Trending AI projects

A collection of agent skills designed to automate UI engineering tasks like accessibility audits, layout reviews, component stress-testing, and design-to-code implementation.

GEOInflow is a self-hosted analytics platform that tracks AI crawler activity and website traffic, helping teams monitor how AI search engines and training bots interact with their content.

A template project that enables AI coding agents to clone existing websites into modern, clean Next.js and Tailwind CSS codebases.

dsh-ops is a Windows‑only DSH plugin that bundles bash 5.3, PowerShell 7, and fast Rust file‑search tools so AI agents can run shell commands with lower token cost and faster execution. It offers configurable components, background‑job management, strict JSON schemas, and is distributed via npm under MIT/Apache‑2.0.

Toonflow is an open‑source, multi‑platform app for creating AI‑generated short films. It provides a unified canvas for scripts, characters, scenes and generated video, supports local or Docker deployment, a plugin marketplace, and can hook any OpenAI‑compatible image, video or LLM service.

Vibe Research is a local-first investment research workstation that integrates global stock market data with AI-assisted analysis to help users perform deep-dive company research, backtesting, and portfolio management.

Diy-LLM is a comprehensive, hands-on educational course that teaches users how to build, train, and optimize Large Language Models from scratch using PyTorch. It provides a structured path covering everything from basic architecture to advanced distributed training and evaluation.

A tool to download and run open-source AI models locally on your own hardware, providing privacy, a simple CLI, a REST API, and official Python/JavaScript SDKs. It acts as a local AI server that powers coding assistants, chat interfaces, agents, and thousands of community integrations.

AirLLM is a Python library that streams LLM layers from disk so only one layer (plus optional adapters) resides on the GPU at a time, enabling inference of models up to 671 B parameters on as little as 4‑12 GB VRAM and fine‑tuning of 125 B models on ~6 GB. It offers optional 4‑/8‑bit weight‑only compression for up to 3× speed‑up and works with most Hugging Face models via a familiar `AutoModel` API.

Heretic is an automatic tool that removes safety‑alignment “censorship” from transformer‑based language models using directional ablation and Optuna‑based hyper‑parameter optimisation, preserving model quality while drastically reducing refusals.

gstack is an open-source framework that turns AI agents into a structured engineering team by providing specialized slash commands for product planning, security auditing, QA, and release management.

Open‑source job‑search platform that crawls 3.3 M+ live IT postings, deduplicates them, offers faceted search, and adds AI‑powered CV matching, fit analysis, and an in‑app assistant. Built in Go, PostgreSQL + pgvector, Meilisearch, SvelteKit front‑end, and optional LLM integration.

A collection of 15 motion design films created by Claude Opus 5.5 writing code for visuals and audio, providing the prompts and source code to replicate the process.

Headroom is a context compression layer for AI agents and LLM applications. It compresses prompts, tool outputs, logs, RAG chunks, and conversation history locally before they reach the model, cutting token usage by 30–90% while preserving accuracy and allowing reversible retrieval of original content.

An LLM-powered trading agent that chats with you over Slack/Discord/Telegram/web, pulls market data, runs backtests, and executes trades across 18 brokers — with a "grounding gate" that verifies every reported number against real tool outputs to prevent AI hallucination.

Bend is an open‑source, high‑performance language that integrates dependent types and a fast proof checker. It lets you write formal `LAWS.bend` specifications and requires AI‑generated code to supply a `PROOF.bend` that the compiler verifies, effectively blocking bugs. The compiler targets C, CUDA, Metal and JavaScript, runs programs on GPUs with automatic parallelism, and claims C‑level speed and sub‑second proof checking. The project includes a CLI, formatter/LSP, benchmarks, and academic papers, but is still early‑stage with many language and tooling limitations.

A set of agent skills that enable AI agents to research top-grossing mobile apps and build high-fidelity, simulator-verified Expo and React Native screens.

LEANN is an open‑source, ultra‑light vector database that stores almost no embeddings. It builds a graph of document chunks, recomputes embeddings on‑demand, and prunes the graph to use ~97 % less storage than traditional vector DBs while keeping search quality. It runs locally on a laptop, supports many data sources (files, emails, chat logs, live streams), offers HNSW/DiskANN back‑ends, integrates via MCP, and works with any OpenAI‑compatible LLM or embedding provider.

A suite of eleven Claude skills and Python tools that simulate a business board, consumer panel, and CFO to validate and plan a startup before launch.

MOSS‑Image is an open‑source 5 B diffusion model for text‑to‑image generation with precise layout control via JSON prompts. It offers two checkpoints (Base and fine‑tuned SFT), a custom Diffusers pipeline, Gradio UI, ComfyUI plugin, and a prompt‑enhancer skill for agents.

An open-source AI office suite that allows users and AI agents to create and edit native Word, Excel, and PowerPoint files with byte-preserving compatibility.

A modern terminal multiplexer and window manager built in Go that features deep integration for monitoring and coordinating AI coding agents across local and remote sessions.

A tool that uses AI and the McDonald's MCP protocol to help users find and compare meal combinations that fit within a specific budget.

A full-featured coding agent CLI with IDE integration, subagents, debugging, and many built-in tools, built on Rust and TypeScript.

LayerFS is a Rust‑based, SQLite‑backed storage engine that gives AI agents cheap, immutable snapshots of their filesystem workspaces. Using content‑addressed storage, content‑defined chunking, and copy‑on‑write, it deduplicates data across many parallel agent runs. A CLI and FUSE container runtime let you create a Store, start a workspace, commit changes as immutable layers, and fork new branches without copying files. The v0.1.6 preview demonstrates strong storage efficiency but lacks durability guarantees and cross‑host sync. It’s a research‑grade foundation for building agents that need fast, versioned filesystem state.

A collection of agent-agnostic skills for AI short-drama production that automates the pipeline from novel adaptation to shoot-ready storyboards and character bibles.

A tool that converts technical books and documents into structured AI agent skills, enabling low-token, hallucination-free referencing of complex sources during work.

An open-source inference framework optimized for fast serving of LLMs, vision-language, and diffusion models, particularly for agentic workloads and RL rollouts.

BIMCraft is an open-source, Rust-based BIM tool that recreates Revit's workflow, supports IFC and .rvt files, and is agent-drivable via MCP. It's in early development but already offers a wide range of modeling, drawing, and analysis features.

Project Maya is a local AI engine that enables you to run massive, high-performance language models like GLM-5.3-Flash and DeepSeek-V4.1-Flash-1017 on your own hardware by intelligently managing memory across your GPU, RAM, and SSD.

CorridorKey 2.0 is an AI video matting tool that extracts foreground subjects from video with optional hint inputs, offering multiple inference modes and HDR support.

WeKnora is Tencent’s open‑source, LLM‑powered knowledge‑management framework. It lets teams ingest documents from many sources, search them with hybrid RAG, run multi‑step AI agents with sandboxed tools (including browser automation), and maintain a collaborative wiki with a knowledge graph. The system is model‑agnostic (29 vendors), supports multimodal embeddings, RBAC, and can be deployed via Docker Compose, Helm/Kubernetes, or a single‑binary Lite mode. MIT‑licensed.

Omnirush‑gui is a cross‑platform desktop application that provides a chat‑based coding assistant powered by frontier LLMs (default GPT 6 Astra). It lets you pick models/effort levels, add reusable skills, and run Git operations (clone, branch, PR, etc.) directly from the UI. The app self‑updates, stores credentials securely in the OS keyring, and is distributed as DMG/AppImage/.deb/.rpm packages under an MIT license.

Erupt is an annotation‑driven Java admin framework that auto‑generates full CRUD UIs from JPA entities. It includes built‑in AI modules (`erupt‑ai` and `erupt‑ai‑claw`) offering multi‑provider LLM integration, role‑based tool security, chat history, and natural‑language control of the server. Fast startup, rich UI components, RBAC, multi‑DB support, and optional commercial extensions make it a turnkey solution for building AI‑enhanced admin panels on Spring Boot.

A free, open-source, and offline speech-to-text application that transcribes voice to text directly into any active text field on your computer.

Prompt Optimizer is a multi‑platform (web, desktop, Chrome extension, Docker) tool that iteratively refines LLM and text‑to‑image prompts, evaluates improvements, and saves reusable prompt assets. It supports major model providers, offers advanced testing (variables, multi‑turn, function calling), and can be self‑hosted with optional password protection.

A curated collection of 100+ open‑source LLM agents, skills, and RAG templates that run with a single API key or locally. Organized into categories (starter agents, advanced multi‑agent teams, voice bots, generative UI, RAG tutorials, etc.), each entry is a ready‑to‑run codebase under Apache‑2.0, with security‑checked CI and weekly updates.

A Python CLI (and WorkBuddy skill) that uses McDonald’s MCP API to generate real‑time group‑meal quotes, book themed party slots, and suggest optimal dates, all with standard‑library code and built‑in rate‑limit handling.

Squadra coordinates multiple AI coding agents (Claude Code, Codex, Cursor) on one project. It maintains a shared registry of tasks, reviews, and decisions, ensuring independent review and proof for each change, with a local dashboard to track progress.

Windows‑MCP is an open‑source MCP server that exposes Windows UI‑automation tools (mouse, keyboard, window management, screenshots, browser DOM mode, etc.) to any LLM‑based AI agent. Installable via PyPI, it runs on Windows 10/11 with Python 3.14+, supports local `stdio` or network transports (SSE/HTTP), and includes optional auth, TLS, IP allow‑lists, and OAuth for secure remote access.

An AI agent-led search engine that aggregates and scores real-time data from social platforms and prediction markets to provide a synthesized brief of the last 30 days of activity.

A tool that uses AI agents to convert static 2D illustrations into animated, rig-ready avatars for streaming, featuring a local editor for fine-tuning and real-time face tracking.

OpenPencil is an open-source, AI-native vector design tool where you describe UIs in natural language and watch them generate on an infinite canvas. It uses concurrent AI agent teams, adapts to different models' capabilities, stores designs as Git-friendly JSON (`.op` files), exports to multiple code frameworks, and includes a built-in MCP server for terminal-based AI control. It runs as a web app or native desktop app (macOS/Windows/Linux) from a single Rust core, with a CLI, collaboration features, style guides, and presentation tools.

A service and agent skill designed to strip multi-vendor AI provenance marks and watermarks from text and a wide variety of file formats for privacy and hygiene.

Co‑op Anything is an MIT‑licensed framework that uses LLM‑driven coding agents to turn any PC single‑player game into a multiplayer‑ready co‑op mod. It provides playbooks, agent skills, and an MCP server so agents can research, reverse‑engineer, test with bots, and package a one‑click Steam‑relay install. Existing co‑op mods are revived or new ones built from scratch for a wide range of engines, with several games (Deus Ex, Dungeon Keeper, S.T.A.L.K.E.R., Fallout NV, Ocarina of Time) already supported.

A DSH plug‑in that shows DeepSeek (and other LLM providers) balance, daily spend, per‑turn cost, and lets you fully customise the little whale bubble UI, with drag‑and‑drop positioning, audio alerts, peak‑off‑peak pricing, and multi‑vendor API support.

A remote access tool for DeepSeek Harness that enables secure, end-to-end encrypted access to your local AI workspaces and tools from mobile, web, or other desktop devices without requiring public network exposure.

A cybersecurity skills router that provides AI agents with structured methodologies and tool routing for reverse engineering, penetration testing, and CTF challenges.