Trending AI projects

Spark-X2.6-4B is a 4-billion-parameter on-device language model that improves agentic, coding, and long-context capabilities, offering configurable thinking levels to balance quality and speed. It runs on a variety of inference engines and is designed for deploying capable AI agents on edge devices.

An open-source AIOps and alert management platform that centralizes, deduplicates, and uses AI to correlate and summarize alerts from multiple observability tools.

A collection of on-device inference frameworks for Apple platforms providing speech-to-text, text-to-speech, and speaker diarization using Core ML.

A genuine open‑source project that provides 22 “Claude Code” skill packs turning Anthropic’s Claude‑Code assistant into a cybersecurity co‑pilot. Each skill is a markdown file with triggers, methodology, templates and optional Python scripts, covering everything from OSINT and vulnerability scanning to AI‑LLM security and purple‑team emulation. Install by copying the `skills/` folders into `~/.claude/skills/` (or a project‑local `.claude/skills/`). Claude then auto‑activates the appropriate skill during conversation, runs helper scripts, and returns ready‑to‑use reports, rules, or commands. MIT‑licensed, with safety gates for offensive actions.

Google's Agent Development Kit for Go is an open-source, code-first framework for building, evaluating, and deploying modular AI agents. It applies software engineering practices—such as versioning and testability—to agent development, and is model-agnostic while optimized for Gemini.

WeMM-Embedding is a family of multimodal models that convert text, images, videos, and visual documents into unified numerical representations for cross-modal search and analysis.

A workflow orchestration framework for building resilient, dynamic data pipelines in Python with built-in scheduling, retries, and monitoring.

A feature-rich web and desktop interface for managing AI image generation and editing, supporting multiple providers, batch processing, and local-first history management.

OpenOSINT is an MIT‑licensed Python package that lets security analysts run 21 OSINT utilities (email, username, breach, WHOIS, IP, DNS, etc.) via a natural‑language LLM interface. It can be used as a REPL, CLI, web chat UI, or as an MCP server for programmatic access. By default it operates in a passive mode (only read‑only API calls); active, potentially noisy tools must be explicitly enabled. The toolset is extensible, supports parallel execution, auto‑generates Markdown/PDF reports, and offers a pay‑per‑use cloud version on Apify. Install with `pip install openosint` and run `openosint` (or `openosint web`).

A Python library for portfolio optimization and risk management that extends the scikit-learn API to provide a unified framework for building, validating, and tuning financial models.

ExLlamaV3 is a Python library for fast, low‑memory inference of large language models on consumer GPUs. It provides QTIP‑based EXL3 quantization (2‑8 bit), parallel and CPU‑offloaded execution, dynamic batching, and multimodal support, with integrations for Hugging Face, Transformers, and an OpenAI‑compatible API via TabbyAPI.

A curated ecosystem of plugins, tools, and infrastructure for DeepSeek Harness, enabling users to extend AI agent capabilities through community-developed extensions.

Laya‑MLX is a Python library that runs ConvAI’s Laya LLMs locally on Apple‑Silicon via the MLX runtime. It provides ultra‑low‑latency *typed‑decision* inference (choice, score, yes/no) without PyTorch or cloud services, includes a router for multilingual/English models, optional compilation, confidence gating, and CLI tools such as a snake‑game demo.

MindOS is a local-first knowledge base that gives you and your AI agents a shared, persistent memory. It stores your knowledge as plain Markdown files, connects agents via MCP and Skills, and helps you turn reviewed conversations into reusable SOPs so agents start from a better baseline.

A native PlayStation 5 application that runs llama.cpp on the console's GPU, providing an OpenAI-compatible API and a dashboard for local LLM inference.

A Windows taskbar widget that monitors usage limits and reset times for various AI coding assistants, including Claude Code, GitHub Copilot, and Cursor, by reading local session data.

DarkMoon is an open‑source, CLI‑driven platform that uses a local or cloud LLM to autonomously run full penetration‑testing campaigns. It discovers the target, dispatches ~50 specialist agents (web, AD, Kubernetes, cloud, AI endpoints, etc.), executes real offensive tools through a logged MCP layer, and records the exact command and raw output as proof. Results are emitted as JSON/Markdown with severity, CVSS and MITRE tags. Integrations exist for GitHub, GitLab, Jenkins, IDEs and monitoring tools. A paid Pro tier adds a web dashboard and automated remediation PRs. The tool runs in Docker, supports privacy‑preserving tokenisation, and was benchmarked to find 57 bugs in OWASP Juice Shop in ~28 min.

Nativ is a SwiftUI macOS app that bundles the MLX‑based `mlx‑vlm` server to run LLM, vision, and speech models locally on Apple‑silicon Macs. It offers chat, image generation, voice dictation, model management, performance dashboards, OpenAI‑compatible APIs, and an extension system, all wrapped in a native UI.

A self-hosted control plane for operating AI agents, allowing users to dispatch tasks, track spend, and coordinate multiple agent runtimes from a single dashboard.

dsh‑worktable is a Cordis plugin for DeepSeek Harness that adds a sidebar drawer, dockable split workspaces, and a built‑in Control‑room dashboard to monitor all projects in real time. Projects can be bound to chat conversations, opened as multi‑pane windows (file explorer, terminal, browser, custom agent‑generated content), and managed via a persistent UI stored in browser localStorage. The plugin works on Windows Web and Desktop (experimental macOS), has no analytics, and is MIT‑licensed.

dsh‑vision‑router is a Node‑based plug‑in for DeepSeek Harness that adds a “👁 Vision” toggle, letting a text‑only LLM call pixel‑level vision tools (crop, OCR, grounding, etc.) via configurable cloud or local vision models, with a free OVHcloud fallback and a local‑only privacy mode.

Lanhu MCP Server is an open‑source bridge that implements the Model Context Protocol, letting AI coding assistants read Lan‑Hu design and requirement documents, extract visual and textual data with a vision‑capable LLM, and share results via a team knowledge board. It runs via HTTP or stdio, offers Docker and one‑click local install, and provides caching, incremental updates, and Feishu notifications.

An AI product design agent skill that gives coding agents design judgment to create high-quality UI, presentations, and graphics instead of generic layouts.

A VitePress-based open-source repository that compiles course notes, code, and online notebooks for Python, LangChain, LangGraph, and AI Coding (Vibe Coding) video courses, with a live GitHub Pages site for easy review and practice.

XPolicyLab is a unified framework that decouples robot policies from evaluation environments, allowing researchers to test and deploy a wide range of models across different simulators and real-world robot setups using a standardized interface.

An AI orchestrator for trading that integrates AI agents, market data, and brokerage accounts into a local workspace for research and execution.

A macOS application for discovering, organizing, and editing AI coding agent skills and agents across multiple tools like Claude Code, Cursor, and Windsurf.

CLI Agent Orchestrator (CAO) is a Python tool that lets you run several AI‑coding command‑line interfaces (like Claude Code, Copilot CLI, etc.) as coordinated agents. It starts a local server, launches each provider CLI in its own tmux pane, and offers a web UI, API, and plugin system for building parallel or sequential code‑generation pipelines.

A resumable long-running task workflow and skill platform for AI coding that provides structured methodologies and scientific evaluation for agentic software development.

sparkDash is a Docker‑based, real‑time web dashboard (React + Express) for monitoring NVIDIA DGX Spark clusters and any GPU‑enabled Linux/Windows host. It streams hardware stats (GPU/CPU, memory, network, energy) and LLM inference metrics (token‑per‑second, cache, latency) via SSH collectors, supports built‑in benchmarks, optional ComfyUI and Hermes Agent integration, Prometheus export, and a mobile‑friendly PWA.

Pixal3D is a high-fidelity 3D asset generator that creates detailed geometry and PBR textures from a single image or multiple views using pixel-aligned back-projection.

Celesto provides secure, isolated, and persistent virtual machine environments for AI agents to safely execute code, browse the web, and interact with desktop applications.

A Chrome extension that turns YouTube videos into learning resources by providing AI-generated summaries, bilingual transcripts, and timestamped notes using the user's own API keys.

An open-source AI agent that automates Android and iOS devices using natural language to perform tasks and scrape data from apps.

A secure, personal AI assistant that stores data locally and uses WASM sandboxing to run tools safely and privately.

DiffSynth‑Studio is an open‑source diffusion‑model engine that unifies inference, low‑VRAM management, quantisation, and LoRA‑style training for image, video, and audio generative models. It supports many state‑of‑the‑art models, offers VRAM‑saving tricks (disk‑/CPU‑offload, NF4/INT8/EntroPack quantisation), and ships companion UI tools (ComfyUI, WebUI). Ideal for researchers and developers who need a single framework to run or fine‑tune large diffusion models on consumer GPUs.

Jev Browser is a Node.js tool that lets an LLM (via TypeSafe’s Jev model) control a headless Chromium browser to accomplish a user‑specified web task. It offers a CLI, a library API, and an MCP server, supports secure password‑fill and cookie injection, provides step‑by‑step confidence traces, and is designed for cheap, observable web automation.

Diri is a native Rust‑based GUI for macOS (beta Linux) that lets you run many AI coding agents—Claude Code, Codex, Gemini, etc.—in parallel, each in its own Git worktree. It provides a single‑pane UI with live status, agent‑to‑agent spawning, session persistence, and local‑first operation, enabling coordinated “swarm” coding workflows.

Kortix is an open-source AI operating system that lets organizations manage, version, and run autonomous agents in isolated cloud sandboxes, requiring human approval for all actions.

LangGPT is a structured, reusable prompt‑design framework that treats prompts like a tiny programming language. It defines a clear template (role, profile, goal, skills, rules, workflow, etc.), supports variables, commands, and conditional logic, and works across major LLMs. The project provides example libraries, ready‑made ChatGPT/Claude assistants, a Claude Code skill, and an ecosystem of related tools (PromptVer, PromptShow, Minstrel) plus model‑specific prompt collections.

iCode is a terminal-based AI agent orchestrator that provides a transparent, privacy-focused interface for managing LLM workflows, file editing, and agent performance monitoring.

Flash Linear Attention is a library that provides hardware-efficient implementations of linear attention, sparse attention, and state space models, designed for training and inference of modern sequence models. It includes a wide range of architectures and fused building blocks verified across NVIDIA, AMD, and Intel hardware.

An open-source singing voice synthesis editor for the UTAU community that provides a modern, cross-platform interface for creating vocal covers and songs.

A pi extension that brings autonomous sub-agents and workflow orchestration to the platform, allowing users to spawn specialized agents that run in parallel or via scripted JavaScript workflows.

ainovel‑cli is a command‑line/TUI tool that orchestrates multiple LLM agents (Architect, Writer, Editor) via a deterministic engine to automatically write long‑form novels (500+ chapters). It features rolling‑window planning, hierarchical summarisation, checkpoint/recovery, multi‑model support, import/export, and configurable rules, all runnable headlessly or in Docker.

Bonsai‑demo is a local demo for PrismML’s Bonsai 2 LLMs (2‑27 B parameters). It provides scripts to install dependencies, download ternary‑quantised GGUF/MLX checkpoints, and launch a chat server (llama.cpp or MLX) with vision, OpenAI‑style tool calling, long‑context (up to 262 K tokens), and an optional Open WebUI UI. The repo is a genuine AI software project.

VPP2 is a world-action model that integrates video prediction with robot policy learning, enabling agents to perform better by anticipating future visual states before executing actions.

A TypeScript library for running the Laya decision model locally via ONNX Runtime to perform high-speed, calibrated classification and scoring tasks without text generation.