Trending AI projects

CubeStudio is an open‑source, MIT‑licensed, cloud‑native AI platform that covers the full MLOps lifecycle—from data ingestion and multimodal annotation, through notebook‑based development and drag‑and‑drop pipelines, to distributed training (including large‑model fine‑tuning) and zero‑code inference serving. It runs on Kubernetes, supports many Chinese‑made accelerators (GPU, NPU, vGPU), offers multi‑tenant resource scheduling, a compute marketplace, built‑in billing, and extensive monitoring.

A marketplace of Claude Code / Copilot plugins that give AI agents the knowledge and automation needed to build, validate, and manage Power BI/Fabric models, reports and custom visuals.

tiny-vllm is a C++ and CUDA implementation of a high-performance LLM inference engine, built from scratch and accompanied by a step-by-step course. It focuses on serving Llama 3.2 1B Instruct with features like continuous batching, PagedAttention, and CUDA-optimized kernels.

vLLM-Moet provides optimized vLLM images and patches to run frontier Mixture-of-Experts models on Blackwell (SM120) consumer GPUs by implementing custom kernels, efficient KV caching, and 2-bit expert quantization.

A detailed guide and working system for converting an AMD BC-250 crypto-mining board into a fast, local AI server. It unlocks the GPU's full 40 compute units, gets Vulkan-based LLM inference running (with speeds up to 129 tok/s), and adds autonomous tools like a Signal chat bot, image generation, and a network monitoring suite.

An open-source AI agent skill set that automates the entire WeChat Official Account workflow from topic selection and writing to HTML formatting and publishing.

A collection of deep learning architectures and model implementations in PyTorch and TensorFlow, providing runnable Jupyter Notebooks for a wide variety of neural network types.

XGrammar is an open‑source library that adds constrained decoding to LLMs, guaranteeing 100 % structural correctness (JSON, regex, custom grammars) with near‑zero overhead. It works across platforms, hardware, and languages, and is the default structured‑generation backend for major inference engines like vLLM, SGLang, TensorRT‑LLM, and MLC‑LLM.

LLMs_interview_notes is a Chinese-language, topic-organized interview question bank for LLM algorithm roles, with 800+ questions across RAG, Agent, RL post-training, GraphRAG, and more. Each question follows a three-part structure (one-sentence answer, deep explanation, likely follow-ups) with difficulty ratings. It targets interview candidates, career-changers, and interviewers, and includes 126 open-sourced detailed explanations.

A provider-neutral Agent Skill that helps developers design, audit, and refactor production-safe agentic harnesses and runtimes.

An AI-native management tool for Milvus vector databases that provides a visual interface for data exploration, cluster monitoring, and AI-driven administration.

A modular high-level library for end-to-end development in embodied AI, used to train and evaluate agents performing tasks in indoor environments.

A leaked source archive of Anthropic's Claude Code CLI, a terminal-based AI coding assistant that can edit files, run commands, and manage git workflows.

An open-source pipeline that automates YouTube video editing by turning raw talking-head recordings into polished videos with AI-generated visuals, audio cleaning, and automated uploads.

A ComfyUI workflow and proof-of-concept that removes Google's SynthID watermarks from AI images by using controlled diffusion reconstruction to replace watermark-carrying pixels.

A comprehensive Chinese learning resource for DeepLearning.AI's Agentic AI course, providing translated content, structured knowledge breakdowns, and detailed code analysis to help users build autonomous AI agents.

tract is Sonos' Rust‑based inference engine that loads ONNX/NNEF/TFLite models, optimises them, and runs them on CPUs, Apple Metal, NVIDIA CUDA, or WebAssembly. It supports streaming (pulsified) inference for low‑latency tasks, offers Python bindings, and is licensed under Apache 2.0/MIT.

A Spec-Driven Development tool consisting of a desktop app, CLI, and AI agent skills to visually track and manage specifications, change proposals, and tasks.

M-3LAB/awesome-industrial-anomaly-detection is a curated, community-maintained list of research papers, benchmarks, and code for AI-based defect detection in manufacturing. It organizes the field by method type (unsupervised, supervised, zero/few-shot, etc.), lists state-of-the-art methods with code, and tracks recent publications by conference. It also highlights several notable resources: AD-Copilot (an AI assistant that reportedly beats humans on real inspection tasks), the MMAD benchmark for evaluating multimodal language models, and the IM-IAD benchmark. The repository is a reference index, not a tutorial — it points to papers and code rather than explaining methods itself.

A collection of academic writing best practices and an automated LaTeX checking tool designed to help researchers avoid common errors in conference paper submissions.

A collection of modular skills for AI coding agents that automate tasks like video production, competitive content analysis, macOS system management, and college application planning.

A scalable Python toolkit for analyzing single-cell gene expression data, supporting preprocessing, clustering, and trajectory inference for datasets with over a million cells.

A learning-based one-stage planner for autonomous drones that integrates perception and trajectory optimization into a single network for agile navigation in dense obstacles.

AI-Meeting is an open-source backend for AI-powered mock interviews. It parses resumes, generates and scores interview questions via multiple AI agents, supports real-time speech transcription and long-text synthesis, and provides a full user and permission system. Built with Spring Boot 3 and Spring AI, it's designed for learning and for building practical job-prep platforms.

This project is a marketplace of specialized skills for Claude Code that enables researchers to automate and streamline domain-specific tasks in the life sciences.

An MCP-based server that allows Large Language Models to search for 12306 Chinese railway ticket information, including transfers and stop-overs.

ChatAPI is a tool that lets you act as an AI assistant by providing an API interface for AI clients to send requests to you, which you can then answer manually or via automated rules.

A Model Context Protocol (MCP) server that gives AI assistants direct, real-time access to official Microsoft technical documentation and code samples to prevent hallucinations.

Apache HertzBeat is an open‑source, AI‑enhanced observability platform that collects metrics and logs without agents, analyses them, and sends alerts via many channels. It offers Docker, package, source‑code, Docker‑compose and Helm deployment options, and provides a UI for building status pages.

A workflow engine for Claude Code that turns it into a multi-model orchestrator, allowing it to dispatch specialized work to models like Codex and Gemini while maintaining context.

A topic modeling technique that leverages transformers and c-TF-IDF to create dense, interpretable clusters of documents.

A toolkit for integrating AI agents with 1C:Enterprise databases via MCP and REST API, featuring an embedded server and data anonymization.

A tool that connects Xiaomi AI Speakers to LLMs like ChatGPT and Gemini, allowing the speaker to answer questions using advanced AI via custom TTS engines.

A FastAPI proxy that converts DeepSeek's free web chat into an OpenAI-compatible API, featuring automated PoW solving and token refreshing.

An optimized inference engine for Qwen3-TTS that provides real-time streaming and high-speed generation across NVIDIA, AMD, and Apple Silicon hardware.

An open-source, enterprise-ready embedded BI platform that combines interactive dashboards, paginated reports, and AI-driven natural language analytics.

Markstream is a family of streaming Markdown renderers designed for AI chat and real-time data, preventing visual flicker and layout instability while content is still being generated.

Distributed Llama is a tool that links multiple home devices over Ethernet into a cluster to run large language models faster, using tensor parallelism and shared RAM. One root node coordinates, worker nodes process slices, and it supports Linux/macOS/Windows with ARM and AVX2 CPUs. It offers one-command setup for models like Llama 3.x, DeepSeek, and Qwen 3, but only works with 2ⁿ nodes and specific quantization types.

A distributed, scalable graph database designed for large-scale graph analytics and millisecond latency in applications like knowledge graphs and recommendation systems.

An open-source directory and scoring system for AI agent tools, MCP servers, and Claude Skills that helps developers find high-quality, security-graded agent capabilities.

onWatch is a cross‑platform daemon that monitors usage quotas for many AI‑coding services (Anthropic, Codex, Copilot, MiniMax, etc.). It stores historical data locally, serves a Material‑Design web dashboard, and offers alerts, a system‑tray widget, and a VS Code extension so developers can avoid throttling and manage AI spending.

Axiom is an AI‑assistant extension for Apple platform development. It provides ~280 prompt‑based skills, ~42 autonomous agents, and a set of command‑line tools (log capture, crash symbolication, UI automation, performance tracing) that integrate with Claude, Cursor, Gemini, VS Code MCP, Pi, and Xcode. Installed as a plugin, it lets developers ask natural‑language questions or run `/axiom:*` commands to automatically diagnose build failures, memory leaks, concurrency bugs, and more, with up‑to‑date support for Swift 6, SwiftUI, and upcoming OS 27 features.

An open-source PyTorch toolkit for Conversational AI that provides recipes and tools for speech and text processing, including speech recognition and synthesis.

Reflect is an open-source, local-first note-taking application that manages Markdown files and provides optional AI-powered search and querying capabilities using user-provided API keys.

noScribe is a free, open‑source desktop app that locally transcribes interview audio, detects speakers, and supports many languages using Whisper and pyannote‑audio. It’s aimed at researchers and journalists who need confidential, high‑quality transcripts.

An open-source vulnerability scanner for LLMs and agent workflows that protects AI systems from jailbreaks, fuzzing, and multimodal attacks.

A comprehensive architectural analysis of the Claude Code agent that reveals how to build reliable AI systems by focusing on deterministic infrastructure, safety, and context management rather than just AI logic.

Latent Box is a community-curated collection of high-quality resources for AI, art, and creative technology, designed to help users discover useful tools without the noise of automated search engines.