Trending AI projects

ARTEX‑Global is an open‑source, LLM‑driven autonomous penetration‑testing platform (Go backend + embedded Next.js UI) that runs multi‑agent attack cycles, stores results in PostgreSQL, and supports English, Korean, Chinese, and Spanish output. It ships with common security tools, a traffic‑recording proxy, and defensive detection guides, and is released under AGPL‑3.0 with strict authorized‑use warnings.

赛博小雅直播版是一个免费开源的数字人直播软件套件：浏览器操作页 + 本地桥接服务 + Vidu 云端渲染，观众弹幕驱动数字人说话和做动作。它提供完整代码但不含任何个人素材（形象/音色/人设/API Key），需要用户自备，约 1 小时可跑通。

A collection of specialized skill sets for AI coding agents that provides best practices, documentation, and command-line capabilities for building video projects with Remotion.

An open-source proxy that provides security, observability, and governance for agent-to-LLM, agent-to-tool, and agent-to-agent communication.

gpt‑instruct is a Python‑based toolkit that bundles versioned prompt files for Codex models and provides scripts to deploy, test, and roll back those prompts using a structured A/B/C (and jailbreak) evaluation pipeline.

AI‑Infra‑Guard (A.I.G) is an open‑source, Docker‑deployable platform for red‑team security testing of AI services. It scans live LLM endpoints, AI‑agent workflows, and model‑control‑plane code for CVEs, jailbreak susceptibility, and agent loss‑of‑control risks, exposing results via a web UI and REST API.

An end-to-end AI production workspace for short dramas that manages the full pipeline from script breakdown and asset consistency to video generation.

YuE2 is an open‑source music‑generation system that turns lyrics + style prompts into an editable symbolic score and then into high‑fidelity audio. It offers three modes (full plan, melody‑only, direct), supports zero‑shot covers via SheetSage2 transcription, and achieves state‑of‑the‑art benchmark scores. The Python pipeline, agent skill, and web demo make it easy to create, edit, or cover songs, all under an Apache 2.0 code license and CC‑BY‑NC model license.

A web-based management dashboard for Tencent CodeBuddy account pools that provides a visual interface for account onboarding, API key distribution, and usage monitoring.

DSH Pocket is a plugin for DeepSeek Harness that enables remote mobile access to your local AI agent interface via secure LAN or public internet tunnels.

Paradev is an open-source desktop app that lets you run multiple AI coding agents in parallel, each in its own workspace with terminals, diff review, and browser previews built in. It supports many common agents out of the box and works on macOS (primary) with experimental Linux support and an iOS companion app.

A handbook of AI side-hustle ideas, resources, and articles for earning extra income, with contribution via GitHub issues.

A community tool that creates multiple independent Claude Desktop windows, each able to use a different API (official Anthropic or third‑party via cc‑switch). It provides per‑window icons, shortcuts, isolated data folders, and shared session syncing on Windows (experimental macOS support).

An AI-powered reverse-engineering agent that uses Ghidra and LLMs to reconstruct and validate C/C++ functions from compiled binaries through an autonomous pipeline.

AI Novel Writer is a desktop app that orchestrates AI‑assisted long‑form fiction writing. It lets you configure any OpenAI‑compatible or Gemini model, manages a structured workflow (premise → characters → outline → chapter blueprint → draft → review → revision → final), stores all project data locally, and offers features like chapter‑level generation, failure recovery, automated review, batch tasks, and a frozen DeepSeek Harness web plug‑in.

An agent skill and design framework for AI coding agents to build premium, scroll-driven websites that avoid generic AI design tropes through strict craft standards and automated verification.

AutoCua is a multi-agent harness that lets AI models operate real computers—desktop apps, browsers, shell terminals, and iOS devices—by using real input events and accessibility tree parsing. It solves the problem of building agents that work across a wide range of software without custom integrations.

A repository of seven AI agent skill bundles that automate cross‑border ecommerce workflows (market research, product strategy, listing creation, image/video generation, ad optimization, review analysis). It provides prompts, templates, small calculation scripts, compliance guards, and end‑to‑end examples, all runnable locally with any LLM. The project is on‑topic for AI/ML tooling.

A browser-based design tool for creating Material 3 Expressive app screens that generates precise prompts for AI coding agents to build the actual application.

Nakama is an AI agent platform that allows teams to manage, configure, and deploy multiple AI agents with unique identities and tool access through a unified dashboard.

A Prompt-as-Code system and template library for GPT-Image2, providing over 500 reverse-engineered cases and structured prompts for stable, controllable image generation.

Mercury Agent is an open‑source, locally‑run AI assistant that lives as a single background daemon. You can chat with it from a terminal, Telegram, Discord, Slack, Signal or a local web dashboard. It keeps persistent, SQLite‑based “second‑brain” memory, enforces per‑folder/file/shell permissions, respects a daily token budget, and can run a variety of built‑in tools (git, shell, web fetch, Spotify, etc.). Specialized sub‑agents called **Mercury Bots** can operate continuously or on schedules, delivering finished files to a folder or chat. The agent works with many LLM providers (OpenAI, Anthropic, DeepSeek, Ollama, Mercury Cloud, …) and falls back automatically. Installation is a one‑line script or an npm package, and the system runs on macOS, Linux, Windows and Android/Termux.

A desktop app and toolkit for running and training AI models locally, offering significantly faster fine-tuning with reduced VRAM usage on consumer hardware.

Accessibility-first browser & computer use for AI agents — one MCP server for every harness (Codex, Claude Code, OpenCode, Antigravity). Instruments the DOM/accessibility tree via CDP instead of screenshot loops: diffed snapshots, stable element refs, trusted input, typed errors. Zero dependencies

Agent2API is a local OpenAI-compatible gateway that aggregates login sessions from multiple AI desktop clients into one API endpoint. It lets you reuse your existing desktop app subscriptions through any standard API client, with automatic token refresh, multi-account rotation, and provider-specific adapters for 17+ platforms.

A domain-general framework that enables AI systems to recursively self-improve by using world models to predict action outcomes and learning from interaction feedback.

A curated collection of production-ready skills for AI coding agents like Claude Code and Cursor, providing specialized workflows for professional web design, video presentations, and image generation.

Cangjie Skill is a Python‑based pipeline that extracts actionable methodologies from books, videos, podcasts, etc., and packages them as reusable “AI Skills”. Using a deterministic RIA‑TV++ workflow it creates capability cards, links them in a graph, pressure‑tests them, and compiles either a single router skill or a compact skill pack. The output can be installed on platforms like OpenClaw, Claude Code or DeepSeek Harness, letting LLM agents invoke the distilled knowledge directly instead of just reading summaries.

OpenDots is an open‑source, self‑hosted template for building persistent AI agents (Dots) that can chat, make voice calls, browse the web, run shell commands, and integrate with Slack or Composio. Each Dot has its own isolated computer (via OpenBot), a human‑in‑the‑loop approval UI, and stores its work in searchable “Spaces” (pages). The stack uses CopilotKit runtime, AG‑UI, TanStack AI, and optional hosted or local CopilotKit Intelligence for LLM calls. It’s an alpha‑stage starter kit for creating specialised AI coworkers while keeping full control over infrastructure and data.

AutoSwarm is a Python framework that automatically builds, runs, and learns from a swarm of AI agents organized as a virtual company. It offers a CLI and a visual Office UI, supports adaptive team creation, Kanban‑driven orchestration, and experience‑based playbook generation, and integrates with external LLM back‑ends like JiuwenSwarm.

Internet Court is a single open skill that bundles ~35 protocols (vendored 92 official skills across 34 owners) so AI agents can discover, negotiate, contract, pay, execute, and — uniquely — *adjudicate* deals with humans out of the loop. Agents agree in advance on a dispute judge, written into the contract. One root SKILL.md routes to sub-skills on demand. Built by a consortium of ~30 companies, MIT licensed for first-party code, installable on 6 agent harnesses.

An open-source, Rust-based digital audio workstation that reimplements the Pro Tools workflow and is fully scriptable via CLI or AI agents.

ARTEX is an AI‑driven autonomous penetration‑testing system built with a Go backend, Next.js frontend, and PostgreSQL. It uses LLM‑powered agents (planner + workers) to explore targets, run real tools, and record findings in a dual‑graph model (global asset graph + per‑task exploration graph). The project provides Docker images, pre‑compiled binaries, and a one‑click install script, plus a mock demo UI. It supports asset sync from ScopeSentry, approval gating, self‑updating, and optional HTTP Basic Auth. Licensed under AGPL‑3.0.

A port of the pstack skill stack for AI coding agents that provides structured workflows and playbooks to ensure code is concise, simple, and verified.

A comprehensive data toolkit for the Chinese A-share market that aggregates 34 sources into a unified interface for AI programming assistants to perform financial analysis.

imozuru is a Claude‑powered research‑assistant skill that builds a source‑first “research map” for unfamiliar topics, crawling the web, following citations, and saving original documents so each claim is backed by a precise file‑and‑line reference.

CozyClay is a browser‑based pre‑visualisation studio built on Three.js/React‑Three‑Fiber. It lets you block a 3‑D scene, pose characters, animate cameras on a timeline, and export a keyframe pack that feeds directly into AI video generators (Seedance, Kling, Veo, MiniMax‑H3, etc.). Optional integration with the Kimodo motion‑generation service adds AI‑driven character animation. The app includes an AI chat panel (supporting ChatGPT, Anthropic, Gemini, OpenRouter, CLIProxy), an MCP server for external control, and a node‑based workflow canvas. Install with a single `npx cozyclay` command; no compilation needed.

Hyperfide is a local agent workflow engine that manages project context, task planning, and memory in a single binary to help AI agents work effectively on complex codebases.

UniEmployee is an enterprise-grade platform for building and managing AI digital employees by codifying business processes, professional expertise, and safety guardrails into observable, multi-step workflows.

OpenQodex is a local code review tool that combines static analysis scanners with AI agents to verify code changes before they are pushed to a repository.

A compact biped robot project providing a full pipeline from MuJoCo-based RL training to real-world deployment on a Raspberry Pi Zero 2 W using ONNX policies.

WinShun is a free, open‑source Windows search launcher that indexes files via the NTFS MFT, supports pinyin and full‑text search, adds clipboard history with OCR, integrates into file dialogs and the taskbar, offers plugins (including an LLM‑based semantic search), and provides an MCP server for AI assistants.

MonoCode is a unified GUI that aggregates various AI coding agents into a single desktop application, allowing developers to manage multiple agent sessions and project workflows in one interface.

QQ Agent is a local AI assistant for QQ that provides session management, memory, persona, and plugin capabilities, currently offering both a stable baseline and an ongoing architectural rewrite.

Explainer for miao-org/awesome-supermimic, a curated list of SuperMimic projects, grounded strictly in the README.

A reverse-engineered OpenAI-compatible API for muse.ai, enabling automated access to its multimodal text, image, and video generation capabilities.

Promptfoo is an open‑source CLI/library for evaluating LLM‑driven apps and red‑team‑ing them. It runs locally, supports many providers, produces comparison dashboards, and integrates with CI/CD and GitHub code‑scanning.

EAGLE‑4 is a source‑available library that adds adaptive speculative decoding (a “drafter”) to LLM serving engines (SGLang, vLLM, TokenSpeed). It automatically sizes the speculative tree based on batch size, hardware and workload, delivering up to ~3× token‑per‑second speed‑up on NVIDIA B300 or AMD MI355X GPUs. The repo provides the `eagle_harness` Python package, installation guides, pre‑built wheels/Docker images, and a support matrix with measured results. License is source‑available with commercial‑use limits.