Trending AI projects

SurfSense is an open‑source, Apache‑2.0‑licensed desktop app that provides an offline AI agent for researching, creating, editing, and converting documents (PDF, Word, Excel, PPT, images, audio, etc.). Users choose any OpenAI‑compatible or local LLM, keep all data on‑device, and can extend functionality with web‑search plugins.

Native app port of SillyTavern (LLM chat frontend) with a Rust/Tauri backend — same UI and data compatibility, no Node.js required, on desktop and mobile.

The repository is a genuine software project that provides Docker images, deployment guides, and raw benchmark data for heterogeneous GPU inference of large mixture‑of‑experts LLMs (e.g., Qwen‑3.8‑Flash‑Next, DeepSeek‑V4.1‑Flash). It shows how to split prefill and decode across different GPU types (RTX 3080, RTX 6000D, DGX Spark, AI Max 395) and reports untuned performance numbers up to 16 k tok/s prefilling and ~442 tok/s decoding.

A tool that lets AI agents control real iPhones and Android devices via screen mirroring, adb, or cloud hosting to automate mobile app interactions.

sprite-gen is a Python CLI (and Codex/Claude skill) that turns a single character illustration into a game‑ready transparent sprite atlas with a manifest, using AI image/video models, optional curation, and many post‑processing utilities.

A Python MCP server that gives AI assistants and other tools instant, offline‑first access to Brazil’s fiscal data (CNPJ, NF‑e, SPED, eSocial, Simples Nacional, 2026 tax reform). No API keys needed; optional premium source adds real‑time official data.

E2B is an open‑source platform that offers cloud‑hosted, isolated sandboxes for safely running AI‑generated code. It ships Python and JavaScript SDKs to start a sandbox, execute shell commands, run code snippets, or control a desktop environment. The runtime can be self‑hosted on AWS, GCP, Azure, or any Linux machine via Terraform, making it a practical building block for LLM‑driven agents, AI coding assistants, and automation bots.

A dynamic training framework built on LLaMA-Factory that provides unified implementations of data selection, mixture, and reweighting to optimize LLM training performance.

A tool for running untrusted workloads in fast, local microVMs with hardware-level isolation, specifically designed for AI agents and automated code execution.

Kero is a cross-platform terminal for coding agents. It runs Claude Code, Codex, dev servers, and test watchers in persistent sessions grouped by project, so quitting the window doesn't halt them and you always know which agent needs your attention.

A privacy-focused desktop writing application for web-novel authors that combines a professional text editor with local-first AI assistance and project management tools.

A visual editor for Matplotlib and AI-generated scientific figures that allows users to tweak layouts and elements visually while keeping the original Python scripts untouched.

An AI co-author toolkit that automates the research lifecycle, from generating reviewer-defensible research ideas to creating posters, videos, and blogs from finished papers.

A completion discipline for AI agents that uses runnable shell gates and evidence-backed ledgers to ensure substantial software work is fully completed and verified.

Agent Teams AI is a free desktop app that lets you assemble, run, and monitor teams of AI agents across many LLM providers. Agents collaborate on a Kanban board, can review code, and you retain control via per‑action approvals and budget analytics. Installable on macOS, Windows, and Linux with no build required.

An MCP server that integrates IDA Pro with LLMs, enabling AI agents to perform interactive reverse engineering, binary analysis, and patching.

A security layer that intercepts and blocks destructive commands or unauthorized secret access from AI coding agents before they execute.

A React-based framework that enables AI coding agents to generate, edit, and manage presentation decks through natural language and code-based workflows.

A concise domain-specific language and compiler built on TVM for developing high-performance GPU/CPU/NPU kernels like GEMM and FlashAttention.

A highly customizable status line formatter for the Claude Code CLI that displays real-time session metrics like token usage, git status, and API costs directly in the terminal.

AssetFurnace is a local, web‑based studio that converts a text prompt or 2‑D image into a rigged, animated, game‑ready 3‑D GLB asset. It supports multiple generation back‑ends (Pixal3D, Hunyuan‑3D, TRELLIS.2, Stable Fast 3D), runs on Apple Silicon, NVIDIA, or AMD GPUs, and includes a Pixel‑Match step that projects the original image’s pixels onto the mesh for crisp details. The tool is installed with a single script, works entirely offline, and records licence provenance for each output.

AutoRAG Agent is a Node‑based, locally‑run RAG system that turns your computer into a searchable knowledge base. It indexes files (PDF, Office, HWP, images, etc.) with BM25 and a built‑in embedding model, supports many external chat sources, and answers natural‑language questions with numbered citations. Available as a CLI, an interactive terminal UI, a desktop Electron app, and a TypeScript library, it emphasizes privacy (no data leaves the machine) and ease of use (no vector‑DB setup).

Horizon is an AI-powered news aggregator that collects, filters, and summarizes content from various sources into personalized, context-rich daily briefings.

A tiny npm proxy that forwards legacy Magic‑MCP calls to the new 21st MCP server, preserving old configs while giving access to the updated AI component generation tools.

An intelligent data querying system based on LLMs and RAG that enables conversational data analysis (ChatBI) by converting natural language into SQL and visual charts.

An open-source TypeScript framework and product stack for building autonomous AI agents that can run locally or in the cloud across web, mobile, and OS distributions.

Harp is a macOS terminal fork of Warp that replaces the tab bar with a sidebar showing the status of Claude Code and Codex AI coding agents, adds optional usage‑limit rings and tmux integration, and otherwise works exactly like the upstream Warp terminal.

Open Intelligent UI is an open‑source demo that recreates ChatGPT’s “Intelligent UI” – LLM‑generated, interactive answers (maps, photos, itineraries, dashboards). Built on the OpenUI generative‑UI standard, it provides a TypeScript component library, 59 built‑in functions, stateful inputs, and can run against OpenUI Cloud, any OpenAI‑compatible API, or a local Ollama model.

Marker is a document conversion tool that turns PDFs, images, and Office files into clean Markdown, JSON, or HTML. It preserves complex layouts like tables, equations, and forms, and can optionally use an LLM to further boost accuracy.

exo is an open‑source platform that lets multiple Macs (or Linux boxes) form a single AI cluster. It auto‑discovers devices, shards models across them using MLX, and speeds up communication with RDMA over Thunderbolt 5. The system exposes OpenAI‑compatible APIs and a web dashboard, supports custom HuggingFace models, and runs on macOS (with a Homebrew cask or DMG) or from source on macOS/Linux.

A fully open-source DIY humanoid robot project providing the hardware designs, RL training workflows, and ROS2 deployment framework needed to build a walking and running robot.

Mark LVI is a cross‑platform desktop AI assistant with voice, avatar, vision and OS‑control features. It can use Google Gemini Live or any local/hosted LLM (Ollama, OpenAI, etc.) together with Whisper for speech‑to‑text and Piper/ElevenLabs for text‑to‑speech. The UI lets you swap engines, add plugins, and manage memory, while safety guards ensure the assistant only reports verified actions.

Pocket TTS is a 100 M‑parameter, CPU‑only text‑to‑speech library that streams audio with ~200 ms latency, supports multiple languages and voice cloning, and offers a Python API, CLI, and local HTTP server. It can be installed via pip (or uv) and runs faster than real‑time on modest hardware, with community‑trained models and several alternative runtimes for browsers and embedded devices.

OpenReality converts phone videos into persistent 3‑D scenes and exposes them to LLM assistants via 41 MCP tools for measurement, navigation, robot‑training export, and more, with a self‑hostable backend and BSD‑2‑Clause code (non‑commercial model license).

A curated list of 132 example projects, tutorials, and recipes for building LLM-powered applications — covering text agents, voice assistants, RAG systems, MCP integrations, memory-enhanced agents, and fine-tuning — organized by category with links to each project's code.

An 'awesome list' of Claude 5.5-made videos and animations, with model-specific browsing, research findings on style/color/production, guidance for making your own, and links to reusable open-source projects — all backed by statistical analyses and documented maintenance.

An AI coding agent for the terminal that can read and edit code, run shell commands, and integrate with IDEs via the Agent Client Protocol.

A GitHub repo offering plug‑in style “skills” that extend Claude Code or any skill‑aware AI agent with ready‑made ModSmith mods, multi‑agent workflow orchestration, task relaying, prompt‑ambition boosting, and artifact explanation with spaced‑repetition learning.

NVIDIA NemoClaw is an open‑source stack that runs AI agents (OpenClaw, Hermes, LangChain Deep Agents) inside NVIDIA OpenShell sandboxes, providing a CLI, managed inference, network‑policy enforcement, sandbox hardening, and state persistence for safer on‑premise agent execution.

An agent skill that provides automated guidance on fine-tuning design details like typography, animations, and spacing to make user interfaces feel more polished.

A structured investment research framework for Claude Code and Codex that applies value investing methodologies to generate professional, decision-grade financial reports.

LiveKit Agents is a Python SDK for building real‑time, multimodal voice agents (LLM‑driven assistants) that run on LiveKit servers. It provides unified adapters for STT/LLM/TTS, job scheduling, WebRTC/telephony integration, turn detection, MCP tool support, and a built‑in test framework.

The ecom‑details‑image skill is a CLI‑based tool that creates complete e‑commerce visual packages (main images, detail‑page graphics, banners, etc.) using AI image models via the apiz service or Codex’s built‑in imagegen. It can output a structured visual brief and prompt, or directly generate images, applying a “Campaign Style Lock” to keep style consistent across the set. Installation requires the apiz CLI; usage is driven by prompts to an AI assistant or by running `scripts/generate_image.py` with options for prompt source, size, model, and reference images. The repo contains no API keys and warns about cost, async generation, and compliance with content policies.

HTML Anything is a local‑first, agent‑driven HTML editor that turns plain‑text input into publish‑ready HTML using installed AI coding‑agent CLIs. It offers 75 ready‑made templates for various content types, live streaming previews, one‑click export to Chinese social platforms or as .html/.png files, and works without any API keys.

Graphicle is a Rust‑based, local tool that builds a searchable code‑graph and stores project docs, decisions and past coding‑assistant sessions. It serves this data over an MCP server so agents like Claude Code, Cursor or VS Code can request cited excerpts within a token budget, all without telemetry or cloud services. Benchmarks claim it’s dramatically faster and lighter than the Python‑based Graphify, and it can be installed with a single command and managed via a concise CLI.

PocketPal AI is a free, open‑source mobile app that runs quantized LLMs and neural TTS entirely on‑device (iOS & Android). It lets you chat, hear the assistant speak, and use built‑in tools (calculator, date‑time, HTML rendering) without any cloud connection. Models are downloaded once from Hugging Face and executed via llama.cpp (CPU/GPU/NPU). Users can create custom “Pals” or download community assistants from PalsHub. Developers can extend the app by adding new Talents, TTS engines, or locales; the codebase is a standard React‑Native project with native bridges to llama.cpp and ONNX Runtime. MIT‑licensed, privacy‑first, and fully offline.

SmartSub is an open-source desktop tool that packages video-to-subtitle transcription, translation, proofreading, TTS dubbing with voice cloning, and subtitle burning into a single pipeline, supporting both local offline processing and AI-agent automation via MCP/CLI.

A real-time object detection and target locking system for Delta Force using YOLOv14 to identify game characters while filtering out environmental false positives.