Trending AI projects

Dormice is a self‑hosted sandbox platform for AI agents. It offers permanent, idempotent sandboxes that freeze to a few megabytes when idle, can be archived to S3, and expose an E2B‑compatible API. Install with a single script, manage via a tiny web console or the `dor` CLI, and use the native TypeScript SDK (or the official `e2b` SDK) to run commands, move files, and control lifecycle. Ideal for teams that want cheap, long‑lived agent sandboxes on their own hardware.

X‑AnyLabeling is a cross‑platform desktop application for AI‑assisted annotation of images, video, text, and 3D point clouds. It bundles dozens of state‑of‑the‑art vision models (YOLO, SAM, Grounding‑DINO, etc.) and supports local or remote inference, offering tools for detection, segmentation, OCR, pose, VQA, and more, with export to common dataset formats.

Cloudflare Computer is a Cloudflare‑hosted virtual filesystem backed by a Durable Object (SQLite) that can be accessed via three pluggable backends: a full Linux container (FUSE mount), a lightweight Bash shell, or a JavaScript worker. It lets AI agents and edge services run tools, store intermediate files, and share state across invocations, with stable core APIs and experimental container/worker backends.

Occamy-1.0 is an open-source, compact agentic language model (35B total, 3B active) designed for long-horizon co-work tasks like tool use, file manipulation, and multi-step workflows. It achieves leading results on co-work benchmarks while keeping inference efficient.

AID Studio is an open-source, self-hosted platform that streamlines the creation of AI motion comics and short dramas by integrating scriptwriting, storyboarding, and multi-model AI generation into a single workflow.

A developer tool for inspecting and testing Model Context Protocol (MCP) servers through web, CLI, and terminal interfaces.

Agent Interview Hub is a free, open-source interview prep library for AI Agent engineers targeting Chinese big tech and top overseas AI companies. It contains 300+ answered questions, company-specific folders (Alibaba, ByteDance, OpenAI, Anthropic, etc.), 6 timed coding challenges, a 16-week study roadmap, and 35 knowledge documents — all readable online via GitHub Pages or locally via Markdown. It also includes a script that automatically collects real interview reports from public sources. MIT licensed.

A curated awesome-list of essays, papers, patterns, and templates on 'harness engineering' — the discipline of building the scaffolding (context, tools, verification, memory, sandboxes) around AI agents. Maintained by ai-boost and licensed CC0.

MonkeyCode is an open-source AI office workbench (product name: 百智办公 MonkeyAI) where teams describe tasks in natural language and AI delivers finished office documents, code, or designs. It includes team resource sharing, usage billing, permission auditing, and usage statistics. It can be self-hosted with a one-command install (min 1 CPU + 1 GB RAM), supports bringing your own models and MCP-connected tools, and offers desktop clients for Windows, macOS, and Linux. Free tier covers up to 10 seats; paid tier adds cloud PC and support. Licensed under AGPL-3.0.

TrailSnap is an open‑source, self‑hosted AI‑enhanced travel photo album. It automatically extracts ticket data, recognizes faces/scenes, indexes images for smart search, and lets a large‑language model write travel diaries. The app can be run via desktop installers or Docker (server + PostgreSQL + AI micro‑service + frontend).

A reward-guided post-training framework for diffusion models that uses on-policy self-distillation to create explicit intermediate targets, improving image quality and training efficiency.

An open-source desktop app that downloads video and audio from 1000+ sites, transcribes them locally, and uses AI to summarize and search the content.

Figwright is a free, two-way MCP server + Figma plugin that lets AI coding agents read Figma designs as code and edit the canvas directly — without a paid Dev Mode seat.

A collection of tutorials, code examples, and deployment guides for building edge-native AI applications using Liquid Foundation Models and the LEAP SDK.

An agentic video creation framework that automates the entire pipeline from narrative planning and scriptwriting to storyboarding and final video generation.

Internet Court is a single open skill that bundles ~35 protocols (vendored 92 official skills across 34 owners) so AI agents can discover, negotiate, contract, pay, execute, and — uniquely — *adjudicate* deals with humans out of the loop. Agents agree in advance on a dispute judge, written into the contract. One root SKILL.md routes to sub-skills on demand. Built by a consortium of ~30 companies, MIT licensed for first-party code, installable on 6 agent harnesses.

Matrix-Game 4.0 is an open‑source hybrid system that combines a traditional game engine with a causal neural‑rendering model, delivering real‑time, high‑fidelity visuals for three ready‑made scenes. It runs on high‑end NVIDIA GPUs (A100, RTX 5090) and is released under Apache‑2.0.

ColliePWA is a self-hosted mobile web interface that allows developers to interact with and manage terminal-based AI agents from their phone, providing a native-like experience for terminal multiplexers.

SAM 3 (Segment Anything with Concepts) is Meta’s 848 M‑parameter foundation model that lets you prompt an image or video with free‑form text (or visual exemplars) and receive masks, boxes, and scores for *all* matching objects. It combines a DETR‑style detector and a SAM 2‑style tracker, introduces a presence token for fine‑grained prompt discrimination, and is trained on >4 M auto‑annotated concepts. The repo provides installation steps, example notebooks, and a new SA‑CO benchmark (270 K concepts) for evaluation.

A community-written practical handbook for WorkBuddy, organized around real tasks — from first use to building reusable AI-agent teams — with case studies, contribution guidelines, and a VitePress-based website.

Mesh2Motion is an open-source web application that allows users to import 3D models, assign skeletons, and apply animations for export, serving as a flexible, community-driven alternative to tools like Mixamo.

A comprehensive framework that adds structured professional workflows and Chinese-language support to 26 popular AI coding tools, ensuring AI agents follow rigorous development methodologies.

MuScriptor is an open-source, multi-instrument music transcription model that converts audio recordings into MIDI files and sheet music.

An enterprise AI digital employee platform that provides orchestration, deployment, and runtime capabilities for AI agents with support for on-premise deployment.

halogen‑flash‑server is a containerised inference server tuned exclusively for the Qwen 3.8‑Flash‑Next model on an AMD Strix Halo GPU. By using custom kernels for that exact hardware it delivers sub‑2‑second first‑token latency and up to 94 tokens / s decode speed, even with 30‑96 k‑token contexts—ideal for low‑latency AI agents. The repo ships a ready‑to‑run Docker/Podman image, OpenAI‑compatible HTTP API, extensive docs, and a Discord community for setup and benchmarking.

A GPU-accelerated terminal and SSH workspace that integrates AI CLI tools like Claude Code and Codex into a unified native desktop environment.

A screen parsing tool that converts GUI screenshots into structured elements, enabling vision-based AI agents to accurately identify and interact with interface components.

A curated collection of copy-paste prompt templates that make NotebookLM (or Kael.im) generate presentation slides in specific visual styles — from newspaper editorial and manga to premium Apple-style mockups. Each template is a detailed design brief (colors, fonts, layout rules) sourced from social media creators. Includes a bonus citation-checker tool and is licensed CC BY 4.0.

Maki is a terminal‑based AI coding assistant that minimizes LLM token usage through tools like a tree‑sitter index, sandboxed code execution, and sub‑agent task delegation. It supports dozens of model providers, offers a fast Rust TUI, Lua‑based plugins, and safety features such as permission‑aware command parsing and SSRF protection.

A repository that extracts, version‑tracks, and publishes every system‑prompt string used by Anthropic’s Claude Code code‑assistant, providing raw prompt files, token counts, and a changelog for over 500 fragments.

A real-time shared workspace that connects multiple AI agents and apps, eliminating the need to manually copy-paste context between different AI tools.

An open‑source, ESP32‑based quadruped robot that’s 3‑D printable, Wi‑Fi‑controlled via a JSON API, and includes tools for animation (Sesame Studio), simulation, and a Python companion app.

A visual status dashboard for Claude Code that displays session metrics, token usage, and media playback controls directly above the terminal prompt.

A command-line tool that adds an OCR text layer to scanned PDF files, making them searchable and copy-pasteable while maintaining original image resolution.

Univer is an open‑source Office SDK that lets developers embed spreadsheet, document and presentation editors into their own apps. It offers a canvas‑based renderer, formula engine, plugin‑first architecture and a unified Facade API that works in browsers and Node.js, enabling both UI‑driven and headless (AI‑agent) use‑cases.

Busabase is an open‑source, local‑first database + workspace that lets AI agents (Claude Code, Codex, Cursor, Gemini, etc.) and people share structured records, docs, skills, and tiny apps. Every agent write is stored as a Change Request with diff, author, and history, enabling durable, auditable multi‑agent workflows. It runs via a simple `npx busabase server` command (or Docker/desktop), supports templates, and offers an API (MCP/OpenAPI) for any custom agent.

Rescript is an open-source, transcript-based media editor: drop in video/audio, get a local transcript with word timestamps and speaker labels, then edit by deleting text — the media follows. All processing happens on-device. Runs in the browser or as a desktop app (macOS/Windows/Linux). It supports filler/silence removal, timeline split/trim, multiple export formats, and NLE/DAW timeline export. Licensed under PolyForm Noncommercial (prior versions remain MIT).

A reactive Python notebook that ensures consistency and reproducibility by automatically running dependent cells and storing notebooks as pure Python files.

An open-source humanoid robot project providing the CAD, simulation models, and software needed to build and customize a 1.2m bipedal robot.

TraceMemo is a desktop app that reads your local WeChat database, builds searchable indexes (including OCR for images), and optionally adds LLM‑powered natural‑language search. It provides group‑chat summaries, speaker statistics, member‑exit monitoring, text‑to‑speech, and automation pipelines, all while keeping data on‑device. Available for Windows x64 and macOS (Apple Silicon & Intel).

An open-source AI orchestration framework for building production-ready RAG systems, autonomous agents, and LLM applications in Python.

An autonomous AI agent that enables LLMs to fix GitHub issues, find security vulnerabilities, and perform custom coding tasks.

A plug‑in for the open‑source AstrBot framework that lets the bot automatically start a conversation after a period of silence. It supports private and group chats, configurable timing, context‑aware prompts, TTS, message splitting, persistence, and includes its own WebUI for monitoring and configuration.

safelabs‑eval is an open‑source Python framework that red‑teams AI agents. It sends 300 curated adversarial prompts (30 per OWASP‑inspired ASI01‑ASI10 categories) to a target agent—via HTTP or a Python callable—and scores the replies with pure‑Python pattern detectors. The tool is framework‑agnostic (supports LangChain, CrewAI, AutoGen, LlamaIndex, OpenAI Agents SDK, Google ADK, Semantic Kernel via optional adapters), produces human‑readable or JSON reports, and requires no extra LLM calls. It helps developers discover prompt‑injection, data leakage, tool‑use safety, hallucination, and other agent‑specific risks before deployment.

dsh‑purge is a community plugin for DeepSeek Harness (web and desktop) that adds a UI dock for applying, restoring, or uninstalling custom jailbreak prompts and host policies. It lets security researchers test the separation between model refusals and Harness’s sandboxing, works only with Harness 0.2, and is intended for local, authorised red‑team evaluation.

OpenKnowledge is a free, open‑source markdown editor that feels like Google Docs/Notion but runs locally. It offers true WYSIWYG editing, desktop (macOS/Windows/Linux) and web UI, and side‑by‑side AI assistance from Claude, Codex, Pi and other LLMs via an MCP/CLI bridge. Built‑in git/GitHub sync lets teams share notes, and the tool supports AI‑driven wikis, second‑brain knowledge graphs, and embeddable HTML specs.

An open-source, local-first writing workspace that integrates AI agents with local sources and semantic search to help users draft and revise documents while preserving their personal voice.

Overlay Studio 是一个免费工作台，通过20种预调好的动效卡和AI语义编排，让口播视频的动效制作变成“挑卡对时间”而非“AI现场发挥”，并支持可视化编辑和透明层导出，直接导入剪映。