Trending AI projects

ZCode is a multi-platform AI programming workbench that provides a unified interface for AI agents across desktop, web, and terminal environments, supporting local and remote development workflows.

keypaste is a local, KeePass-compatible password and secret manager that stores logins and environment variables in a KDBX vault and lets you approve or deny each secret request from an AI agent. It also runs commands with injected variables, generates strong passwords, and keeps an audit log of all agent access.

AstrLink is a local privacy gateway for AI agents that detects and redacts sensitive information on your device before sending requests to upstream API providers. It also unifies multiple AI subscriptions and provides tools for model routing and usage tracking.

Harbor is a framework for evaluating and optimizing AI agents and language models by running them in parallelized, containerized environments across various cloud providers.

OpenPencil is an open‑source, cross‑platform design editor that reads native Figma (`.fig`) and Pencil (`.pen`) files, offers AI‑driven design creation (plug‑in any LLM), and provides a powerful CLI plus an MCP server for headless automation. It ships as a tiny Tauri desktop app and a PWA, supports real‑time WebRTC collaboration, and includes a Vue SDK for embedding custom editors.

llmfit is a cross‑platform CLI/TUI (and optional web/API) tool that auto‑detects your CPU, RAM, GPU and accelerator setup, then ranks hundreds of open‑source LLMs (including various quantizations) by how well they fit, their estimated speed, quality and context length. It lets you benchmark models on your own hardware, share results, and consume recommendations via JSON or a REST API, making it easy to pick and deploy a model that actually runs on your machine.

rossark-motion-skill is a skill pack for AI agents that teaches them to animate drawings, characters, and interfaces with lifelike, non-mechanical motion. It provides animation engines, live face components, tools for analyzing reference videos, and a set of principles that fix common agent animation mistakes like teleporting faces, dead stops at poses, and over-synchronized movement.

A developer-centric API and integration testing tool that auto-generates tests and data-mocks from real user traffic using eBPF.

SuperPlane is an open‑source platform that orchestrates AI coding agents, source‑control, CI and other dev‑ops tools into a visible, repeatable “factory” workflow, turning backlog issues into ready‑to‑review pull requests while preserving execution history and cost data.

VLX-Seek is a fine-grained perception VLM that replaces coordinate generation with region-token retrieval for more stable and efficient object localization in embodied AI.

An AI Agent Skill that distills the complete TCM thinking system of Ni Haixia, enabling AI to perform Six Meridian syndrome differentiation and prescription selection based on 3.5M words of distilled knowledge.

LCU is a tool that decouples the official ChatGPT desktop app's computer-use runtime, allowing users to integrate screen and application control into other AI agents and development harnesses.

Capri is an interactive desktop companion that visualizes the activity of your coding AI agents, provides system-level assistance, and offers a character-based chat interface for managing your development workflow.

This project is a community-maintained Simplified Chinese translation of Matt Pocock's agent skills — a collection of small, composable instruction sets that help coding agents handle real engineering tasks such as grilling requirements, test-driven development, code review, and codebase architecture improvement. It keeps all technical names and commands in English to preserve compatibility, while translating the narrative instructions into Chinese to reduce friction for Chinese-speaking developers and models.

Oh Story is a skill pack that equips Claude Code‑style programming agents with a full pipeline for Chinese web‑novel creation – market‑scanning, de‑construction of best‑sellers, outline generation, chapter writing, AI‑flavor linting, and cover image creation – all running on the agent’s existing model without extra GPU.

A local NVR with real-time AI object detection for IP cameras, designed for tight integration with Home Assistant.

Awesome GitHub Copilot is a community-maintained catalog of Copilot customizations — agents, instructions, skills, hooks, workflows, and plugins — with a searchable website, a Learning Hub, and simple plugin installation via the Copilot CLI. It also includes a cookbook for Copilot API recipes and clear third-party safety guidance.

把 AI 客户端的系统提示词换成你自己写的，支持 15 种客户端，Windows 图形界面，免费开源，安全可控。

A privacy-first AI meeting assistant that captures, transcribes, and summarizes meetings locally on your own infrastructure to ensure data sovereignty.

RedKnot is a model-aware execution framework that optimizes long-context LLM serving by using head-aware KV reuse, sparse FFN/MoE execution, and SegPagedAttention to reduce TTFT and compute costs.

An open-source, self-hostable version of Overleaf enhanced with professional features like AI assistants, GitHub sync, and academic reference integrations.

A curated library of 161+ specialized AI assistants (subagents) for Claude Code, organized into 10 categories covering everything from core development and language expertise to infrastructure, security, data/AI, and niche domains. Installable as plugins, manually, or via scripts, with a strict no-promotional-PRs policy and sponsor-supported.

An AI agent skill that transforms articles, reports, and papers into professional image-based PowerPoint presentations by planning outlines, applying visual styles, and assembling generated slides into a .pptx file.

Ultralytics is the official Python library for the YOLO family of computer‑vision models (up to YOLO‑26). It offers a simple CLI (`yolo`) and a Python API to train, evaluate, predict, and export models for detection, segmentation, pose, depth, classification, oriented‑bbox detection, and video tracking. Pre‑trained models are provided for standard datasets (COCO, Cityscapes, ImageNet, etc.), and the package can be installed via pip, conda, Docker, or from source.

A comprehensive guide and technology stack for building, launching, and scaling profitable AI applications while minimizing infrastructure costs.

Agent Relay is an infrastructure toolkit that enables engineering teams to coordinate multiple AI coding agents, share project context, and enforce deterministic workflows across tools and platforms.

Timestral provides inference‑only Python code for Mistral AI’s family of transformer‑based time‑series forecasting models (5M‑209M parameters). It supports multivariate probabilistic forecasts with covariates, returns 21 quantiles per step, and runs on CPU or GPU. Models are hosted on Hugging Face; the repo is Apache‑2.0 licensed, weights under a Modified MIT license.

A recurrent Transformer architecture that feeds the previous token's decoder state back into the current input to enable better length generalization on algorithmic tasks.

Microduck replica is a community project that reverse‑engineers the Pollen Robotics Microduck biped robot, providing editable CAD, reconstructed PCB schematics, BOMs, assembly videos, and a software fork with IMU‑I²C support and a web‑debug UI, enabling hobbyists and researchers to build and experiment with a reinforcement‑learning‑driven walking robot.

FluidVoice is a macOS‑only, open‑source dictation app that provides real‑time transcription using a range of local speech models and an optional on‑device AI layer (Fluid Intelligence) for smart formatting. It supports command‑voice control, per‑app prompts, global hotkeys, and privacy‑first operation, and can be installed via Homebrew or built from source.

An automation assistant for Arknights that uses image recognition and OCR to automate daily tasks, resource farming, and base management.

Rune is a GPU-accelerated native IDE and terminal multiplexer that supports AI-assisted programming and encrypted remote access across devices.

An open-source security infrastructure platform for managing secrets, certificates, and privileged access, featuring a specialized vault for securing AI agent API access.

A middleware tool that lets users replace the temporal upscalers and frame generation technologies in games that already support DLSS, FSR, or XeSS.

Serena is an open‑source server that gives AI coding agents IDE‑like, symbol‑aware capabilities (search, rename, move, safe edit, debugging). It works via the Model Context Protocol, supports >40 languages through LSP or a JetBrains plugin, and is installed with the uv tool. Licensed under MIT (core) and GPL‑3.0‑or‑later.

HEX is a local-first voice dictation app for Mac, Windows, and Linux that transcribes speech on your device and pastes the result into the app you're using. It runs entirely on-device by default, with optional cloud features for rewriting and voice commands.

Open Duet is an open-source, self-hosted workspace for creating and managing AI agents that can work as a team, remember context, run scheduled routines, and operate with approval controls. It connects to external model engines like Claude Code or Codex CLI.

Kapitsa is a fast, locally-run AI text engine for Apple silicon Macs, built for coding agents. Its key innovation, KFlash, checks a tree of a drafter model's guesses in one pass to generate text faster while keeping output identical to plain decoding. It runs Qwen, Nemotron, and Recursive Flint models, and the README benchmarks it as the fastest decoder in 25 of 27 model-and-Mac pairings against nine other engines.

A high-performance programmatic video engine that renders TypeScript and HTML projects into video using a native Rust GPU compositor instead of a browser.

A self‑hosted MCP server that lets AI assistants read/write Xiaohongshu (login, post images/video, search, comment, like, fetch profiles, etc.). Install via binary, Docker, or source, log in once, start the service on port 18060, and connect any MCP‑compatible client (Claude Code, Cursor, VS Code, Open Claw, etc.).

AI-For-Beginners is a 12-week, 24-lesson open-source curriculum by Microsoft that teaches AI fundamentals with practical exercises, quizzes, and labs, using TensorFlow and PyTorch. It covers symbolic AI, neural networks, computer vision, NLP, and ethics, and is available in 50+ languages. It explicitly excludes business AI, classic ML, Azure services, and deep math, directing learners to other resources. The project welcomes contributions and has a Discord community.

Kev is a family of small, specialized decision models that provide calibrated, structured outputs for classification and scoring tasks, designed as an open-source, self-hostable alternative to proprietary decision-model APIs.

An agent orchestration framework that allows you to manage a crew of autonomous coding agents from a single interface, handling parallel tasks, context management, and supervision.

ModLens is a lightweight vision plug‑in for DeepSeek Harness that lets you paste images directly into a chat with a text‑only LLM. It routes the picture to one of several built‑in or reusable vision services (Gemini, OpenAI‑compatible, Anthropic, Antigravity‑CLI, Claude‑CLI, Kimi‑CLI) and returns a structured description for the model to use. Installation is a single command, and the plugin works without extra proxies or config changes.

An interactive, chapter-based tutorial for learning prompt engineering with Claude 3 Haiku, featuring lessons, exercises, a playground, and an optional Google Sheets version.

An open-source orchestrator for agentic workloads that provides durable, lightweight processes (Actors) with integrated SQLite storage and ultra-fast cold starts.

A CLI and Web UI tool that simplifies LLM fine-tuning and post-training, enabling users to train models on consumer hardware using features like layer streaming.

CVEngine is a free, open-source, self-hosted tool that continuously discovers your internet-facing assets, matches them against currently exploited CVEs, verifies findings with safe read-only checks, and gives you a prioritized risk score with evidence. It combines asset discovery, vulnerability scanning, and threat intelligence into one pipeline, and supports optional AI triage with your own keys or local models.