Trending AI projects

An AI Agent support library for E-language that enables AI to automatically read and edit source code via a local MCP server and integrated IDE tools.

A pipeline that converts locked-camera videos into clean FK animations for Mixamo characters, removing the need for mocap suits or manual keyframing.

director is an AI‑agent skill that guides a language model through the full video‑production pipeline—script, visual style, characters, voice, shot‑by‑shot prompts—and uses LibTV or 即梦 CLI tools to generate ready‑to‑edit video clips. It supports three modes (Animated Explainer, Storytime Animation, Cinematic Drama) and is installed as a skill directory for agents like Codex.

An open, intuition-first textbook covering maths, computing, and AI across 18 chapters, born from the author's personal notes that helped friends land roles at DeepMind, OpenAI, and Nvidia. Includes an MCP server for AI assistants, a learning philosophy aimed at "the Darwins of the world," and a two-phase study technique.

Scout is an open‑source TypeScript library that lets you give a natural‑language web task to an AI‑driven browser agent. It works with any chat‑completion model you provide, runs on Outcrawl’s cloud (residential IP, captcha solving, replay recordings) or locally via Playwright, and returns structured results plus a step‑by‑step replay.

An open-source Python package for exploring, visualizing, and analyzing human neurophysiological data like MEG and EEG.

Gaze is an on‑device facial authentication suite for Linux, offering a daemon, CLI, GUI, PAM module, and desktop‑environment extensions. It runs face detection (SCRFD), alignment, ArcFace embedding, and MiniFASNet‑V2 liveness checks locally via ONNX models, supports TPM‑backed template sealing, and integrates with GNOME, KDE, Cinnamon, and other lock screens. Installable via native packages or a one‑line script, it can use CPU or hardware accelerators (OpenVINO, AMD NPU) and is fully configurable through `/etc/gaze/config.toml`.

A curated collection of academic papers on humanoid robot learning, organized by task and highlighting research with real-world experiments and open-source code.

A Chinese-language, community-maintained guide collecting ChatGPT and AI resources — from prompts and apps to model ecosystems, development guides, and frontier topics — with a Telegram community and open contribution model.

HERO is a paste-in rule block for coding agent config files that stops AI assistants from over-defending — adding unread hashes, chasing impossible edge cases, replacing judgment with checklists, and building unneeded scaffolding — so they spend their work on the task actually asked for. It works with Claude Code, Codex, Cursor, Copilot, and other agents that load a config file.

minutes is an open‑source desktop/CLI app that records, locally transcribes, and stores meetings as Markdown files with rich metadata. It exposes a simple MCP API (`search_meetings`) so any LLM agent (Claude Code, Cursor, Codex, etc.) can retrieve relevant excerpts, enabling privacy‑first, agent‑first workflows.

Kiln is an open‑source, local‑first workbench (desktop app + Python library) that lets teams build, evaluate, optimize, and ship LLM‑based products. It covers eval creation, prompt auto‑optimization, RAG pipelines, multi‑agent hierarchies, synthetic data generation, and zero‑code fine‑tuning across 190+ models, with Git‑based collaboration and offline support.

Harness Books is a project publishing two free online books about harness engineering for AI coding agents. Book 1 focuses on Claude Code's runtime structure; Book 2 compares Claude Code and Codex. The books emphasize that the real challenge is managing behavioral consequences, not answer quality. They cover prompts, permissions, context, recovery, verification, and team rules. The project includes source files and build tools, and links to a related practice platform called AgentWay.

A curated, annotated list of 443+ verified resources for building and evaluating AI agents, with a runnable code playbook, deep reading notes, and a must-read starter set.

vbsec is a security scanning tool for AI-generated code that identifies 21 common vulnerabilities by analyzing data flow, providing clear explanations and fixes for developers.

Millennium Harness is a formal-verification gate for AI-generated proofs in Lean and Dafny that rejects all cheats (like `sorry` or `assume {:axiom}`) by checking the frozen specification still holds and no escape hatch is used. It also powers Millennium RSI, a self-improving agent loop where the reward cannot be gamed.

AGY2API is a self-hosted gateway that exposes Google Antigravity models (like Claude and Gemini) through OpenAI, Gemini, and Claude compatible APIs. It manages multiple Google OAuth credentials, supports streaming, and provides a web control panel for configuration and credential management.

Godot-MCP is an AI-powered assistant for the Godot Editor that allows AI agents to control the game engine, enabling them to edit scenes, manage scripts, and debug projects through the Model Context Protocol.

Comfy MCP is a server that connects AI agents to your local ComfyUI installation, allowing you to trigger image generation, manage workflows, and monitor your server directly through your AI assistant.

An open-source alternative to OpenAI Dots that lets you run personal background agents on your Mac using your own API keys and app integrations.

GreptimeDB is an open-source observability database that unifies metrics, logs, and traces into a single columnar engine, allowing users to store data on object storage and query it with SQL.

An open-source robot simulation and training framework that uses GPU parallelization to enable high-throughput data collection and training for manipulation skills.

An open-source research framework for universal metric 3D reconstruction that uses a single transformer model to handle multiple 3D tasks from various input combinations.

MLC LLM is a universal deployment engine and ML compiler that enables high-performance native execution of large language models across diverse hardware and platforms.

A declarative YAML-based framework that wraps SQL to create auditable, version-controlled dashboards and reports, specifically optimized for both human auditors and AI agents.

DuckLake is a DuckDB extension that enables an open Lakehouse format by storing metadata in a relational catalog and data in Parquet files, supporting SQL-based updates, time travel, and schema evolution.

A set of Node‑based plugins (mods) for Anthropic’s Claude Code that add profiling, context analysis, timeline snapshots, loop detection, pinned notes, and model‑effort visualisation. Each mod is installable via Claude’s marketplace and includes unit tests and TypeScript type‑checking.

Qt AI Skills is a repository of packaged “skills” that let LLM‑based coding assistants (Claude Code, Codex CLI, GitHub Copilot, Gemini, etc.) perform Qt‑specific tasks—code reviews, documentation generation, test creation, profiling, and design‑to‑code conversion—through a standardized `SKILL.md` format. The repo also provides a small MCP server for fast Qt API look‑ups. Installable directly into supported AI tools, these skills give Qt developers deterministic, Qt‑aware automation without leaving their AI‑assisted workflow.

Funplay MCP for Cocos is a Cocos Creator editor extension that embeds an HTTP MCP server, enabling AI assistants (Claude, Cursor, Copilot, etc.) to run JavaScript in the scene or editor, inspect assets, capture screenshots, and automate workflows via JSON‑RPC.

RSIAgent is a research‑grade, training‑free framework that lets LLMs recursively improve by exploring digital environments. Three agents (Curriculum, Actor, Verifier) run a broad‑then‑deep self‑improvement loop, storing distilled procedures in persistent memory that is frozen for downstream tasks. The repo includes full OSWorld‑V2 and Agents’ Last Exam (ALE) benchmark integrations, scripts for setup, validation, and batch execution, and documentation of the architecture and results.

A real-time webcam application that detects facial expressions and hand gestures to trigger matching meme overlays, featuring a virtual camera for use in video calls.

A lightweight reverse proxy that translates Command Code API requests into OpenAI and Anthropic-compatible formats, featuring built-in device fingerprinting and memory management.

DeepAnalyze is an open‑source, agentic LLM (8 B parameters) that can run the entire data‑science workflow—cleaning, analysis, modelling, visualisation, and report generation—without human intervention. The repository provides the model, a 500 K instruction dataset, and ready‑to‑run demos (Web UI, Jupyter UI, CLI) plus an OpenAI‑compatible API, all deployable via vLLM.

A Node‑JS server that adapts OpenAI Codex to the Agent Client Protocol, exposing Codex’s code‑generation capabilities through ACP’s standardized request/response format and supporting advanced AIR extensions, sub‑agent sessions, and rich I/O.

A versatile AI-powered tool that translates books and documents into bilingual versions by leveraging various LLMs and translation services while preserving original file formatting.

A collection of Python sample codes and documents providing a technical guide for beginners to learn autonomous vehicle control algorithms and software architectures.

A Codex Skill that converts business documents and research data into high-density, editable, consulting-style PowerPoint presentations using a structured analysis and blueprinting process.

A marketplace of Claude Code plugins (skills) with a production‑hardened meta‑skill (`skill‑creator`) that helps you build, validate, and package new skills, plus dozens of ready‑made skill suites for docs, macOS/iOS dev, financial data, GitHub ops, and Claude Code power‑user workflows.

An AI video prompt analysis tool that reverse-engineers existing videos to extract prompts, helping creators replicate and adapt successful AI video styles and scripts.

A high-performance gradient boosting library for decision trees that provides native support for categorical features and fast prediction speeds.

A self-hosted, local-only NVR and AI computer vision software for home and office monitoring with object and face recognition.

A collection of specialized skills and tools for AI agents to handle French accounting, taxation, and administrative bureaucracy.

A scalable Kubernetes scheduler that optimizes GPU resource allocation and fairness for large-scale AI and machine learning workloads.

PageLM is an open-source AI education platform that converts documents and lecture notes into interactive study tools like quizzes, flashcards, and AI-generated podcasts.

TE MAN is a comprehensive ComfyUI plug‑in that adds unlimited‑canvas workflow, prompt‑enhancement nodes (Qwen, MiniMax H3), audio/video editing, a 3‑D director panel, an asset library and a chat‑assistant “构想台”. It works with local diffusion models and cloud APIs, targeting artists and prompt engineers who want an all‑in‑one UI for image, video and audio generation.

A platform providing persistent and secure sandboxes for AI agents powered by Firecracker microVMs to enable safe code execution.

A browser-native, local-first slide editor that uses Web AI and WebGPU to provide AI-powered slide generation, translation, and presentation tools without a backend.

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.