Open Source Projects

Cross-session memory system for AI coding assistants. A persistent memory service for AI programming agents built on the iii engine, automatically capturing and compressing agent actions into retrievable memories, then injecting relevant context into new sessions so the agent can resume from where it left off without re-explanation. Supports multiple coding assistants including Claude Code, Cursor, Gemini CLI, and Codex CLI, provides over 50 MCP tools and 12 hooks, achieves 95.2% retrieval recall on LongMemEval-S, and reduces token usage by 92%.

Learn AI engineering from scratch. This is a free, open-source, comprehensive AI engineering course consisting of 20 stages, 435 lessons, and approximately 320 hours of study. It covers topics from mathematical foundations such as linear algebra, through machine learning basics, deep learning, Transformers, large language models, Agent engineering, to multi-Agent systems. Each lesson requires coding and testing from mathematical derivations to deeply understand their principles and mechanisms.

AI coding assistant token usage analyzer. A tool to track token consumption across 20+ AI coding tools. Reads session data directly from disk without API keys and calculates call costs via LiteLLM. Supports analytics by task type, model, tool, project, and provider, with optimization suggestions, model comparisons, cost analysis, and CSV/JSON export.

An open-source alternative to Claude Design. This project supports local execution and deployment to Vercel without requiring any subscription or account. It automatically detects 10+ AI programming CLI tools in PATH as design engines, featuring 30+ composable skills and 70+ brand-level design systems.

An AI desktop assistant that integrates into your daily life. This open-source AI personal assistant features an expressive desktop mascot, can attend online meetings on your behalf, and continuously thinks and remembers in the background. It's ready-to-use, user-friendly, requires no complex setup, stores data in Markdown format into SQLite and Obsidian-compatible knowledge bases, supports integration with Gmail, Notion, GitHub, Slack, Jira, and automatically syncs latest data and updates memory every 20 minutes.

A toolkit for creating videos with HTML. Enables developers to produce videos using HTML, built-in frameworks like GSAP, anime.js, Lottie, and Three.js, supporting real-time preview/editing, background removal, speech synthesis, and subtitles, with seamless integration for tools like Claude Code, Cursor, Codex, and Gemini CLI.

Make AI-generated documents worth printing. This project is a document layout design toolkit for AI programming tools, addressing the issues of poor formatting and inconsistent styles in AI-generated documents. Includes 10 built-in templates such as one-pagers, resumes, slides, portfolios, and landing pages.

A skill package that helps AI programming tools stop random coding. Created by资深 developer Matt Pocock, this toolkit works with Claude Code, Codex, and others, installable via a single command. It includes over ten composable skills—such as requirement alignment, shared documentation, red-green refactor TDD, issue diagnosis, problem decomposition, and architecture optimization—to help developers harness AI through engineering practices instead of aimless vibe coding.

hermes-agent: An AI agent that grows smarter over time. This open-source, self-evolving AI assistant learns and evolves with the user, featuring an internal learning loop to automatically create and optimize skills based on past tasks, persistent cross-session memory, flexible model switching, and integration with multiple messaging platforms.

Your personal AI news radar. An AI-powered news aggregation and newsletter generation tool that collects content from multiple sources including Hacker News, Reddit, RSS, Telegram, and GitHub. It deduplicates, filters by AI scoring, enriches context, and generates summaries to produce bilingual (Chinese-English) daily reports. Reports can be published to GitHub Pages or delivered via email, Feishu, Slack, etc., with support for Docker deployment and scheduled execution via GitHub Actions.

A tool that enables multiple AI programming assistants to work simultaneously. This AI-powered desktop tool is designed for parallel multi-project Vibe Coding, with native support for Claude Code and Codex. Built with Tauri, React, and TypeScript, it integrates multi-project management, task tracking, session management, a virtual terminal, code editor, and Git, eliminating the need to switch between terminals, editors, and Git clients. Developers can manage Vibe Coding tasks across multiple projects within a single interface, significantly boosting productivity.

paseo: A platform to manage multiple AI coding assistants anytime, anywhere. A unified management platform built for Claude Code, Codex, and OpenCode, supporting clients on iOS, Android, desktop, web, and CLI, with features like voice control, cross-device collaboration, and isolated development.

Use AI to convert documents into editable PowerPoint presentations. This project generates editable .pptx files from documents in formats such as PDF, Word, URL, and Markdown, with support for adjusting text, layout, and style. It includes multiple canvas formats, supports custom models, and can be integrated with AI tools like Claude Code, Cursor, and VS Code Copilot.

A behavioral guideline to reduce AI coding assistant errors. Inspired by Andrej Karpathy's observations on LLM coding pitfalls, this project distills four principles—think before coding, simplicity first, precise edits, and result validation—and packages them into a skill set. Supports one-click installation into tools like Claude Code and Cursor, effectively reducing issues such as unwanted code changes, over-abstraction, and guessing without asking.

Make AI coding assistants talk less. A skill package that enables AI coding assistants to respond in a caveman-style, minimal text, reducing output tokens by 75% while maintaining technical accuracy.

A toolkit that turns code repositories into knowledge graphs. It scans code repos, Markdown, PDFs, screenshots, and audio-video files to automatically extract concepts and relationships, building a queryable knowledge graph. Supports various AI coding tools like Claude Code, Codex, OpenCode, and Cursor, ideal for developers wanting AI to better understand project context and reduce token usage.

Generate deliverable design drafts from a single sentence. A design toolkit for agents like Claude Code, creating high-fidelity prototypes, presentation slides, infographics, and animations via an HTML workflow. Supports design direction recommendations, brand asset parsing, 5-dimensional design review, and export to formats including MP4, GIF, and PPTX.

A dedicated knowledge base to reduce AI coding hallucinations. This project provides a community-curated, versioned API documentation repository tailored for AI programming tools like Claude Code, supporting on-demand retrieval, local annotation, and feedback loops to effectively minimize hallucinations in AI-generated code.

An AI-powered conversational video creation agent. Built on LLM/VLM, this automated video creation agent allows users to describe their intentions in natural language, and automatically handles material collection, editing, storyboarding, voiceover, and music. It supports conversational refinement, including trimming, reordering, and subtitle editing, and enables one-click saving as an editing skill package for quickly replicating videos of the same style.

Turn AI coding assistants into a virtual development team. A collection of skills that enhance AI coding assistants' capabilities, enabling them to simulate YC partners' questioning styles and roles such as CEO, architect, designer, and tester, transforming Claude Code into an on-demand virtual development team.

An AI operator for your website. This is an embedded GUI agent designed for web developers. Unlike traditional browser automation tools, it requires code integration into the website, enabling users to interact with the page via natural language after integration.

A Python framework for quickly building production-ready agents. This framework enables the development of AI agent applications with simple code, supporting model capabilities, tool calls, autonomous workflows, multiple models, Agent Loop, multi-agent collaboration, and voice conversations.

An LLM-powered intelligent stock analysis system. This tool automatically analyzes and delivers daily insights for A-shares, Hong Kong stocks, and US stocks. It fetches real-time market data via AkShare, Tushare, and YFinance, and leverages large model APIs like DeepSeek to perform multi-dimensional analysis (technical indicators, chip distribution, sentiment) on selected stocks, generating a decision dashboard. Supports scheduled execution via GitHub Actions (no server required) or one-click deployment with Docker.

Build an AI Agent from scratch. This project teaches how to create a Claude Code-like AI Agent tool through 12 lessons, each with a runnable Python file. Starting from the basic Agent loop, it progressively adds features such as tool calling, task planning, sub-agents, context compression, multi-agent collaboration, and autonomous execution, ultimately building a complete AI Agent system.

A zero-dependency, single-file implementation of modern AI mainstream algorithms. This educational project is designed for learning AI algorithms, featuring 30 standalone, runnable Python files with no dependencies, covering topics from basic GPT to fine-tuning (LoRA, PPO) and inference optimization (Flash Attention). Each algorithm is implemented in clear, easy-to-understand code, accompanied by Manim animations for better comprehension and learning.

A viral personal AI assistant. This open-source personal AI assistant, built with TypeScript, can be quickly deployed on macOS, Windows, and Linux, and supports interaction via instant messaging apps like WhatsApp, Telegram, and Slack. With sufficient token allowance, it can work for you 24/7 without stopping.

A minimalist AI Agent toolkit. This TypeScript-based AI Agent toolkit is the foundation of the popular OpenClaw. It provides essential features for developing AI Agents, including unified multi-LLM service interfaces, agent state management, tool calling, interactive CLI, WebUI, and Slack bot integration.

A locally-run intelligent knowledge base search tool. This fully offline smart search engine can retrieve personal documents, knowledge bases, meeting notes, and Markdown files. It integrates lightweight local models, BM25 full-text search, vector semantic search, and re-ranking, offering plug-and-play functionality without internet access and supporting the MCP protocol, making it suitable for AI assistants and Agent workflows.

An AI-driven cultivation world simulator. This LLM-based cultivation simulation game features AI agents as NPCs, each with independent personalities, memories, and behavioral logic. Players assume the role of the "Heavenly Dao," observing and influencing AI cultivators and celestial rules from a god's-eye view, witnessing sects' rise and fall and the emergence of prodigies.

Control Claude Code and Codex from your phone. A tool for remotely operating Claude Code or Codex, allowing you to access and control your AI programming assistant anytime, anywhere, with iOS, Android, and Web clients.

Generate 3D scenes in under a second. This project is the companion code for Apple's open-sourced monocular view synthesis technology, capable of generating high-quality 3D scenes from a single image in a short time. It uses neural networks to regress 3D Gaussian parameters from a single photo, outputting a ply file usable by 3DGS renderers.

A lightweight and efficient ONNX model optimization tool. Implemented purely in Python with no additional compilation dependencies, OnnxSlim simplifies and optimizes ONNX model structures by analyzing and rewriting computation graphs. It automatically removes redundant nodes, dead branches, and unnecessary parameters, reducing model size and improving inference speed while preserving accuracy, ideal for model publishing, deployment, and engineering scenarios.

Reimplementing classic deep learning papers in around 100 lines of code. This project provides minimal, runnable Python implementations of over 60 seminal deep learning papers, helping researchers, students, and developers quickly grasp their core ideas and implementation details.

AI-powered PPT generation tool. Based on the Nano Banana Pro API, it automatically creates well-structured and beautifully formatted presentations from user ideas, outlines, or documents (e.g., PDF, Markdown), with support for content refinement via conversational interaction.

A memory system for AI programming tools. Designed specifically for AI coding assistants, it uses Git and JSON for persistent storage, providing long-term, structured memory to address context loss in extended, complex programming tasks.

Official Python client for Hugging Face. This open-source project provides official Python tools for downloading, uploading, and managing models, datasets, and Spaces on the Hugging Face platform.

Visual explanations of large model technologies. This project includes over 100 diagrams systematically covering large language models and reinforcement learning, including LLM/VLM architectures, training algorithms (RL, RLHF, GRPO, DPO, SFT, CoT), performance optimization, and RAG.

Automated evaluation framework for RAG applications. An open-source framework designed to evaluate and optimize RAG applications, providing objective, quantifiable results and supporting automatic generation of test datasets.

anarlog: A locally-first AI note-taking and meeting assistant. This offline-capable app uses Ollama to perform speech transcription and summary generation entirely on-device, supporting meeting recordings, real-time transcription, note organization, and intelligent summarization.

A plug-and-play memory engine for AI agents. This Python project provides memory capabilities for AI agents, integrating graph databases, knowledge graphs, and vector databases. With just 5 lines of code, it enables persistent, multimodal memory for AI agents, supporting storage and retrieval of past conversations, documents, images, and audio transcripts.

An AI desktop assistant at your fingertips. This context-aware desktop AI assistant automatically captures and understands on-screen content without requiring screenshots, copying text, or switching apps. Simply press a key to summon the AI for tasks like Q&A, translation, and problem-solving.

Control your phone with natural language. This is an LLM Agent-based mobile automation framework that operates Android devices or emulators via natural language, supporting mainstream large models like DeepSeek, OpenAI, and Gemini. Requires installing DroidRun Portal on the phone to collect UI information and execute commands, then uses ADB to send data to the DroidRun framework on a computer, which interacts with the LLM to generate execution instructions.

Let AI draw architecture diagrams for you. A Next.js-based web app integrating AI with draw.io diagramming capabilities. Generate, edit, and optimize flowcharts and architecture diagrams through conversation, featuring animated connections, screenshot replication, version history, and more.

An open-source AI browser based on Chromium that runs AI agents locally, serving as an open-source alternative to ChatGPT Atlas, Perplexity Comet, and Dia. It retains Chrome's familiar interface and extension compatibility while enabling AI-powered browser automation, smart Q&A, and support for custom LLM services or local large models.

A training-free caching acceleration framework for DiT models. This project provides a unified caching acceleration solution for Diffusers, supporting nearly all DiT diffusion models including Qwen-Image-Lightning, Qwen-Image, HunyuanImage, Wan, FLUX, etc. It enables efficient caching acceleration with minimal code changes, significantly improving inference speed without retraining models.

Build real-time knowledge graphs for AI agents. A framework designed for AI agents to construct and query real-time, time-aware knowledge graphs. It continuously integrates dynamic data from user interactions, structured or unstructured sources into coherent, queryable knowledge graphs. Supports incremental updates, efficient retrieval, and historical queries, ideal for developing interactive, context-aware AI applications.

An ultra-low storage footprint vector database. This open-source lightweight vector database achieves minimal storage usage by computing embeddings on demand. Users can build powerful, fully private retrieval-augmented generation (RAG) systems on personal devices (e.g., laptops), enabling semantic search across local files, emails, browser history, chat logs, and other data sources.

A hands-on tutorial for building an AI coding assistant from scratch. This project teaches you how to develop a local AI coding assistant using Go and the Claude API, starting with a simple chatbot and progressively adding features like file operations, command execution, code editing, and search.

parlant: An LLM agent framework designed for real-world control. It enables precise behavior management in complex business scenarios by defining agent rules in natural language, ensuring AI stays on task and avoids hallucinations or deviations.

SwanLab: An AI model training tracking and observability platform. A tool designed for tracking, logging, analyzing, and collaborating on AI model training, helping researchers optimize training processes and improve team efficiency. It offers training visualization, automatic logging, hardware monitoring, experiment management, and multi-user collaboration via a simple Python API and intuitive interface. Integrated with 40+ mainstream training frameworks, suitable for large model training, computer vision, audio processing, AIGC, and more.

Out-of-the-box local speech transcription tool. An open-source solution for real-time speech-to-text, translation, and speaker separation, complete with server and Web UI, deployable privately with a single command. Built on the latest incremental streaming technology, it enables ultra-low-latency meeting transcription and cross-language communication without internet dependency or frontend coding.

An open-source AI virtual companion that brings anime-style characters (waifus) and agents to life. Supports real-time text and voice chat, can play games with you like Minecraft and Factorio, and offers web and desktop applications.

An invisible second brain. This is a discreet desktop AI assistant that doesn't appear in screen recordings, screenshots, or on your desktop. Supports real-time reading of screen and microphone content, turning everything you see and hear into searchable, reviewable structured knowledge.

A visual AI workflow builder. An open-source platform for developers and enterprises to build AI agents and workflows. It packages LangChain's core capabilities (chains, tools, memory, vector stores, etc.) into reusable components and uses React Flow for visual workflow editing, enabling users to design, debug, and deploy complex AI workflows without coding.

An open-source real-time multimodal conversational AI framework designed for real-time AI audio-video interaction scenarios. It features a user-friendly web interface for creating, customizing, and deploying conversational AI agents with voice, vision, and virtual avatar capabilities.

AI programming assistant task management board. A kanban tool designed for AI coding agents, unifying popular AI programming assistants like Claude Code, Gemini CLI, and Codex. It integrates kanban tasks, Git repositories, and AI coding agents, supporting multiple AI agents to automatically complete bug fixes, feature development, project initialization, and documentation generation.

A tool to instantly convert code repositories into an AI-friendly format. It quickly transforms any GitHub repository into a plain text summary suitable for large language models. Easy to use—just replace "hub" with "ingest" in the GitHub project URL to get the text summary.

Dynamically visualize neural networks with Python. This Python library, based on Manim, creates animations and visualizations for machine learning concepts. Simply write basic Python code to generate animations of neural network architectures, convolution operations, dropout processes, and more, aiding in understanding and demonstrating complex machine learning principles.

A beginner-friendly LLM fine-tuning library. This Python toolkit enables efficient fine-tuning and optimization of large language models (LLMs) using dynamic quantization and memory optimization techniques, improving training speed and reducing GPU memory usage by 70%-80%. It supports various hardware setups, LLMs, and long-context tasks, and includes interactive Jupyter Notebook examples for easy hands-on experience, lowering the barrier to LLM fine-tuning.

A high-performance AI inference engine exclusively for MacBook. Designed specifically for Apple M-series chips, uzu is a lightweight, high-performance AI model inference engine that leverages Apple's hardware capabilities to accelerate inference speed, offering simple, easy-to-use APIs for one-click deployment of efficient local large model services.

An out-of-the-box intelligent captioning assistant. This is a smart video subtitle tool powered by large language models, featuring a clean interface, easy operation, and support for speech recognition, intelligent proofreading, and automatic generation of multilingual subtitles.

A terminal-based Claude coding assistant. This officially open-sourced AI coding assistant by Claude integrates into the terminal, understands the entire codebase, and helps developers efficiently complete various coding tasks using simple natural language commands.

A command-line tool for Google Gemini. This official open-source tool integrates the powerful capabilities of Google Gemini into the terminal environment. Based on a million-token context, it can understand the architecture and logic of large codebases, and supports multimodal input/output, Google Search, and MCP features.

A tutorial on the principles and practice of LLMs from scratch. This project is a systematic guide for beginners to learn the fundamentals and applications of large language models (LLMs). Through detailed tutorials and hands-on examples, it progressively guides readers through NLP basics, the Transformer architecture, and the core principles of pre-trained language models, enabling them to implement and train their own LLMs.

A lightweight vLLM built from scratch. This Python-based project implements a lightweight large language model inference engine in just over 1,000 lines of core code. It features clear structure, ease of understanding, inference speed comparable to the original vLLM, and integrates optimizations such as prefix caching, tensor parallelism, and Torch compilation.

A tool for optimizing AI prompts. This front-end-only prompt optimizer helps users quickly create higher-quality prompts. Supports multiple mainstream AI models and custom API addresses, with real-time comparison of results before and after optimization.

Bring static photos to life. This project can quickly generate high-quality, vivid videos from a single photo, excelling at capturing and reproducing facial expressions, and supports animation generation for both human and animal portraits.

A simple and easy-to-use multi-platform AI inference deployment framework. Designed with a directed acyclic graph, it abstracts preprocessing, inference, and postprocessing as nodes, supporting optimizations like pipeline and task parallelism. Compatible with multiple backends including TensorRT, OpenVINO, and MNN, it supports mainstream models such as text-to-image, large language, and detection models, enabling cross-platform deployment with a single codebase.

Open-source LLM application evaluation framework. A framework for building evaluation, testing, and monitoring platforms for LLM applications, featuring an intuitive web interface that logs all LLM calls during development and production, supports feedback scoring, test case storage, hallucination detection, and CI/CD integration, suitable for applications such as RAG chatbots, code assistants, and complex agent pipelines.

Easily create your local AI digital twin. This project is a platform focused on training AI with personal data, helping everyone build, train, and own their own local AI digital twin. It uses Hierarchical Memory Modeling (HMM) and Me-Alignment algorithms to integrate your knowledge, interests, and preferences into the AI, creating a smarter, more personalized assistant.

An open-source AI code editor based on VSCode, serving as an open-source alternative to Cursor. It emphasizes user privacy by sending data directly to model providers and supports integration with local models or major LLM services such as OpenAI, Claude, DeepSeek, Gemini, and others.

A high-performance virtualization and control framework designed for AI agents, providing lightweight, near-native-speed virtual containers. It enables AI agents to automatically perform computer-based tasks—such as operating applications, browsing the web, writing code, and automating complex workflows—in fully isolated macOS or Linux environments, with built-in support for multiple mainstream LLMs.

Easily generate long AI videos on a laptop. This predictive neural network architecture for video generation delivers high-quality results on resource-constrained devices by compressing context into a fixed length, making computational cost independent of video duration for efficient, resource-friendly generation.

AI-generated codebase tutorials. This project uses AI to transform code repositories into easy-to-understand tutorials by analyzing GitHub repositories or local code directories, automatically generating detailed learning guides to lower the barrier for beginners to understand and get started with large codebases.

Automatically selects cost-effective GPU options for AI training and inference. This open-source, cross-cloud AI and batch job scheduling platform enables users to run deep learning, distributed training, inference, and batch tasks across K8s, on-premise clusters, and major cloud providers (AWS, GCP, Azure, etc.) via a unified interface. It automatically finds the cheapest available GPU/TPU/CPU resources, supporting queuing, automatic fault tolerance, resource sharing, and cost optimization.

Fine-tune large language models using WeChat chat records. This project personalizes a digital clone by fine-tuning LLMs with individual WeChat conversations, offering an end-to-end solution from chat data to model fine-tuning, text generation to voice cloning, and training to deployment. Enable your digital twin to not only "speak like you" but also "sound like you."

Microsoft's open-source AI Agent tutorial for beginners. This project is an AI Agents tutorial created by Microsoft specifically for beginners, divided into 10 lessons that include detailed tutorials, videos, and sample code.

A powerful tool for cross-framework conversion of machine learning code. It converts machine learning models, tools, and libraries from one framework to another. Developers can convert code using simple functions, supporting mainstream frameworks like TensorFlow, PyTorch, and JAX.

Hand-crafting various AI algorithms and models using Excel. This project implements and demonstrates core algorithms and concepts of AI and deep learning, enabling beginners to gain hands-on experience and understand the underlying principles. It covers matrix multiplication, MLP, RNN, Transformer, ResNet, and more, lowering the barrier to AI learning in a unique and accessible way.

An easy-to-use multi-platform LLM chatbot. Built on Python, this project is an LLM chatbot and development framework that supports integration with various messaging platforms and large language models. It provides a visual management panel and a flexible plugin extension mechanism, featuring rate limiting, whitelisting, keyword filtering, image understanding, and speech-to-text.

An open-source, low-code RAG platform built on LangChain and LlamaIndex. It supports one-click Docker deployment and features a user-friendly web interface with data upload, model management, internet access, and prompt templates, enabling users to build, debug, and deploy RAG applications without writing code.

An open-source LLM evaluation framework. This is a framework for evaluating large language models, capable of testing model performance across various tasks. It provides over 60 academic benchmarks and supports multiple model frameworks, local models, cloud services (like OpenAI), hardware acceleration, and custom tasks.

An AI tool for removing image watermarks. This is a TensorFlow-based open-source image inpainting tool that removes watermarks from images with a single command.

Connect "old" devices to the digital world. Using affordable hardware like ESP32 (under €10) and TensorFlow Lite, this project enables automatic recognition of meter digits and data transmission, effortlessly transforming traditional utility meters (water, gas, electricity) into smart devices.

A Python library for structured LLM outputs. Built on Pydantic, it provides data validation and type annotations to convert natural language results into structured data. It supports multiple LLM services, automatic retries, and streaming responses.

An open-source LLM vulnerability scanner. This project is an open-source tool by NVIDIA for scanning LLM vulnerabilities, supporting the detection of potential security issues and undesirable behaviors, including hallucination, data leakage, prompt injection, misinformation, toxic content generation, and jailbreaks.

A development framework for injecting memory capabilities into LLM applications. This project is a Python framework for building LLM applications with memory functions, supporting the creation of agents with long-term memory and persistent state, and integrating various LLM API services.

A zero-barrier AI face-swapping application. A lightweight, ready-to-use tool that requires no GPU and supports offline use. Easily try new looks (hairstyles, outfits, etc.) with one click.

A high-efficiency LLM inference and serving engine. Designed to address slow inference speeds and low resource utilization, this easy-to-use engine is built on PyTorch and CUDA. It combines memory optimization algorithms (PagedAttention), computation graph optimization, and model parallelism to significantly reduce GPU memory footprint and maximize multi-GPU performance. vLLM is seamlessly compatible with Hugging Face models and supports efficient execution on various hardware platforms (GPU, CPU, TPU), making it suitable for real-time Q&A, text generation, and recommendation systems.

Convert PDF to Markdown using vision models. This project leverages vision model APIs to transform PDF documents into Markdown. It works by converting source files (e.g., PDF, DOCX) into images, processing them via a vision model, and aggregating the results into a complete Markdown file.

Krita AI painting assistant plugin. An AIGC plugin developed specifically for Krita to provide a more convenient and controllable image generation experience. Users can easily perform operations such as inpainting, outpainting, upscaling, and adding or removing objects by simply selecting an area and entering a text prompt. It supports local execution, Stable Diffusion, ControlNet, IP-Adapter, and custom checkpoints.

Cross-platform machine learning model viewer. A visualizer for neural networks, deep learning, and machine learning models, supporting various model formats including ONNX, TensorFlow Lite, Core ML, Keras, Caffe, Darknet, and PyTorch.

A flexible and easy-to-use YOLO deployment tool optimized for NVIDIA devices. It integrates TensorRT plugins and CUDA technology to provide C++ and Python APIs, significantly improving inference speed and usability. It supports multiple YOLO versions and is suitable for various scenarios such as object detection, instance segmentation, pose estimation, oriented object detection, and video analysis.

Small vision language model. This is a small vision language model that runs on resource-constrained devices, capable of understanding and generating natural language descriptions related to images, supporting image recognition, captioning, and question answering.

A comprehensive practical guide to prompt engineering. This tutorial aims to help users master the skills of communicating with Large Language Models (LLMs). It covers prompt engineering techniques from basic to advanced, accompanied by detailed implementation guides and sample code.

A powerful Python library for Natural Language Processing. This industrial-grade NLP library supports tokenization and training for over 70 languages. Written in Python, it enables annotation, parsing, and text classification, and supports model packaging and deployment.

A minimalist neural network implemented in C. This project demonstrates how to implement a minimal neural network from scratch. Using less than 200 lines of code and the C standard library, it implements a minimalist neural network capable of classifying handwritten digits from the MNIST dataset.

A toolkit for optimizing and deploying deep learning models. This open-source toolkit from Intel aims to accelerate and optimize deep learning model deployment. It helps developers deploy trained models to various hardware platforms, supporting frameworks such as TensorFlow, PyTorch, and ONNX.

Kubernetes troubleshooting AI assistant. This project leverages LLMs to automatically analyze Kubernetes cluster issues and provide diagnostics and optimization suggestions. It generates reliable diagnostic reports by reading cluster status data and configurations.

A data framework for Large Language Models. Designed specifically for LLM applications, it helps developers easily integrate private data with LLMs. It provides data connectors to support indexing from various sources such as APIs, PDFs, documents, and SQL, simplifying data ingestion and querying operations. This enables even beginners to perform LLM context enhancement with just a few lines of code.

A collection of Retrieval-Augmented Generation (RAG) tutorials. This project provides tutorials for over 20 advanced RAG techniques, including implementation guides and sample code, with regular updates. It covers retrieval queries, context augmentation, Fusion Retrieval, hierarchical indexing, context compression, knowledge graph integration, and more.

A lossless upscaling tool for videos and images. This project integrates various super-resolution algorithms (such as Waifu2x, Anime4K, Real-ESRGAN) to effectively improve resolution, offering usage via GUI, Docker, and CLI.

A Python library for enhancing LLM context continuity. It provides a memory layer for mainstream LLMs, storing sessions and context from user interactions with real-time dynamic updates. This enhances AI personalization, making it suitable for personalized LLM applications requiring long-term memory, such as learning assistants, medical assistants, and virtual companions.

An API aggregation gateway for large language models. This project provides a unified and fast API, enabling developers to easily access over 200 LLMs, and supports features such as multimodal, automatic retries, and load balancing.

An AI tool for one-click generation of short videos. This project is an AI video generation tool based on large model services. By providing just a theme or keyword, it can automatically generate high-definition short videos. It features a user-friendly web interface and supports batch generation, setting video duration, and both landscape and portrait orientations.

Omost: A minimalist text-to-image tool. This project leverages LLM programming capabilities to automatically refine text-to-image prompts for users, generating high-quality images from brief user inputs. It also supports features like local image modifications, such as turning a dragon in an image into a dinosaur, significantly lowering the barrier for creating satisfactory images without the need for complex prompts.

AI-based Python web scraper. This is an AI-driven Python web scraping library that leverages the capabilities of LLMs to automatically scrape data from target websites based on prompts.

An open-source AI face swapping and enhancement tool. This powerful tool supports replacing faces in images/videos with another person's face, improving facial and background clarity, and offers a user-friendly web interface (WebUI) and low-barrier CPU processing options.

A tool for simplifying API calls to large models. This project unifies the interfaces of various AI large models and services into OpenAI's format, making it easier to switch and manage different AI services/large models. It also supports setting budgets, limiting request frequency, managing API Keys, and configuring OpenAI proxy servers.

Framework for building applications based on LLM. LLM refers to large deep learning models pre-trained on big data, and this project allows LLMs to access new datasets without retraining. It provides various modules needed for developing LLM applications, including connecting to context sources and performing inference with LLMs, making it easy to develop a variety of LLM-powered applications such as chatbots, knowledge bases, and information extraction.

A framework specifically designed for fine-tuning LLMs. This open-source project makes it easy to fine-tune large language models, supporting various models such as LLaMA, Mistral, and ChatGLM, as well as features like incremental pre-training, reward model training, and precision levels. It introduces advanced algorithms and practical fine-tuning techniques, while also providing convenient monitoring tools for users to track their experiment progress.

A tool for running various LLMs locally. This is a utility written in Go that allows you to install, start, and manage large language models with a single command on your local machine. It supports models like Llama 3, Gemma, Mistral, and is compatible with Windows, macOS, and Linux operating systems.

Chat with your SQL database. This project uses LLM+RAG+database technology to allow users to query SQL databases through natural language, answering your questions with generated SQL.

An open platform for training and evaluating large language models. This platform is designed for training, deploying, and evaluating large language models, allowing you to deploy and assess various models locally. Additionally, it provides an online platform for evaluating large models, where users can ask the same question to two different models and choose which one they believe performs better. During this process, you can use dialogue bots like Claude and ChatGPT for free.

A generative AI tutorial for beginners. This is a free, 18-lesson course on generative AI for beginners, open-sourced by Microsoft. It covers everything needed to create generative AI applications, including an introduction to generative AI and LLMs, prompts, building text generation applications, chat applications, image generation applications, and vector databases.

A one-stop desktop application for experiencing LLMs. This AI conversation desktop app supports running open-source LLMs locally and connecting to ChatGPT services. It is ready to use out of the box, has a clean interface, and is not hardware-intensive. Features include setting up proxies, connecting to ChatGPT, one-click download/access to large models adapted for the current computer configuration, and offline operation. It is compatible with Windows, Linux, and macOS operating systems.

Run code on your computer with LLM. This project allows large language models to execute code locally, supporting programming languages such as Python, JavaScript, and Shell. Essentially, the large language model acts as an interpreter, understanding your intent, converting natural language into corresponding code scripts, and running them. After installation, users can operate their computer through a chat interface in the terminal, such as creating and editing images, videos, and files, and controlling Chrome for searches.

3D visualization of GPT large language models. This project demonstrates the working principles and inference process of GPT-like large language models through 3D visualization.

From zero to neural network hero. This is a course on neural networks starting from the basics, including videos, exercises, and accompanying source code, designed to help beginners gradually master the fundamental concepts of neural networks and deepen their understanding through example codes.

An open-source library for conversational data analysis. This project combines AIGC with data analysis, allowing users to ask questions about their data in natural language and receive corresponding answers. First, the data needs to be imported in a pandas format, then configure the OpenAI TOKEN to start interacting with the data through conversation and chart drawing, without the need to write code.

Bring AI-generated images to life. This is a library for creating animated GIFs in Stable Diffusion, supporting the conversion of most open-source models into animation generators. Turn text-generated images into moving GIFs.

Effortlessly DIY image text, customizing your creative designs. This project offers two modes: text generation and text editing. It can generate images that blend text and visuals based on prompts, ensuring the accuracy of the text. It also supports editing the text in uploaded images and regenerating the images. Supports multiple languages including Chinese, English, Japanese, and Korean, suitable for poster design, logo design, creative doodles, and meme creation.

Open-source video translation and dubbing tool. This project can translate videos from one language to a specified language, and automatically generate and add subtitles and dubbing in the corresponding language.

Real-time interactive AIGC image generation. This project can generate AIGC images at an astonishing speed, reaching up to 100 images/second with a single RTX4090 GPU. It simplifies data processing through stream batching, reduces computational redundancy using the Residual Classifier-free Guidance (RCFG), enhances GPU utilization with a random similarity filter, and achieves parallel processing by optimizing IO queues. Additionally, it leverages various model acceleration tools to explosively increase the speed of AIGC image generation.

一条命令自动移除图像背景。该项目通过 AI 技术，可自动移除图片和视频中的背景。

机器学习路线图。这份机器学习思维导图，包含了机器学习相关的问题、学习步骤、工具、底层数学知识、教程资源等，为如何学习机器学习指出方向。

通过强化学习训练 AI 玩 Pokemon。该项目使用 Python + RL 从零训练了一个玩「宝可梦红」的 AI，同时作者还提供了配套的讲解视频，以及如何在本地运行和自定义训练的教程，快来上手试试吧！

计算机视觉 AI 工具库。该项目简化了对象检测、分类、标注、跟踪等计算机视觉的开发流程。开发者仅需加载数据集和模型，就能轻松实现对图像和视频进行检测、统计某区域的被检测数量等操作。

自动检测数据集中错误数据和标注的框架。该项目基于置信学习（confident learning，CL）算法，实现了自动检测出机器学习数据集中的各种问题，提高数据集质量训练出更好的模型，支持图像、文本、音频类型的数据。

一个基于节点流程的 AI 绘图操作界面。该项目将 Stable Diffusion 流程分成多个节点，通过拖拽各种节点构成图像生成到处理的工作流，支持 Stable Diffusion 1.x 和 2.x 版本、组合各种模型、根据 PNG 图片生成完整的工作流等功能。

一款针对 AI 项目的数据版本管理工具。基于 Git 的数据版本管理工具，版本化机器学习项目的数据和模型。可用于比较代码、数据、参数、模型或性能图，共享机器学习项目的数据或重现结果。

机器学习：LLM/VLM 训练与工程。该项目是作者训练开源 BLOOM-176B 大模型和 IDEFICS-80B 多模态模型的经验总结，还提供了大量可以直接拿来用的代码和脚本，希望能够帮助你成功训练大型语言模型和多模态模型。

将截图转化为 HTML 代码的工具。该项目可以将屏幕截图转化为 HTML/JS/Tailwind CSS 代码，它使用 GPT-4 Vision 生成代码、DALL-E 3 生成相似的图片。

AI 系统 & 深度学习教程。该项目主要是关于人工智能、深度学习系统设计的教程，内容涵盖 AI 基础知识和概述、AI 芯片、AI 编译器原理、AI 框架核心技术等方面。

轻量级人脸识别和分析库。该项目是 Python 写的人脸识别和面部属性分析框架，可根据人脸图像智能识别年龄、性别、情绪等信息。

微软开源的 LLM 提示编程语言。该项目是微软开源的编写 prompt 的模版语言，用于编写复杂、可复用、易于管理的 prompt，支持 GPT-4、LLaMa 等多种大语言模型。

An elegant library for solving hCaptcha challenges. Independent of browser extensions and third-party anti-captcha services, it leverages machine learning to tackle hCaptcha. It utilizes ResNet and YOLOv8 for classification and detection, and packages models via ONNX to lower the barrier to entry. This enables users to run the code on CPU-only devices without installing dependencies like PyTorch, providing a seamless and elegant hCaptcha solution.

An open-source text-to-music library by Meta. It generates high-quality, high-fidelity audio and music from text prompts—such as whistling in the wind or a pop dance track for a beach scene—with stunning results.

An out-of-the-box image generation tool. Designed to combine the strengths of Stable Diffusion and Midjourney, it features easy installation and intuitive operation, eliminating the need for complex parameter tuning. Users simply enter prompts to generate images on par with Midjourney's quality. It supports local deployment and offline use, with minimum requirements of 8GB RAM and a 4GB NVIDIA GPU.

Yida Xu's Machine Learning Course. Open-sourced by Professor Yida Xu of Hong Kong Baptist University (HKBU), this project provides lecture notes and video links on machine learning, probabilistic models, and deep learning.

A real-time AI face-swapping application for live streaming and video. It performs real-time face swapping on subjects in webcam feeds and local video files, making it ideal for PC live streaming and video production.

An open-source deep learning training optimization library by Microsoft. Training large language models (LLMs) is notoriously time-consuming and expensive. Leveraging ZeRO++ technology, this project partitions model states across GPUs during training, boosting throughput to reduce training time and costs.

ModelScope is an AI Model-as-a-Service sharing platform. Open-sourced by Alibaba DAMO Academy, it offers developers a one-stop solution for model exploration, inference, training, deployment, and application, serving as China's equivalent to Hugging Face.

An introductory open course on decision intelligence. This course kickstarts your journey into decision AI through an in-depth exploration of Proximal Policy Optimization (PPO), a classic deep reinforcement learning algorithm. Featuring videos, theoretical materials, exercises, code, and application examples, it helps beginners quickly master how to apply PPO to solve a wide range of decision intelligence problems.

Microsoft's open-source introductory AI tutorial. This completely free, 12-week, 24-lesson course is designed for absolute beginners. It covers the history and fundamentals of AI, mainstream frameworks, computer vision, and NLP.

Edit images interactively using GANs. This is the official source code for DragGAN, enabling image manipulation via mouse dragging. Anyone can easily modify the pose, expression, shape, and layout of objects in an image by precisely controlling pixel movement. For example, you can make a standing dog in a photo sit down.

Google's open-source, cross-platform machine learning framework. This ML toolkit deploys easily across mobile, web, desktop, and IoT devices, featuring models for object detection, image classification, face recognition, gesture recognition, text classification, language detection, and audio classification.

AI-generated game character animations. This project leverages raw motion capture data to produce more natural and controllable animations, solving animation challenges for bipeds and quadrupeds. It autonomously generates complex actions such as sitting, jumping, opening doors, and martial arts moves without human intervention.

An open-source face restoration algorithm by Tencent. It restores low-resolution, blurry, and damaged facial images, delivering exceptional results, particularly in enhancing facial details and clarity.

An AI model that connects multiple sensory data streams. Open-sourced by Meta AI, this novel multimodal model enables seamless translation across six distinct modalities, including images, text, and audio. For example, given an audio clip of a train, it can automatically generate corresponding images, videos, and text.

Deep Learning Tuning Playbook. Distilled from the model tuning experiences of Google Brain researchers, this guide covers starting new projects, improving model performance, and practical training insights. It is ideal for engineers and researchers with foundational machine learning knowledge who are interested in optimizing deep learning models.

An AI project that brings drawings to life. This open-source animation library from Meta AI Research animates your artwork, enabling everything from children's doodles to professional cartoon characters to mimic your movements.

A text-prompted audio generation model. It generates audio from text prompts and supports background sound specifications; for example, the prompt "I love HG (laughs)" yields speech accompanied by hearty laughter. It also simulates music and sound effects, offering a smarter alternative to traditional text-to-speech.

A deep learning tool for restoring old photos. Open-sourced by Microsoft, this project repairs damaged vintage photos with remarkable results.

An open-source bilingual conversational language model by Tsinghua KEG. Based on the GLM architecture, this Chinese-English model features 6.2 billion parameters and supports inference on a single 2080 Ti GPU.

An open-source decision AI platform by OpenDILab. This PyTorch-based general-purpose decision intelligence engine offers developers 60+ algorithms and 40+ environment types, supporting customized training and real-world applications such as game AI, autonomous driving, and biological sequence prediction.

Run LLaMA large models on a laptop. This project enables smooth CPU inference for LLaMA models, supporting macOS, Linux, and Windows.

A tool for packaging machine learning models into containers. It automatically packages required environments and dependencies via configuration for seamless deployment, eliminating the hassle of writing Dockerfiles and managing CUDA, while also launching an HTTP API for easy invocation.

A WebUI for the Stable Diffusion model. This project enables browser-based use of Stable Diffusion, supporting features such as text-to-image and image-to-image generation, text embeddings, and image resizing.

A collection of deep learning paper implementations, featuring PyTorch implementations of neural networks and related algorithms with line-by-line code annotations.

A Chinese pre-trained version of the OpenAI CLIP model. Trained on a massive Chinese dataset of ~200 million image-text pairs, this project provides pre-trained models of various sizes and a technical report, enabling users to perform Chinese image-text feature extraction and retrieval with just a few lines of code.

一款可以给老旧照片上色的项目。该项目基于深度学习实现了对黑白图像和视频进行着色的功能。

一款轻量级的机器学习可视化工具。该项目是用于可视化和跟踪机器学习实验的工具，通过几行代码就可以实现跟踪、比较和可视化机器学习实验。

快速上手 AI 理论及应用实战。该教程包含学习 AI 必备的数学基础，机器学习实战小项目、深度学习入门、自然语言通用框架 BERT 实战，以及大量数据集。

Netflix 开源的视频质量评估算法。一种将人类视觉模型与机器学习结合的评估视频质量的方法，目的是改善观众们的观看体验。

根据文本生成 3D 人体运动的模型。论文《Human Motion Diffusion Model》第一作者开源的 PyTorch 实现。

OpenAI 开源的多语言识别系统。该项目是强大的自动语音识别系统，支持包括中文在内的多种语言识别。尤其是在快语速、口音、背景噪音等场景，依旧表现出色，能够达到极高的准确率。

可以根据文字生成图片的模型。只要输入一段文字描述，就能得到一张由 AI 生成的图片，除此之外该项目还支持将粗糙的草图转化成精致的艺术图片。

自建 GitHub Copilot 服务。它采用 NVIDIA 的 Triton Inference Server 的 SalesForce CodeGen 模型，自建 AI 编码辅助服务。支持接入 VSCode Copilot 插件，使用起来十分方便。

OpenMMLab 开源的目标检测工具箱。基于 PyTorch 的目标检测开源工具箱，支持 Faster R-CNN、Mask R-CNN、RetinaNet 等主流算法。

简单易用的 NLP 框架。基于 PyTorch 的 NLP 框架，支持文本命名实体识别(NER)、词性标注(PoS)、词义消歧和分类。项目中包含详细的使用教程，介绍了如何标记文本、训练语言模型等。

根据文字生成图片的 AI 模型。可以按照文字提示自动生成图片，我试了下感觉生成的图片有些“抽象”。

解释机器学习模型输出的库。它基于博弈论中的 Shapley Value 理论，将所有特征视为贡献者，然后计算每个特征对于模型输出结果的贡献。

全面的机器学习教程库。一份包含 30 多个 Jupyter Notebook 的集合库，内容涵盖机器学习所需的 Python 基础，数据操作、清洗、分析、可视化常用的库和工具，以及经典机器学习、NLP、计算机视觉等算法，一份面面俱到的机器学习入门教程

用 SQL 开启机器学习的数据库。把机器学习引入 SQL 数据库将模型作为虚拟表（AI-table），从而省去了数据准备、预处理等步骤，可以直接用 SQL 查询时间序列、回归、分类预测的结果，实现简化机器学习开发流程的效果

拯救词穷的字典。由清华大学 NLP 实验室开源，可以根据你的意思返回相关词汇，有效解决词穷、话到嘴边说不出来的窘境。[在线尝试](https://wantwords.net/)

加速训练 NeRF 模型的项目。英伟达开源的最快只需 5 秒，训练出一只狐狸 NeRF 模型的技术。即根据静态的 2D 图片通过神经网络建模，快速训练出可以放大且从任何角度观察都清晰的图片