Trending AI projects

FrontierAgent is an open‑source Python runtime that provides a terminal UI for running autonomous LLM agents. It supports a single‑agent ReAct loop or a coordinated team of parallel sub‑agents, all operating inside a sandboxed filesystem with approval‑gated file changes, interactive instruction queuing, and full session tracing. The framework is model‑agnostic (any OpenAI‑compatible endpoint), includes web‑search and shell tools, and ships an evaluation harness for many research‑oriented benchmarks. Install via `uv`, configure an API key, and launch with `frontier-agent --mode react|agent_team`. Licensed under Apache 2.0.

Hy-Nomas is a Tencent open-source model that generates interactive video worlds from text and camera inputs, runs on a single RTX 4090, supports LoRA fine-tuning, and includes an interactive web demo.

LightRAG is an open‑source Python framework that blends vector embeddings with a knowledge‑graph index for fast, scalable Retrieval‑Augmented Generation. It supports multiple chunking strategies, multimodal document parsing, incremental updates, and a variety of storage back‑ends, and ships with a REST API and Web UI.

Skill Training lets an agent learn professional craft from finished human work without fine-tuning or labels. By corrupting a screenplay into an outline, reconstructing it, and turning the gap into editable Markdown skill cards, it improves writing behaviors that transfer to other models.

A desktop enhancement tool for WorkBuddy and CodeBuddy AI clients that adds multi-account management, automation, and UI customization via CDP injection.

An AI-powered cybersecurity automation platform that integrates 150+ security tools into an MCP framework for autonomous penetration testing.

ASu‑skills is an open‑source plug‑in suite for AI coding assistants that guides Chinese‑language job seekers through a full hiring workflow—from finding real open‑source contributions to polishing a résumé, matching it to a JD, auto‑filling an application form, rehearsing interview questions, and tracking offers. It works on platforms like Claude Code, Codex, Cursor, etc., and is released under the MIT license.

OpenCreator is a local-first AI workspace that combines agentic workflows with visual tools for video, image, and text creation, allowing users to manage complex media production tasks through a unified interface.

Waza is an open‑source collection of eight “skills” (e.g., /think, /check, /hunt) that embed common engineering habits into AI coding assistants like Claude Code and Codex. Each skill provides a clear trigger, a scripted workflow and verification steps, letting the model produce disciplined, project‑aware outputs. Installation is a one‑liner (`npx skills add …`) and the skills can be chained or extended with optional guard‑rail rules.

A command-line tool that enables programmatic creation and editing of CapCut and JianYing projects by directly manipulating local draft files, allowing for seamless integration with LLM agents and automated video production pipelines.

An open-source, self-hostable design editor similar to Canva that allows users to create professional graphics, videos, and PDFs without paywalls or account requirements.

Animate is a Claude Code skill that lets you create short, stylised animated videos entirely from code. You describe a subject, pick a built‑in or custom visual style, and the system builds a deterministic frame‑by‑frame animation, adds voice‑over and music, renders with a headless browser and ffmpeg, and outputs an MP4 with quality checks.

A curated directory of over 5,000 community-built skills for the OpenClaw local AI assistant, helping users discover and install tools to automate tasks and extend agent capabilities.

ReArk is a desktop reverse‑engineering suite for HarmonyOS, Android, iOS, and native binaries that adds AI‑assisted, natural‑language analysis via the ReArk Agent.

A collection of human-reviewed instruction sets (skills) for AI agents to ensure they produce idiomatic, secure, and high-performance Go code.

Zero-Mem is a memory framework for LLM agents that performs all memory operations (construction, organization, retrieval, calibration) without any LLM calls or tokens, using only an encoder and NER for computation. It builds an entity–context graph and temporal hierarchy over original interaction traces, retrieves evidence via query-conditioned routing, and uses a reader LLM only for final answer generation. Reported results show GPT-4o-mini achieving 59.15 F1 on LoCoMo and 72.07 F1 on HotpotQA 56K, with 57.6% lower memory-operation time than the LightMem baseline.

Pi-OptChat is an extension for the Pi AI assistant that adds persistent, long-term memory and subagent capabilities by organizing conversation history into a summary tree.

An AI-powered McDonald's fan certification tool that uses official MCP interfaces to generate personalized meme cards, suggest meals based on real nutritional data, and facilitate coupon claiming.

A versatile, cross-platform audio AI toolkit that enables local execution of speech-to-text, text-to-speech, and speaker analysis models on everything from mobile devices to embedded hardware.

Wan2.2 is a collection of large-scale video generative models that produce high-resolution cinematic videos from text, images, or speech using a Mixture-of-Experts architecture.

A comprehensive open-source handbook providing architectural guidance, engineering practices, and governance strategies for building and scaling enterprise-grade AI agents.

LeRobot is Hugging Face’s open‑source PyTorch library for robot learning, offering a hardware‑agnostic robot API, a unified dataset format (video + Parquet), and a catalog of state‑of‑the‑art imitation‑learning, reinforcement‑learning, VLA, world‑model, and reward models. It supports many real robots, simulation benchmarks, and extensible third‑party plugins, with tools for data collection, training, evaluation, and a web UI for end‑to‑end workflows.

A visual feedback tool that allows developers to annotate UI elements and provide AI coding agents with precise CSS selectors and positions to locate the exact code being referenced.

REA is a Node‑JS platform that connects LLM agents to reverse‑engineering tools (Hopper, Ghidra, IDA, etc.) and static analysis utilities for binaries, JavaScript/Electron, .NET, Android, firmware, web captures and more. It runs a local MCP service, registers a skill with supported agents, and returns concrete evidence (pseudo‑code, assembly, module graphs, screenshots) that the AI can use to explain or re‑implement a target feature. Install via `npx rea‑agents setup`, then ask questions or use the CLI (`rea …`).

OpenChatCut is an open-source, local-first video editor where AI agents edit a real multitrack timeline via natural language or MCP, producing editable projects instead of locked one-shot generations.

A library of 77 AI‑ready, structured prompts and templates that let product managers and their LLM assistants run classic PM frameworks (discovery, road‑mapping, PRDs, stakeholder mapping, AI‑product design, career coaching, etc.) with minimal setup.

A curated, opinionated AI agent learning roadmap (maintained by a Datawhale member) that turns scattered resources into an actionable todo list — emphasizing coding agents, harness engineering, personal agents, modern protocols, and evaluation/safety over legacy role-play frameworks.

PostHog is an open‑source platform that collects product events, errors, logs, and traces, then offers analytics dashboards, session replay, feature flags, experiments, surveys, data pipelines, and AI‑powered insights—all self‑hostable or available via a free cloud tier.

ai-agents-from-zero is a free, open-source, Chinese-language tutorial for learning AI agent development from scratch to enterprise level. It combines systematic chapters (LLM basics, low-code platforms, LangChain/LangGraph frameworks, RAG/Agent projects, fine-tuning, engineering standards), runnable example code, an interview question bank, and two completed real-world projects (an NL2SQL e-commerce query system and a multi-agent research tool). It focuses on the Python ecosystem (Python + LangChain + LangGraph, not Java), targets both beginners and working engineers, and is continuously updated toward 2026.

Respan Gateway is a universal adapter that lets you use any AI SDK to talk to any AI provider, with built-in translation, routing, tools, and observability.

icm-architect is an agent skill that designs any process, idea, or problem into an ICM workspace — a folder structure that acts as agent architecture — or restructures existing folders into one. It works in any agent that reads SKILL.md, replacing orchestration code with interpretable structure.

A skill package for AI agents that automates short‑drama creation: lock model/aspect‑ratio, generate script, dialogue, assets, storyboard, and produce a ready‑to‑feed 7‑section prompt (or MiniMax H3 format). Includes validation scripts, offline storyboard board, and FACS‑based micro‑expression tools for consistent AI video production.

A Claude Code skill for reverse engineering Android apps to extract HTTP APIs and recover original Kotlin class names from obfuscated binaries.

A computer vision toolkit that provides model-agnostic building blocks for data loading, visualization, and dataset management to accelerate the development of vision applications.

A local-first desktop app that uses AI to identify viral moments in long-form video or audio and automatically crops them into vertical short-form clips.

MOVO is a self-hosted enterprise platform that wraps the DeepSeek Harness agent runtime with a user workspace, administrative controls, and document processing capabilities for production-grade AI deployments.

A plugin for Claude Code that lets multiple people share the same live session, seeing each other's prompts, replies, and tool results, while only the host's machine executes the work. It enables real-time pairing and review with host-controlled permissions, works in Desktop and terminal, and can be self-hosted.

An open-source multimodal AI creative workbench that combines an infinite canvas with AI image, video, and audio generation tools to streamline visual iteration.

CUDA Rust is NVIDIA’s open‑source stack that brings safe, idiomatic Rust to NVIDIA GPUs. It provides driver‑level bindings, a runtime (`cuda-core`, `cuda-async`) and two kernel programming models: **cuda‑oxide** (low‑level SIMT, similar to CUDA C++) and **cutile‑rs** (high‑level Tile model for tensor‑centric code). The project is early‑stage research, licensed under Apache‑2.0, and targets developers who want Rust’s safety guarantees while writing GPU kernels.

An open-source, multi-platform AI coding agent that automates development tasks like code reviews, bug fixing, and project navigation across terminals, IDEs, and chat applications.

IndexTTS is an open‑source zero‑shot multilingual text‑to‑speech system that can clone a voice from a single reference audio. The latest version (2.5) supports Chinese, English, Japanese, Spanish, and Arabic, and provides fine‑grained control over emotion, speaking speed, and pronunciation (Pinyin/CMU phonemes/Japanese Kana). It runs fast with half‑precision (BF16/FP16) inference, optional DeepSpeed acceleration, and can be served at scale via a vLLM recipe. The repo includes pretrained checkpoints, a web UI, and a Python API, and is released under Apache‑2.0.

A high-performance Vietnamese text-to-speech system featuring instant voice cloning, real-time streaming, and an OpenAI-compatible API for on-device deployment.

A collection of engineering knowledge and agent-ready skills for automating the deployment, dependency management, and configuration of Computer Vision applications across diverse hardware platforms.

Mobilerun is an open-source framework that enables AI agents to control Android and iOS devices, allowing for automated mobile workflows through natural language commands.

EAC design world is a local video/design studio plugin for DeepSeek Harness that lets AI and humans collaborate on creating videos, websites, slides, documents, and images with real files, previews, and exports. It includes a design library, motion shots, recording, editing, color grading, subtitles, and more, and can run standalone or inside Harness. It requires Node.js 22+, FFmpeg 7+, and optionally Chrome/Edge; it’s AGPL-3.0 licensed.

DSH Voice AI Girlfriend is a plugin that turns DeepSeek Harness into a talking companion: it listens through your microphone, responds with a cloned voice, shows a lip-synced digital human avatar, and can push replies to QQ.

GEO 医生 is an open‑source skill/plugin for AI assistants (Claude Code, WorkBuddy, etc.) that audits a website’s AI‑crawler accessibility, extracts verified product facts, generates realistic user questions, samples answers from major LLMs, analyses recommendation gaps, auto‑writes and publishes SEO‑friendly content, and can retest weekly via GitHub Actions. Install with a single chat command or run a one‑line Node.js audit; the tool is MIT‑licensed and focuses on improving AI‑search (GEO) visibility for Chinese and global products.

An open-source evaluation framework for AI applications and agents that provides a library of ready-to-use graders and tools to generate custom evaluation rubrics.