Trending AI projects

ZLUDA is a drop-in replacement for CUDA that enables unmodified CUDA applications to run on non-NVIDIA GPUs with near-native performance.

A Python speech-to-text library that provides fast, real-time transcription with integrated voice activity detection and wake word support.

A self-hosted, OpenAI-compatible LLM router that automates model selection based on cost and capability, providing a unified API for multiple AI providers.

Jeff is a 0.8 B “System 1” model that makes ultra‑fast, calibrated classification decisions. It uses lightweight LoRA adapters for specific domains (guard, triage, spam, etc.) and can act as a fast‑path filter in front of a large LLM (e.g., Qwen 3.8‑27B), handling most queries in ~30 ms and delegating only uncertain cases. Weights and adapters are open on Hugging Face, GGUF files enable llama.cpp use, and a simple HTTP/Python API (`jeff‑serve`) lets you plug Jeff into any service.

A coding-agent workflow that reduces failures by 45% by using a multi-layered agent hierarchy to review, fix, and recheck work.

CAD-MCP is a Model Context Protocol server that connects AI assistants like Claude or Cursor to CAD software (AutoCAD, GstarCAD, ZWCAD). It exposes 35 tools for drawing, editing, querying, and managing layers and blocks, so users can create and modify drawings by typing plain-language commands instead of using the CAD interface manually.

A 15TB collection of diverse numerical physics simulations and a benchmarking suite designed to train and evaluate machine learning surrogate models for computational sciences.

A curated, source-linked library of copy-ready prompts for AI video and image models, each with real output previews, tags, and practical editing recipes.

apple‑hig is a skill for AI agents that provides a compact, source‑reviewed copy of Apple’s Human Interface Guidelines and a `/duo` slash‑command that walks developers through adapting an existing iOS app for the new folding iPhone Duo. It also offers an optional “enforce” review loop. Install by copying the provided prompt or manually placing the `apple-hig/` folder in the agent’s skill directory.

An AI-powered divination platform that uses ChatGPT to perform Tarot readings, astrology analysis, and dream interpretation with a responsive, user-friendly interface.

A framework for building a persistent, AI-driven operating system using Claude Code, organized into four layers: Memory, Agent, Pulse, and Screen. It enables automated routines, intelligent file management, and a read-only dashboard to keep your work organized and accessible.

Tokless is a zero-config pipeline that automatically wires together specialized coding tools and memory management for AI coding agents, ensuring they work efficiently across various platforms.

An MCP server that enables AI coding agents to control and interact with Roblox Studio, allowing for automated debugging, script execution, performance profiling, and asset management.

A technical book providing a deep architectural analysis of the Agent Harness, the runtime framework that powers AI agents, covering everything from conversation loops and tool systems to memory management and multi-agent orchestration.

Audio Separator is a Python library/CLI that uses deep‑learning models (MDX‑Net, VR, Demucs, MDXC) to split music files into stems such as vocals, instrumental, drums, etc. It supports CPU, CUDA, Apple Silicon (MPS/CoreML), and experimental DirectML, offers Docker images, and provides a model catalog with filtering options.

Fara1.5 is a family of native computer-use agent models that can visually perceive and interact with web browsers using mouse and keyboard inputs to automate complex tasks.

OpenCompany is an open‑source desktop app that lets you create autonomous AI “employees” – small teams of LLM‑driven agents – that can read email, write code, post to social media, handle payments, etc. You define a job, drag specialist agents onto a visual canvas, connect them to tools (Gmail, GitHub, Stripe, browsers, Android emulator, …), and the employee runs locally (or in Docker) using any LLM you provide. State, memory, and skills are editable, and the system works without a subscription.

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.

A prompt-based toolkit that turns AI assistants into learning companions by forcing users to think and reason through code rather than just generating solutions.

BallonsTranslator is a desktop application that automates manga translation using OCR, AI‑based translation (including LLMs), and image‑inpainting, then lets users fine‑tune the result with visual and rich‑text editors.

DSH‑X is a community‑built desktop launcher that helps you install, run and manage multiple local instances of DeepSeek Harness, with version control, plugin/pack handling, MCP‑skill integration and a themed UI for Windows and macOS.

A high-performance Core ML runtime for Laya models that enables low-latency, energy-efficient structured decision-making on Apple Silicon without token generation.

A local proxy that lets Claude Code use multiple LLM providers (Codex, Kimi, Grok, OpenCode Go, Cursor) by translating Anthropic‑style API calls, handling auth, routing, streaming and offering a TUI monitor.

CuPy is a library that provides a drop-in, GPU-accelerated alternative to NumPy and SciPy, allowing for faster numerical computing on NVIDIA and AMD hardware.

A Python-based UI framework that allows developers to create web-based interfaces for robotics, ML, and dashboards without writing HTML or JavaScript.

A community-maintained, work-in-progress list of ~64 open-source AI hacking/security agents, sorted by release date, with docs links, research paper references, and some benchmark notes — aimed strictly at educational and authorized testing use.

A cross-platform GUI application for AI-powered video frame interpolation and upscaling, supporting multiple GPU backends and a variety of specialized models.

FluxVLA Engine is an open‑source, end‑to‑end platform for training, evaluating, and deploying vision‑language‑action (VLA) models for robots. It offers unified configuration, modular model back‑ends, benchmark suites (LIBERO, RoboCasa, RoboDojo), and deployment tools (Docker, Jetson Orin, Triton, ZMQ) with real‑robot demos on Franka and Oli.

A runtime for running applications and AI agents in secure, lightweight sandboxes across local, cloud, and browser environments.

An open-source, MIT-licensed tool that removes Gemini's visible watermark from AI-generated images and videos using a mathematically exact Reverse Alpha Blending algorithm (not AI inpainting). It processes everything locally in the browser or via CLI, offers multiple integration methods (online tool, Chrome extension, userscript, CLI, SDK), auto-detects watermark position, and supports both images and videos. It only handles Gemini's specific visible watermark, not invisible ones like SynthID, and comes with legal and usage warnings.

A curated, open-source library of modular AI agent skills designed to help legal professionals automate tasks like contract review, regulatory compliance, and legal drafting with expert-vetted workflows.

AI Collection is a curated directory of 3,569 generative AI applications across 80 categories, with descriptions, screenshots, links, and pricing. It includes top picks, latest additions, translations in 5 languages, and a companion website (thataicollection.com). MIT-licensed with ~9.2k GitHub stars.

A structured, code-first educational resource that explains how to build the 'harness'—the software infrastructure required to turn LLMs into functional, autonomous AI agents.

Strands Box is an open‑source sandbox that isolates AI agents and enforces Dogwood‑written semantic policies on file, program, and network actions, with unified rule evaluation, credential injection, and audit logging.

An AI Agent skill that redesigns GitHub READMEs by optimizing content hierarchy and creating project-native SVG visuals to make a project's value immediately clear.

pi‑vcc is a deterministic, algorithm‑only conversation compactor for the Pi coding‑assistant. It replaces Pi’s LLM‑based summariser with fast extraction of goals, file changes, commits, commands, open issues, and user preferences, reducing session size by ~98 % with millisecond latency and no token cost. The plugin works automatically or via `/pi‑vcc` commands, keeps the compacted history searchable through `vcc_recall`, and is configurable via a JSON file. MIT‑licensed.

A collection of SQL and AI-based models for detecting financial risk, fraud, and regulatory compliance issues in lending and corporate finance.

Izwi is a local-first voice AI suite (desktop app, web UI, CLI, and server) that runs speech recognition, text-to-speech, chat, diarization, and voice cloning entirely on your machine, exposing an OpenAI-compatible API without needing cloud services or API keys.

Apache Doris is an open‑source, MPP‑style analytics database that delivers sub‑second SQL queries on streaming, lakehouse, and hybrid (text + vector) data, making it suitable for real‑time dashboards, observability, and AI‑driven similarity search.

apfel is a Swift‑based CLI and local OpenAI‑compatible server that exposes Apple’s built‑in Foundation Models LLM on Apple‑silicon Macs. It runs 100 % on‑device, supports pipe‑friendly prompts, JSON/schema output, tool‑calling via MCP, and a chat REPL, making it useful for private scripting and local AI development.

An MCP server that uses tree-sitter AST parsing to let AI agents retrieve precise code symbols instead of entire files, reducing token costs by up to 99%.

MAMMA is a markerless multi-person 3D motion acquisition system that reconstructs accurate human poses from multi-view video footage.

A native, GPU-accelerated SSH client and remote operations workspace that integrates terminal, file management, and AI-assisted automation into a single, resource-efficient application.

A free and open-source LLM fine-tuning framework that streamlines post-training for text, vision, and audio models using a single YAML configuration.

A library providing state-of-the-art parameter-efficient fine-tuning methods to adapt large pretrained models using minimal compute and storage.

A standardized AI skill that distills traditional Feng Shui knowledge into a conversational consultant for environmental analysis and auspicious date selection.

A local, AI-powered workbench for math teachers to digitize question banks and typeset professional A4 exam papers using LLMs and LaTeX.

A self-hosted media manager that integrates searching, downloading, tagging, and organizing for music, movies, and TV, replacing multiple separate tools with a unified pipeline.