Trending AI projects

A library of 'skills' (context packs) that help AI coding assistants use the Gemini API correctly, with three specific skills for different use cases, installable across multiple coding tools.

A self-contained, offline AI studio that lets you run image generation, text chat, and speech tools locally on your own hardware without needing cloud accounts or complex installations.

World‑Intelligence MCP Server is an open‑source Python platform that gathers free public‑API data from 30+ domains (finance, geopolitics, climate, military, cyber, health, etc.), stores it in SQLite and a Qdrant vector store, and exposes it via the Model Context Protocol for AI agents. It includes a Streamlit dashboard, CLI, PDF/HTML reporting, geofence alerts, and cross‑domain analytics, all built around 139 “tools” that fetch and analyse the data.

MoGe is a model for recovering accurate 3D geometry, including metric depth and point maps, from single open-domain images.

An LLM‑driven agent that scrapes a URL, drafts Twitter/LinkedIn posts, lets a human review them, and then publishes via Arcade or native OAuth. Built on LangGraph, it supports optional Slack, Supabase, GitHub, and YouTube integrations and is configurable through prompt files.

A comprehensive, beginner-friendly collection of tutorials and code demos for learning LLMs, AI image generation, and model fine-tuning using standard APIs and free cloud platforms.

humanize (hmz) is a Python‑based CLI/web tool that orchestrates AI coding agents (Claude, DeepSeek, Kimi, litellm) into reusable flows, offering token‑usage‑aware execution, a macOS desktop app, and easy installation via the `uv` package manager.

AI Homepage is a Chrome extension that uses your Anthropic API key to send a brief summary of your recent browsing (up to 8 pages) to Claude, which then generates a personalized new‑tab page in any style you describe.

MOSS is an open-source bilingual (Chinese-English) conversational language model with 16B parameters that supports multi-turn dialogue and external tool calling via plugins.

A set of custom Python nodes for ComfyUI that streamline Stable Diffusion workflows—seed handling, context routing, LoRA loading, prompt building, and muting/bypassing groups of nodes—allowing cleaner graphs and GPU‑time savings.

A TypeScript framework for building, testing, and deploying MCP servers and interactive AI-powered applications with native React view support.

Conduit is a native Flutter mobile client for iOS and Android that connects to Open WebUI, OpenAI-compatible endpoints, Ollama, Apple's local models, and self-hosted agents like Hermes. It solves the poor mobile experience with self-hosted AI by offering reliable streaming, native rendering, secure credential storage, and offline-capable notes and chats.

An AI prompt management tool providing over 5,000 ready-to-use prompts to help users get expert-grade outputs from AI tools like ChatGPT and Cursor.

Reader is a tool that converts URLs, PDFs, and Office documents into LLM-friendly Markdown and provides web search capabilities that return full page content instead of just snippets.

This is a curated list of tools, extensions, and resources for Gemini CLI, an open-source AI agent that runs in the terminal. It helps developers discover everything from alternative interfaces and forks to agent orchestration tools, MCP servers, and Neovim plugins.

A self-hosted, cross-device platform for managing AI agents and workspaces, featuring a task board and an open API for workflow integration.

CodeNomad is a desktop and browser workspace that sits on top of OpenCode V2, offering project windows, AI‑driven chat sessions, Git worktree management, voice input, remote server mode, and extensible SideCar web tools.

AstrBot plugin that continuously learns from group chat interactions to evolve a bot's personality, slang, social relations, and memory, with a review pipeline and WebUI dashboard. It injects approved learning into LLM requests without replacing AstrBot's core logic.

A self-hosted API relay service for Claude that enables multi-account management, cost-sharing, and private access to Anthropic's models.

NVIDIA Personal AI Router (PAIR) is a cross‑platform desktop/service app that discovers local machines, installs/manage LLM inference engines (Ollama, LM Studio, llama.cpp), and routes OpenAI‑/Anthropic‑/Ollama‑compatible API calls to the best‑suited node. It keeps all traffic on the LAN, supports Windows, Linux, macOS, and both x86‑64/arm64, and provides both GUI and terminal interfaces.

A reference implementation of the Jacobian lens for decoding internal language model activations into vocabulary tokens to analyze internal representations.

AIStudioToAPI is a self‑hosted bridge that turns Google AI Studio’s web UI into OpenAI, Gemini, and Anthropic‑compatible HTTP APIs. It automates a Camoufox browser to log in, supports multiple Google accounts, tool calls, streaming, and offers a web console. Install via npm or Docker, configure with environment variables, and call the familiar `/v1/*` or `/v1beta/*` endpoints to access Gemini’s multimodal models.

OpenClip is a macOS utility that provides a contextual popup menu for selected text, allowing users to quickly run scripts, perform math, search, or trigger AI actions without leaving their current application.

learn-workbuddy is an open-source teaching blueprint that rebuilds the architecture of the WorkBuddy desktop AI assistant in 24 hands-on Python lessons, each adding one mechanism (agent loop, sidecar, memory, audit sandbox, etc.). It solves the engineering gap between CLI agents and production desktop agents — long-lived processes, context limits, safety, and multi-provider support — with fully offline demos and a minimal standard-library reference implementation.

Laya is a desktop AI command centre that aggregates notifications from Slack, Gmail, GitHub, Jira, Notion, Outlook and calendars. It normalises events via n8n, processes them with local or cloud LLMs (Ollama, Claude, GPT, Gemini, etc.) to produce draft “Action Cards”, and lets you approve, edit or reject them. Features include multi‑persona routing, interactive coding workspaces, on‑demand research, hybrid vector‑plus‑keyword search, cost budgeting, analytics, and a chat sidebar. Built with Tauri (Rust), Svelte, Python FastAPI, LiteLLM, SQLite + ChromaDB, and runs locally‑first with optional cloud model support.

A native macOS menu bar application that provides real-time monitoring of Claude AI usage limits, session windows, and API costs across multiple profiles.

GhidrAssistMCP is a Ghidra extension that runs a Model Context Protocol (MCP) server, exposing 100 reverse‑engineering tools (disassembly, decompilation, analysis, etc.) via HTTP/SSE. It lets AI assistants or automation scripts query and control Ghidra—both in the GUI and headless—using a standardized JSON API.

A verification tool for AI coding agents that drives real running applications to confirm if a feature actually works, providing verdicts based on internal app state and network traffic.

Unbrowse is an open‑source TypeScript SDK and CLI that let AI agents call a website’s native HTTP API instead of using a full browser. It discovers or learns “capabilities” (recorded request flows) for sites, runs tasks via a single verified API call, and integrates with hosted MCP services and plugins for major LLMs, offering large speed and cost gains.

Curvine is an open‑source, Rust‑based distributed file system that adds a POSIX‑compatible, multi‑tier cache (memory → SSD → HDD) on top of any cloud object storage. It targets AI workloads—training, inference, and large‑scale agent platforms—by providing fast, low‑latency file access via FUSE, S3/HDFS gateways, SDKs, and a native Kubernetes CSI driver. The system uses a Raft‑replicated master for metadata, worker nodes for cached I/O, and claims up to 5 billion small files per cluster with ~100 µs latency and >100 k QPS. Benchmarks show it outperforms JuiceFS and Alluxio in metadata ops and throughput, while using far less memory and CPU.

A collection of tools, export recipes, and Swift utilities for converting and running PyTorch models on Apple devices via the Core AI framework, including agent plugins for developer assistance.

dshfind is a community site for DeepSeek Harness (DSH) that offers structured lessons on DSH principles, a plugin marketplace that automatically aggregates GitHub repositories tagged `dsh-plugin`, and best-practice guides with community rankings. It solves discovery and education gaps for the DSH ecosystem.

A lightweight ONNX Runtime implementation of Kokoro-TTS that enables fast, multi-language text-to-speech generation on CPU and GPU.

A control-room framework for language-agent societies that allows users to observe, pause, and intervene in AI agent simulations in real-time.

A DIY ESP32‑CAM project that runs TensorFlow Lite on‑device to read analog utility meters (water, gas, electricity). It captures images, extracts the dial, runs a tiny neural net, and publishes the reading via MQTT, InfluxDB, REST API or Home Assistant. Includes a web UI, OTA updates, and 3‑D‑printable housings.

ICM is a Rust‑based, SQLite‑backed memory system for AI coding agents. It stores and recalls project‑specific facts without needing LLM calls, supports keyword or embedding‑based search, integrates with many coding assistants, and achieves benchmark accuracy comparable to larger services.

A tool that parses multi-language codebases into a knowledge graph to enable natural language querying, surgical code editing, and dead code detection.

An open-source framework for making fast, typed semantic decisions by reading model logits directly instead of generating text, reducing latency and parsing overhead for AI agents.

ok-script is a Python-based computer vision automation framework that enables developers to create cross-platform scripts for Windows applications, games, and Android devices.

A Java SDK for the Model Context Protocol that enables Java applications to interact with AI models and tools through a standardized, interoperable interface.

An open-source Python framework to evaluate, test, and monitor ML and LLM-powered systems, helping developers detect data drift and ensure output quality.

A community-maintained collection of configuration files that help website owners block AI crawlers from accessing their content.

Hands-On AI Engineering is a collection of practical AI projects across agents, OCR, audio, multimodal, RAG, and fine-tuning. It's MIT-licensed, welcomes contributions, and emphasizes production-ready, learn-by-doing examples.

A TypeScript SDK that enables AI agents to control Chrome via JavaScript execution and the Chrome DevTools Protocol to automate web tasks.

An LLM-agnostic AI coding agent by JetBrains that automates bug fixes, feature implementation, and PR reviews via the terminal and CI/CD pipelines.

A container-native workflow engine for Kubernetes that orchestrates parallel jobs, specifically optimized for machine learning pipelines and data processing.

A fast and memory-efficient implementation of the attention mechanism for transformer models, reducing GPU memory IO to increase speed and efficiency.

A scalable reinforcement learning algorithm that uses world models to master diverse control tasks with a fixed set of hyperparameters.