Trending AI projects

Deep-Live-Cam is an AI-powered tool that enables real-time face swapping for live video streams and recorded media using only a single reference image.

一个自托管的 OpenAI 兼容反向代理网关，将腾讯 CodeBuddy 账号包装成统一 API，支持多账号池调度、会话粘性、自动签到/成长任务、Web 管理面板等，让开发者用标准 OpenAI 客户端调用 CodeBuddy 能力。

A tool that protects websites from AI scrapers by serving them poisoned training data and self-referential links to degrade the quality of AI models.

HyperQwen is an optimized serving stack for Qwen models that enables high-speed inference and long-context support on consumer GPUs through custom vLLM patches and speculative decoding.

ReelMimic is a local, agent-based animation tool that analyzes the style and pacing of a reference video to help you create original 2D animations through an iterative, multi-agent production pipeline.

A collection of 12 Claude Code / Codex “skills” that let the Claude AI write, audit, and (optionally) publish LinkedIn posts, comments, replies, profile text, and related marketing assets. The bundle works via Claude’s plugin marketplace, Codex CLI, OpenClaw, Hermes, or any agent that reads `SKILL.md`. Optional integrations with Apify (scraping), Publora (auto‑posting), and Pixfaro (image generation) add data fetching and media creation. Designed for founders, marketers, and teams, it includes a founder‑specific angle library and a 7‑day content planner. MIT‑licensed and installable with a few CLI commands or through Claude’s UI.

EdgeRock‑4B is an open‑source 4 B‑parameter multimodal model that achieves strong text‑and‑image reasoning while keeping its attention memory fixed (72 MiB per sequence). It uses a recurrent summary, a 1 k‑token sliding window, and a learned “Rock Memory Anchor” bank to retrieve distant information. On a single Nvidia Hopper 80 GB GPU it decodes at 233 tokens/s for 128 k‑token inputs and reaches 5.6× the throughput of a comparable full‑attention model. Benchmarks show 82.9 % mean pass@1 on AIME, 77.2 % average on 29 text tasks, and 84.9 % visual accuracy, all with constant‑size memory.

OpenChamber is an open‑source, cross‑platform workspace that lets developers run AI coding agents, set automated goals, compare multiple model runs, review changes with guided walkthroughs, and ship code—all from desktop, web, VS Code, or mobile.

An automated video generation tool that uses AI agents to convert scripts into stylized, high-quality educational videos by rendering frames via code and browser automation.

An open-source knowledge base and learning roadmap for AI Agent engineering and research, designed to help developers move from basic LLM usage to building production-grade agents and securing professional AI roles.

Karakeep is a self‑hosted bookmark‑everything app that stores links, notes, files and archives, adds AI‑driven tagging/summarisation (via OpenAI or local Ollama models), provides full‑text & semantic search, and offers browser extensions, mobile apps, a CLI and REST API for flexible use.

A family of open-source text-to-speech models providing low-latency voice cloning and multilingual speech generation across various model sizes for GPU and CPU deployment.

A monitoring system that detects emerging AI models and tools by tracking signals from X, GitHub, and Hugging Face before they appear in traditional keyword tools.

A comprehensive library of 98 specialized skills that enable AI agents to autonomously conduct AI research, from initial ideation and experiment execution to paper writing.

A personal, Chinese‑language repo documenting prompt engineering, jailbreaks, memory tricks, and tooling for using LLMs in adult‑oriented, long‑term human‑machine relationships.

A tool that removes over 20 common patterns of generic AI-generated writing to help authors maintain a personal, human voice in their AI-assisted drafts.

GBrain is a self‑hosted memory daemon that adds a citation‑rich, editable knowledge base to any LLM‑powered agent. It stores facts with sources, supports corrections, and can be queried via keyword, semantic search, LLM‑driven synthesis, or a typed knowledge graph. The brain can be attached to existing agents (Claude Code, Codex, Grok Bot, Muse, etc.), run as a standalone service exposed over MCP/Tailscale, or used as a shared corporate brain with fine‑grained permissions. Installation is via Bun (`bun install -g github:garrytan/gbrain`) and a one‑command `gbrain setup <harness>` that wires the daemon into the agent. It enables agents to answer questions directly (e.g., meeting prep) with sourced, synthesized replies, turning raw document search into actionable knowledge while keeping data under your control.

Claude‑Obsidian is a Python‑based, local‑first toolkit that integrates Claude Code (or any Agent Skills‑compatible LLM) with an ordinary Obsidian vault. It captures sources, creates linked Markdown notes with provenance metadata, and lets you query the vault for evidence‑backed answers. All operations are preview‑first, atomic, and fully traceable, keeping your files under your control.

A desktop application that automatically turns documents into a structured, interlinked personal wiki using LLMs, featuring a knowledge graph and automated deep research.

Sno Station is a local-first orchestration layer that connects multiple AI coding agents into a single, memory-sharing squad capable of handovers, peer reviews, and automated self-improvement.

screenshot‑to‑code is an open‑source app that converts UI screenshots, mock‑ups, or short screen recordings into ready‑to‑run front‑end code (HTML/CSS, React, Vue, etc.) using LLMs such as Gemini, GPT‑5.5, and Claude, with optional asset extraction and visual preview.

A minimal, hackable harness for training LLMs end-to-end on a single GPU node, designed to produce compute-optimal micro-models for under $100.

GitDiagram is an AI-powered tool that automatically generates interactive architecture diagrams and narrated explainer videos for GitHub repositories, providing an MCP server for AI agents to query codebase structures.

OpenBitFun is a cross-device, highly customizable agent workspace that allows users to manage, extend, and run AI agents across desktop, mobile, and remote environments.

An open-source inference engine that compiles and tunes kernels on your specific hardware to run open-weight models up to 2x faster than llama.cpp.

AI Safety HOT是一个持续更新的AI安全信息聚合平台，追踪新闻、论文和事件，提供可视化观测台和MCP接口，让用户和Agent一站式获取AI安全动态。

An open-source library of ready-to-use AI writing prompts and agent skills for academic papers, gathered from top researchers and students. It provides copy-paste templates for translation, polishing, logic checking, de-AI-izing, and more, plus guides for agent skills.

A universal provider proxy that lets you run any LLM with Codex, Claude Code, Claude Desktop, and Grok Build by translating their APIs.

An open-source robotic arm platform designed for learning and developing Embodied AI, featuring full hardware blueprints and deep integration with ROS2, Isaac Sim, and LeRobot.

An open-source Python framework for orchestrating multi-agent AI systems, combining autonomous collaborative crews with precise, event-driven workflows.

A tiny, educational autograd engine that implements backpropagation over a scalar-based DAG to help users understand the internal mechanics of neural networks.

Trail (formerly Graft) is a TypeScript CLI that creates a local markdown‑based graph of a repository—summaries, key code excerpts, and typed links—then injects this knowledge into coding agents (Claude Code, Cursor, etc.). By doing the heavy lifting once, it reduces token usage, tool calls and latency (up to 4× cheaper, 3× faster) while preserving or even improving correctness. The tool works locally, supports 23 languages, integrates with many LLM providers, and saves team “learnings” for future sessions.

iPhone Use lets coding assistants like Codex, WorkBuddy, or 豆包工作 control your real iPhone over USB using natural language. It installs and manages WebDriverAgent, provides MCP tools for tapping, scrolling, typing, and reading screens, and shows a live phone screen in the sidebar.

Agenvoy is a self-hosted, open-source AI agent that runs on your computer, turning conversation into completed work—research, file operations, automation, and scheduling—while keeping your data local and sharing sandboxed tools with other AI agents via MCP.

A curated directory and automated monitoring dashboard for AI API relay services that provide free credits and trial access to various language models.

Open‑Claude‑UI is a reverse‑engineered, MIT‑licensed React component library that faithfully reproduces Anthropic’s Claude desktop UI. It provides 30 ready‑to‑use, high‑fidelity components, a Tailwind theme, motion presets, Web‑Audio synthesis, macOS desktop integrations, accessibility hooks, and an offline Storybook preview, all documented with a detailed reverse‑engineering guide.

Claude Cookbooks is a collection of code snippets and guides that help developers integrate Claude into their applications, covering everything from basic text tasks to advanced tool use and multimodal capabilities.

**深入理解 AI Infra** is an open-source (Apache-2.0) book by Bojie Li on AI infrastructure — the hardware, networks, and systems software that power LLMs — using a “derive design from constraints” method centered on data movement (five questions: what, how much, how many times, through where, who waits). It has 12 chapters, a Python calculation tool for resource estimates (no GPU needed), per-chapter experiments, and is available in four languages (Chinese original, English, Traditional Chinese, Russian) as PDF/EPUB/online. Motivating metaphor: LLM is the new “operating system,” AI Infra the new “computer architecture.” Repository uses Git LFS (~20 GB, but default clone skips those files); web/PDF/EPUB are built from Markdown sources in `manuscripts/`.

AnythingLLM is a self‑hosted, full‑stack chat application that lets you connect any LLM (local or hosted), ingest your own documents, and interact via a web UI with AI agents, multi‑user support, and optional Docker/desktop deployments.

PolyStrata is a free, open-source server that runs large AI models like Qwen3.6 and GLM 5.3 on a normal PC, with speeds faster than llama.cpp and support for OpenAI and Anthropic APIs. It keeps data on-device and works with coding agents like Claude Code.

Distilly is an open‑source Python project that builds “Person Profiles” from messages, documents, or public content and packages them as reusable Agent Skills. It supports three person families (colleague, relationship, celebrity), integrates with many AI‑agent platforms (Claude Code, Hermes, OpenClaw, Codex, DeepSeek, Pi, Grok Build, OpenCode), and offers versioned, source‑grounded profiles for use in personal or team agents.

Codync is an open-source platform that turns various AI coding agents into persistent, messageable bots, allowing you to manage and delegate coding tasks from your phone or desktop.

A Claude Code plugin that adds a customizable, real-time status dashboard to your terminal, displaying context usage, rate limits, active tools, and agent progress.

A high-performance implementation of OpenAI's Whisper speech-to-text model that uses the CTranslate2 engine to achieve faster inference and lower memory usage.

ACE‑Step 1.5 is an open‑source music‑generation foundation model that combines a language‑model planner with a Diffusion‑Transformer audio decoder. It can create full songs (10 s‑10 min) in under 10 seconds on a RTX 3090, runs with <4 GB VRAM for base models, and supports LoRA fine‑tuning from a handful of tracks. The repo provides Gradio and REST interfaces, launch scripts for all major OSes, multiple model sizes (including a 4 B XL version), and extensive control over lyrics, style, tempo, and instrumentation.

An autonomous AI penetration testing system that uses a multi-agent architecture and dual-graph tracking to automate the discovery and exploitation of security vulnerabilities.

Claude Ads is an open‑source AI skill for Claude Code that audits ad‑account data from twelve major platforms, producing evidence‑backed pass/fail reports and optional drafts for planning, creative, monitoring, optimisation and testing. It is read‑only by default, runs via simple slash commands, and includes an optional UI companion (Cockpit). The project is in development‑candidate status and distributed under the MIT license.

Reason Network is a peer‑to‑peer daemon that lets AI coding assistants (Claude, Codex, Cursor, Gemini, etc.) on different developers’ machines discover each other, share status, and exchange messages instantly, all without a central server.