion for your WHOOP strap that lets you reclaim your biometric stream. Instead of relying on the cloud, NOOP pairs directly with your device over Bluetooth to store heart rate, recovery, and sleep data securely on your own hardware using SQLite. You can even import your existing history for a complete, private dashboard with transparent, on-device analytics and helpful automations. It is completely account-free, subscription-free, and open for you to explore. Own your metrics and keep your health data truly yours by running it all locally today.

📰 https://news.ycombinator.com/item?id=48476432

🆔 @hackernewsgithubprojects

📦 thejudge26/c-asterisk-alpha

Is This The Secret To Super Fast Machine Learning?

Imagine getting the simplicity of Python with the raw power of a systems language. Meet C-Star, a custom programming language designed to bridge that gap by compiling directly to native machine code using LLVM. This project solves the problem of Python's slow execution speeds in data-heavy tasks by utilizing native I/O bypasses and a statically typed architecture. In rigorous benchmarks, C-Star outperformed Python by nearly one hundred times on complex neural network tasks, matching the performance of C++ while keeping the code readable. It is a brilliant example of how choosing the right compiler infrastructure changes everything.

🆔 @hackernewsgithubprojects

📦 p2r3/babel-usb

The Infinite USB Drive: Build Your Own Library of Babel

Ever wanted a USB drive that holds technically infinite data? This project turns an affordable ESP32-S3 development board into a digital Library of Babel, allowing you to explore a vast, generated filesystem. By leveraging the microcontroller's hardware USB support and MTP protocol, you can flash this firmware to create a unique, interactive experience that pushes the limits of storage imagination. It even includes a custom C-based library for handling UTF conversions, keeping the backend efficient. Grab your board, set up your workspace, and start navigating through an endless sea of virtual files today!

📰 https://news.ycombinator.com/item?id=48478766

🆔 @hackernewsgithubprojects

📦 uditgoenka/autoresearch

Automate Your Workflow with Claude Autoresearch

Stop manual tinkering and turn your AI assistant into a relentless improvement engine. This repository brings autonomous goal-directed iteration to tools like Claude Code, allowing you to set a specific metric and watch as the agent works through an iterative loop of modify, verify, and learn. Whether you are debugging errors, running security audits, or shipping features, it uses atomic changes and automatic rollbacks to ensure your progress is measurable and reliable. By leveraging thirteen specialized commands and a built-in safety system, you can effectively optimize your projects while you step away and return to finished, high-quality results.

🆔 @hackernewsgithubprojects

📦 proliferate-ai/proliferate

Master Coding Agents with Proliferate

Struggling to manage multiple AI coding agents? Meet Proliferate, the open source IDE designed to run agents like Claude Code, Codex, and Gemini in parallel within a single workspace. It solves the fragmentation problem by providing a unified environment where agents can even delegate tasks to one another. With native support for worktree workspaces, git review, and shared skills, you can finally coordinate complex development flows across both local and cloud sandboxes. Whether you are working solo or with a team, Proliferate streamlines your agent-driven workflow. Level up your coding productivity by centralizing your AI toolkit today.

🆔 @hackernewsgithubprojects

📦 ggml-org/llamabarn

Run Local AI Models Easily on macOS with LlamaBarn

Ever wanted to run powerful AI models right on your Mac without sacrificing your privacy or complex setup? Meet LlamaBarn, a lightweight menu bar application that brings local LLMs to your fingertips. By leveraging the power of llama.cpp, it creates a local server that connects seamlessly with your favorite chat interfaces, code editors, and command line tools. It automatically suggests models perfectly sized for your hardware and handles everything in the background, ensuring your data never leaves your device. If you want a zero-configuration, private way to supercharge your workflow with local AI, LlamaBarn is your new best friend.

🆔 @hackernewsgithubprojects

📦 https-deeplearning-ai/deeplearning-ai

Master AI Agents with DeepLearning.AI

Struggling to track what your AI agents are actually doing under the hood. Stop digging through messy logs and raw text. This repository serves as the official hub for DeepLearning.AI course materials, offering hands-on companion code for learners. You will find practical tools like the Agent Visualizer, which turns complex agent loops into clean, color-coded panels that display every thought and tool call in real time. Whether you are experimenting with ChromaDB for semantic search or building custom coding agents, these resources provide the perfect sandbox to sharpen your AI engineering skills. Dive in and start building your future.

🆔 @hackernewsgithubprojects

📦 junhengl/mpc-rl

Train Humanoids Smarter with MPC-Guided Reinforcement Learning

Imagine teaching a humanoid robot to walk or manipulate objects by giving it a mathematical shortcut instead of letting it learn from scratch. This repository introduces an efficient way to combine reinforcement learning with a GPU-batched model-predictive controller. By solving optimal plans for thousands of environments in parallel, the system provides real-time guidance that shapes rewards and informs the robot's decisions. This approach enables stable locomotion and complex tasks like pushing boxes much faster than traditional methods. It is a powerful framework for anyone looking to scale robot training and master complex dynamic movements with precision.

🆔 @hackernewsgithubprojects

📦 paavo5/ordergrad

Optimize Beyond the Mean with OrderGrad

Are you tired of your reinforcement learning agents settling for mediocre performance? Standard training often focuses on the average outcome, but real-world tasks often demand robustness against risk or a focus on extreme success. OrderGrad changes the game by letting you optimize beyond the mean using order-statistic gradient estimation. It provides a plug-and-play reward transformation that works with standard methods like PPO or GRPO to target specific objectives like tail risk, medians, or best-of-K performance. With multi-backend support for NumPy, PyTorch, and JAX, you can start shaping your model’s reward distribution today to build more reliable and capable AI.

🆔 @hackernewsgithubprojects

📦 opengvlab/future-l1

Future-L1: AI That Imagines Before It Predicts

Ever wonder if AI can truly plan for the future in video? Future-L1 is an innovative approach for multimodal models that enables compact, latent visual reasoning. Instead of relying on wordy text explanations or generating heavy, full-frame images, this system alternates between language tokens and continuous latent visual spans. This allows the model to effectively imagine upcoming video events in a compressed, efficient way. By using a specialized visual-gain selection method and unique reinforcement learning rewards, it delivers state-of-the-art performance in complex video prediction tasks. It is a powerful leap forward in making AI smarter, faster, and more visually intuitive.

🆔 @hackernewsgithubprojects

📦 12britz/awesome-free-models

Stop Paying For AI: The Ultimate Free Tool Repository

Running advanced AI shouldn't require a credit card or a massive subscription fee. This repository is a goldmine for anyone looking to harness powerful technology for free. It serves as a comprehensive, verified directory of open-weight models, local inference engines, and free API tiers from top providers. You will find everything from cutting-edge coding assistants to sophisticated chatbot interfaces that you can run entirely on your own hardware for maximum privacy. Whether you want to experiment with local tools like Ollama or access elite models via cloud platforms, this collection provides the perfect toolkit to get started today.

🆔 @hackernewsgithubprojects