Trending AI projects

A comprehensive, offline-first personal web application that provides a persistent, multi-modal environment for long-term emotional interaction with AI models.

A Python‑based, markdown‑driven framework that records threat‑hunting investigations using a four‑step LOCK template, makes them searchable, and adds AI‑ready context. It provides a CLI, a validation skill (GATES), and an optional local MCP server so LLM assistants can read past hunts, generate hypotheses, and decide whether findings should become production detections.

TorchGeo is a PyTorch library that makes working with satellite and aerial imagery easy. It offers ready‑made geospatial datasets, smart samplers for patch extraction, multi‑spectral pre‑trained models, and tight integration with PyTorch Lightning, letting researchers train and evaluate remote‑sensing models with just a few lines of code.

A cross-platform AI chat client that unifies local on-device LLM inference via llama.cpp with cloud API access for a flexible, privacy-focused chat experience.

ATLAS is a research project that uses large language models to automatically translate textbook mathematics into formal Lean 4 libraries, providing reusable, machine‑checkable mathematics for theorem proving and AI research.

An open-source, bring-your-own-key AI chat client for iOS, Android, and web that lets users interact with multiple model providers directly without a subscription.

A configuration management layer for the Pi coding agent that allows users to package, version, and automatically generate custom agent harnesses called Genomes.

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

A visual automation bot for the QQ Classic Farm game that uses OpenCV template matching to automate farming, seed management, and social interactions.

A highly scalable, general-purpose workflow orchestrator from Netflix that provides workflow-as-a-service for managing massive data and machine learning pipelines.

A portable mixed-precision linear algebra library providing high-performance SIMD kernels for multiple CPU architectures to prevent numerical overflow and reduce binary size.

An AI agent skill that directs the creation of cinematic, high-performance websites using a disciplined pipeline of art direction, asset generation, and strict anti-slop linting.

An opinionated workflow layer for the Hermes Agent that provides a curated set of skills and agents to autonomously design, build, ship, and monitor software projects.

Dirac is an open‑source AI coding agent that works in VS Code, the terminal, or any ACP‑compatible editor. It lets you give a high‑level goal (e.g., “refactor the logging system”) and then autonomously plans, inspects ASTs, makes precise hash‑anchored edits, runs sub‑agents, and verifies results—all while you can steer it, set low‑verbosity mode, and control permissions. It supports dozens of LLM providers, custom tools, and parallel batch operations, and benchmarks show it completes multi‑file refactoring tasks at the lowest token cost among comparable open‑source agents.

A Home Assistant integration that connects Frigate's AI-powered object detection and NVR capabilities to a smart home hub for automated security monitoring.

FastClaw is a Go‑based, single‑binary runtime that lets you create, manage, and run multiple AI agents with their own personalities, memory, and skills. It supports any OpenAI‑compatible LLM, sandboxed tool execution, channel bots (Telegram, Discord, Slack, etc.), cron‑style scheduling, and a full web UI + CLI + OpenAI‑compatible API. Data is stored in SQLite (or Postgres) and configuration is env‑only. The project is source‑available under a community license.

PixelDiT and PixelDiT2 are NVIDIA/University‑of‑Rochester research projects that implement diffusion‑transformer models operating directly in pixel space for image generation. They provide class‑to‑image and text‑to‑image models (up to 1.3 B parameters, 1024 px), pretrained checkpoints, a pip‑installable library, and integration with ComfyUI.

NVIDIA cuML is an open‑source library that re‑implements scikit‑learn‑style machine‑learning algorithms to run on NVIDIA GPUs, offering up to 50× speed‑ups. It provides a native `cuml` Python API, an automatic acceleration layer (`cuml.accel`), and multi‑GPU support via `cuml.dask`.

Pre-built Python bindings for OpenCV that allow users to install the computer vision library via pip without needing to manual compilation.

A Python library for multi-agent reinforcement learning research that provides a standardized API and a diverse set of environments, similar to a multi-agent version of Gymnasium.

A Vim and Neovim plugin that integrates GitHub Copilot to provide AI-powered coding suggestions based on natural language prompts.

Tardigrade is a TypeScript, Effect‑TS‑based framework for building composable, event‑sourced AI agents. It lets you define immutable state with atoms, derive views and actions with effect atoms, and run agents on Cloudflare Workers or a self‑hosted Celld fleet.

A multi-core AI agent framework that simulates the GLaDOS personality, featuring persistent memory, vision, emotional state tracking, and autonomous decision-making.

go‑openai is an unofficial Go client library that wraps the OpenAI API (text, chat, embeddings, images, audio, fine‑tuning, etc.). It supports the new Responses API for tool‑calling and multi‑turn state, provides model constants, Azure compatibility, streaming, and typed error handling, and includes runnable examples.

A production-ready MCP server that connects AI assistants to Nextcloud, enabling LLMs to manage notes, calendars, contacts, and files via natural language.

An infrastructure layer that provides a single, shared persistent memory database with vector search for all your AI tools, eliminating fragmented data across different AI clients.

A batteries-included agent harness for TypeScript/JavaScript that provides a ready-to-run agent with built-in planning, filesystem tools, and sub-agent delegation.

HydroGym is a comprehensive reinforcement learning platform that provides a standardized interface for training agents on complex fluid dynamics and flow control simulations across multiple CFD solver backends.

A serverless, agentic CMS self-hosted on Cloudflare that allows users to publish content to a website, RSS, and JSON feeds.

A set of AI skill instructions that transforms ordinary photos into minimalist, poetic hand-drawn illustrations with heavy white space and tiny handwritten captions.

The repository provides open‑source hardware designs and assembly instructions for building high‑performance, locally‑run AI workstations (from a 2‑GPU desktop to an 8‑GPU rack). It pairs the hardware with open‑source inference software (Grid, vLLM, Ollama, llama.cpp) so users can run large language models on‑prem, avoiding cloud costs and preserving data privacy.

A comprehensive setup for Claude Code that integrates Obsidian and Graphify to provide persistent long-term memory and token-efficient codebase awareness.

VLA-JEPA is a Vision-Language-Action model that enhances robotic control by integrating a latent world model to better predict and execute actions.

A terminal and JSON CLI tool, written in Rust, that indexes the local conversation history of dozens of different AI coding agents into a single searchable database. It provides high-speed full-text search, optional local semantic search, and a 'robot mode' API so AI agents themselves can query past sessions. It emphasizes privacy (everything local), robustness (atomic index updates, quarantine of corrupt data), and performance (sub-60ms searches).

A model optimization framework that uses compression algorithms like quantization and pruning to make AI models faster, smaller, and cheaper to run.

Rust bindings for the PyTorch C++ API (libtorch), enabling the use of PyTorch's tensor operations and deep learning tools in Rust.

A real-time dense feed-forward scene reconstruction system for 3D mapping and localization, featuring optional open-set 3D object detection.

Edit Mind is an open‑source, Docker‑based app that indexes local video files with AI (transcription, face/object detection, embeddings) and lets you search the content via natural‑language queries using ChromaDB. It runs fully locally, supports optional GPU acceleration, and offers a web UI plus a commercial desktop wrapper.

senpi is a TypeScript monorepo that rebrands pi-mono's coding agent as a saner, extension-first runtime with built-in intent gating, compaction, permission controls, and more. It serves as the runtime for the Dori AI assistant and offers a lightweight alternative to OMO as a single CLI binary.

An OpenTelemetry-native evaluation and observability framework that uses LLM judges to trace agent failures and optimize the cost-quality tradeoff in AI applications.

MCP Scanner is a Python security tool that scans Model Context Protocol (MCP) servers, their tools, prompts, resources, and associated code for malicious behavior. It combines Cisco AI Defense API, YARA rules, and LLM‑based analysis, supports CLI and REST API modes, integrates vulnerability scanning, VirusTotal checks, and supply‑chain analysis for Python and npm packages, and offers extensive configuration and reporting options.

wallstreet-explainer is a Skill for AI tools that automatically creates polished financial-style videos. It handles footage sourcing, person extraction, data visualization, camera moves, voiceover, and rendering, following a strict minimalist style. It includes scripts, style guides, and a full sample video. Code is MIT-licensed, but third-party content in the sample is not.

A high-performance, real-time voice activity detection system for conversational AI agents that provides lower latency and higher precision than industry standards like Silero VAD.

WhisperSubTranslate is a desktop app that locally transcribes video audio using whisper.cpp and then translates the subtitles either offline with Tencent’s Hy‑MT2 models or via online services like DeepL or OpenAI. It runs on Windows (portable ZIP) and can be built for Linux, supports GPU acceleration, and keeps all data on the user’s machine.

KohakuTerrarium is a Python framework for building AI agents. It provides a reusable runtime with controllers, tools, sub-agents, session persistence, and composition tools, so you can create custom agents quickly without rebuilding the underlying machinery.

An enhanced YouTube client that integrates Google Gemini for AI video summarization alongside ad-blocking and advanced download tools.

AutoGluon is an automated machine learning (AutoML) library that finds the best combination of models for tabular, time series, and multimodal data to achieve high predictive performance with minimal code.

YT Short Clipper v2 is a Windows‑only desktop app that automatically turns a long YouTube video into vertical (9:16) short clips. It downloads the video and its original subtitles, feeds the transcript to an LLM to pick highlight segments, cuts and re‑frames each segment, burns word‑by‑word captions, and can upload the results to Repliz. Built with React + TypeScript, Tauri (Rust), and a Python side‑car using yt‑dlp, FFmpeg, and MediaPipe, it supports any OpenAI‑compatible model or an in‑app wallet‑based AI service. The project is MIT‑licensed, beta‑stage, and Windows‑only.