Trending AI projects

A high-performance fork of llama.cpp that introduces advanced quantization types, experimental model support, and optimized inference kernels for CPU and GPU.

NaturalVoiceSAPIAdapter is a Windows SAPI 5 text‑to‑speech engine that lets any program use Microsoft’s neural (natural) voices—from Windows 11 Narrator, Edge’s Read Aloud, or Azure Speech Service—by installing a small DLL and optional Azure key.

FastFlowLM is a lightweight NPU-optimized runtime that runs large language models, including vision and audio models, on AMD Ryzen AI NPUs. It offers a simple CLI and server with OpenAI-compatible API, supports long contexts up to 256k tokens, and installs in under 20 seconds.

luobo-ppt is an Agent Skill that turns an article, topic, or plain PowerPoint into a natively editable, styled presentation. It generates decks from scratch, beautifies existing ones without changing a word, and delivers real editable .pptx files with strict quality checks and 14 built-in visual styles.

A tool that turns a single JSON configuration into animated, professional-looking architecture diagrams for social media, designed to be used as a skill for AI agents.

An open-source personal AI assistant platform that combines LLMs, RAG, and autonomous agent loops into a single self-hostable binary.

A multi-physics simulation package for modeling and simulating rigid and deformable bodies, granular dynamics, and fluid-solid interactions, specifically supporting robotics and embodied AI.

MxPage is an open‑source desktop app (Electron + Next.js) that lets users generate e‑commerce product hero images, multi‑page detail graphics and translated assets using LLM‑driven planning and text‑to‑image models. It supports batch processing, platform‑specific prompts, local editing, cost monitoring, and a resumable V4 workflow. The tool runs on Windows/macOS, requires an OpenAI‑compatible API key, stores data locally (SQLite), and is MIT‑licensed.

Quivr V2 is a durable, open-source engine that ingests content streams and makes them instantly searchable and monitorable using hybrid search and a modular plugin architecture.

Juggler is an open‑source visual workbench that lets developers run AI coding agents next to their code. It provides a tree‑based conversation UI, full inspection of every LLM request/response, vendor‑neutral model support, and a JavaScript extension system for custom tools.

A collection of AI agent skills that enable LLMs to search YouTube, extract transcripts, and browse channel data via TranscriptAPI.

An autonomous financial research agent that uses task planning and real-time market data to perform deep financial analysis.

A CLI tool and Python library for interacting with various remote and local Large Language Models, featuring SQLite logging and plugin-based extensibility.

A library that ensures LLMs return reliable, structured JSON by using Pydantic for validation, type safety, and automatic retries.

TheBigPromptLibrary is an open, educational collection of real system prompts, custom instructions, jailbreak prompts, and protection prompts for major AI chatbots, plus articles and tools about extracting and securing them. It explicitly allows only lawful, educational use.

OptiScaler is a middleware tool that allows users to replace or swap upscaling and frame generation technologies in games, enabling the use of alternatives like FSR, XeSS, or DLSS regardless of the game's native implementation.

A professional C++ SDK for real-time collision detection and multi-physics simulation used in robotics, VR, and machine learning.

A minimal implementation of a Vision Language Model (VLM) that allows users to train a 65M parameter multimodal model on a single consumer GPU in just 2 hours.

An intelligent assistant for ComfyUI that automates the generation, debugging, and optimization of image generation workflows using an LLM-powered agent.

An open-source desktop application built with Tauri and React that provides a graphical interface for working with AI agents via the Goose backend.

A browser automation skill for coding agents that enables them to write and execute custom Playwright scripts for testing and complex web flows.

NVIDIA PhysicsNeMo is a PyTorch-based framework for building and scaling AI models for scientific and engineering simulations, supporting diverse data structures like meshes and grids.

An open-source Android document scanner that uses local machine learning for automatic document detection and perspective correction to create private PDFs.

A desktop application that creates persistent AI characters with a continuous identity across text chat, a tabletop RPG engine, and an autonomous OS-level agent.

A headless CLI client for the Agent Client Protocol (ACP) that provides a standardized interface for managing sessions and interacting with multiple AI coding agents.

The official Go SDK for the Model Context Protocol (MCP), providing the tools to build standardized clients and servers for AI model interaction.

A Chinese translation of the book 'Build a Large Language Model (From Scratch)', providing a technical guide and code examples for understanding and implementing LLM architectures.

A local, open-source implementation of the Jev programming model that provides typed decisions and probabilities from an LLM without generating text tokens.

A library of expert-verified, engine-specific skills and architectural patterns designed to help AI coding agents build professional-grade games in Godot and Bevy.

Kody is a Cloudflare Workers‑based platform that provides persistent memory, credential storage, code packages, and job scheduling for AI agents via the Model Context Protocol (MCP). It offers a Remix UI, OAuth‑protected endpoints, and a four‑worker architecture (origin, platform, runtime, jobs) to serve individual or team accounts. The repo is a full‑stack TypeScript monorepo (Nx) with local dev support, extensive docs, and a functional‑source license.

An adaptive input field that uses AI to morph into specialized UI cards (like timers or checklists) based on the user's intent as they type.

An open-source, 3D-printed tendon-driven robotic hand designed for affordable dexterous manipulation research and reinforcement learning.

An agentic framework that generates high-quality educational videos by synthesizing executable Manim code rather than pixels, ensuring mathematical precision and coherence.

MCP Apps is a framework that enables developers to build and render interactive UIs like charts and forms directly within AI chat interfaces using the Model Context Protocol.

Destructive Command Guard (dcg) is a Rust‑based hook that protects developers from accidental destructive commands issued by AI coding agents (Claude, Codex, Gemini, Copilot, etc.). It matches commands against 50+ rule packs (filesystem, Git, cloud, databases, containers, etc.) and blocks them with clear explanations, offering per‑agent profiles, CI scanning, and fast SIMD‑accelerated filtering.

A deterministic codebase scanner that identifies React-specific issues in state, performance, and security, specifically designed to help AI coding agents write better code.

z-skills is a collection of reusable local Agent Skills covering web scraping, video download, document parsing, audio generation, Markdown-to-Word/PDF, grounded Q&A, and more—designed to plug into Codex or Claude Code for Chinese content creation and automation. Each skill includes trigger words, scripts, and evaluation examples so agents can reliably carry out complex workflows.

Source content for Modal's GPU glossary, with dual licensing (CC BY 4.0 for glossary text, MIT for the rest) and a community Chinese translation.

An agentic framework that enables AI to design buildable LEGO models by generating Python scripts that produce LDraw 3D CAD source code.

A high-performance web scraping and crawling engine that converts URLs into clean Markdown or JSON for AI agents and RAG pipelines.

memsearch provides persistent, searchable memory for AI coding agents (Claude Code, Codex, DeepSeek Harness, OpenClaw, OpenCode) by storing conversation summaries in Markdown, indexing them with Milvus, and offering hybrid vector‑BM25 retrieval plus optional remote reranking. It works via zero‑config plugins for users and a CLI/Python API for developers, supporting local ONNX embeddings, cloud Milvus, background note maintenance, and automatic distillation of repeated workflows into reusable agent skills.

Veridrop is an open‑source Python tool that checks whether AI‑API relay/proxy services for Claude, OpenAI or Gemini are genuine, retain advertised capabilities (PDF, tool use, long context, etc.), and follow the official response specifications. It runs a suite of protocol‑specific probes, compares results against official baselines (including a cryptographic signature check for Claude), and produces a weighted score and detailed report. Available as a free web UI (veridrop.app) and as a self‑hosted CLI/FastAPI service, it is licensed under AGPL‑3.0.

Autonomous Company is a multi‑agent framework that runs a continuous LLM‑driven loop (Claude Code or OpenAI Codex) to research, code, deploy and market small products. It defines 14 expert personas (CEO, CTO, designer, etc.), forms temporary AI teams per task, persists work in plain‑text files, and provides a local dashboard for product catalog, logs and human‑in‑the‑loop control. The system works on macOS and Windows WSL, is started via `make start` or a PowerShell script, and ships with 22 example applications.

A CLI that runs AI coding agents in an OS sandbox (bubblewrap/Landlock/seccomp on Linux, sandbox-exec on macOS), scoped to prevent accidents — not to stop attackers. The project directory is read-write; the host is hidden by default, with opt-in flags for credentials, network, GPU, display, etc. Includes a strict `--lockdown` mode and a "phantom credential" proxy that keeps real API keys out of the sandbox. A mature, well-documented Rust project — not a list, tutorial, or demo.

A developer SDK providing APIs to build AI agents that can write code, manage files, and execute terminal commands in local or ephemeral workspaces.

A high-speed decision-making layer for LLMs and agents that uses semantic routing to classify user intent and trigger workflows in milliseconds, avoiding the latency of standard LLM calls.

Apache TVM is an open machine learning compilation framework that optimizes and deploys models into minimum deployable modules using a Python-first approach.

UniFi MCP provides MCP servers and plugins that let AI agents query and control Ubiquiti UniFi network, camera, and access systems via natural language, with safety, lazy tool loading, and remote relay options.