Runnable, documented projects worth studying and extending.
OthersAionUiPWA free and open-source AI collaboration platform with a built-in agent engine, supporting unified management of 16+ encoding agents such as Claude Code, Codex, Gemini CLI, and more, offering 24/7 automated tasks, remote access, and cross-platform support
OthersMonkeyCodePWChangting Technology's online AI development platform features a built-in cloud development environment, supporting mainstream large models such as GPT, Claude, GLM, Kimi, and DeepSeek. It drives natural language code development, review, and deployment, and supports enterprise private deployment
OthersLangExtractPWGoogle's open-source Python library uses LLMs to extract structured information from unstructured text, supports precise source tracking and interactive visualization, and has 36K+ stars
AI and AgentsClaude Agent Teams UIPWAI Agent team orchestrates desktop applications: You are the CTO, and the AI Agent is your team. They handle tasks autonomously, communicate with each other, and review code—you just need to look at the board and drink coffee. Supports 75+ large model vendors such as Claude, Codex, and OpenCode.
AI and AgentsSquad - AI Agent TeamsPWAI development team tool powered by GitHub Copilot: generate frontend, backend, test, lead, and other role agents with a single command, persist them in repo files, learn codebases across sessions, and collaborate to accelerate development.
Otherscc-connectPWBridge local AI encoding agents (Claude Code, Cursor, Gemini CLI, Codex) to messaging platforms (Feishu, DingTalk, Slack, Telegram, Discord, LINE, WeCom), allowing you to chat with AI development assistants anytime, anywhere, without the need for a public IP
OthersWeKnoraPWTencent's open-source LLM knowledge platform: converts original documents into queryable RAG Q&A systems, autonomous inference agents, and self-maintained wiki knowledge bases, supports synchronization of Feishu/Notion/Yuque data sources, and integrates 20+ large models
OthershatchetPWAn orchestration engine for backend tasks, AI Agents, and persistent workflows, supporting complex task DAGs, retry strategies, rate limits, and observability. Implemented in Go, suitable for agent scheduling in production environments
OthersWarpPWAn AI-native terminal development environment with built-in Agent capabilities, supporting natural language-driven terminal operations, intelligent command completion, and workflow automation, redefining the way developers interact with terminals
OthersAnubisPWThe open-source HTTP Request Soul Arbiter intelligently analyzes request features to identify and intercept AI crawlers, protecting website content from being harvested by large model training data, supporting custom rules and whitelists
Development and engineeringCode Review Graph - Code Knowledge GraphPWBuilding a native knowledge graph for Claude Code, persistently mapping the codebase structure, allowing AI to read only relevant files—code review reduces tokens by 6.8 times, and daily coding can be reduced by up to 49 times.
AI and AgentsDeepSeek-TUIPWThe terminal's native DeepSeek encoding agent is built on the DeepSeek V4 million-token context and prefix cache. Supports real-time streaming output of the chain of thought, MCP protocol, sub-agent orchestration, workspace rollback, and three interaction modes (Plan/Agent/YOLO). Single binary file, developed in Rust, no Node/Python runtime required.
AI and AgentsiOS Simulator Skill for Claude CodePWClaude Code's iOS emulator skill pack enables AI coding agents to build, run, and interact with iOS applications. Mainly used as an xcodebuild wrapper, it can automatically screenshot and verify the UI, run tests, and debug SwiftUI layouts, saving a large amount of context tokens.
AI and AgentsToken-OptimizerPWLLM Token Optimization Tools: detects ghost tokens, fixes context quality degradation, and maintains semantic integrity during compression. Helps developers save costs and avoid information loss during long-context conversations.
AI and AgentsLTX-2PWThe official inference and LoRA training toolkit for the LTX-2 audio-video generation model from Lightricks. Supports text-to-video and audio-to-video generation, open-source and locally deployable, suitable for short video creators and AI video researchers.
OthersOuroborosPWAgent OS — Uses specifications instead of prompts to drive AI Agents, providing an operating system-level framework for declarative task definition and automated execution
Efficiency toolsDocuSeal is an open-source electronic signature platformPWAn open-source electronic signature platform supporting PDF form construction (12 field types), automatic email sending, multiple storage backends (local/S3/GCS/Azure), mobile optimization, and embedded signature components (React/Vue/Angular). One-click Docker deployment supported, batch sending and SSO.
AI and AgentsTabPFN spreadsheet data foundational modelPWThe basic model for table data, trained on pure synthetic data, can directly classify and predict table data. Compatible with the sklearn interface, supports GPU acceleration, requires no feature engineering or data preprocessing, suitable for rapid modeling of small to medium-sized datasets.
OthersFree LLM API ResourcesPWA carefully curated list of free LLM API resources covers dozens of free inference services such as OpenRouter, Google AI Studio, Mistral, Cerebras, Groq, and more, making it an essential reference for developers to access large model APIs at zero cost.
OthersLadybird BrowserPWA truly independent open-source browser, not based on Chromium or Firefox, built entirely from scratch. The goal is to create a modern browser that is free from ad-tracking and respects privacy.
OthersLocal Deep ResearchPWA local in-depth research engine achieves about 95% accuracy on the SimpleQA benchmark. Supports all local and cloud LLMs (llama.cpp, Ollama, Google, etc.), integrates 10+ search engines (arXiv, PubMed, private documents), runs fully locally with encrypted data.
OthersTabPFNPWThe Foundation Model for tabular data, developed by PriorLabs. No feature engineering or hyperparameter tuning is needed to predict table data directly, with performance comparable to traditional methods like XGBoost, suitable for classification and regression tasks.
OthersAionUiPWA free and open-source AI coding agent collaborative desktop application supporting 20+ CLI tools such as OpenClaw, Claude Code, Codex, Gemini CLI, and 24×hour local collaboration. Provides customizable assistant interfaces, session management, and support for multi-agent parallel development.
AI and Agents9Router - AI-encoded smart routerPWFree AI-encoded smart router, connecting Claude Code, Codex, Cursor, and other coding tools to 40+ providers and 100+ models. RTK compression technology saves 20-40% tokens, and the automatic downgrade strategy ensures perpetual uninterrupted flow.