Runnable, documented projects worth studying and extending.
AI and AgentsDeepEP - DeepSeek MoE expert parallel communication libraryPWDeepSeek's open-source MoE expert parallel high-performance communication library provides high-throughput, low-latency all-to-all GPU cores, supports FP8 low-precision computation, and is suitable for large model training and inference acceleration
Othersml-internPWHuggingFace is an open-source ML Agent that can automatically read papers, train models, and publish them to HuggingFace Hub. Based on the Agent architecture, supports end-to-end machine learning research workflow automation.
Othersdaily_stock_analysisPWLLM-driven intelligent analysis system for A-shares, Hong Kong stocks, and US stocks, supporting multi-source market data + real-time news + AI decision dashboard + multi-channel push notifications, can run at zero cost on scheduled runs via GitHub Actions.
Othersadk-samplesPWGoogle's official Agent Development Kit example collection offers ready-to-use agent samples in Python, TypeScript, Go, and Java, covering academic research, customer service, data engineering, financial advisory, and other scenarios.
OthersManifestPWThe intelligent model routing engine automatically selects the best model for personal AI agents, saving up to 70% in costs. Supports intelligent switching between multiple models, automatically routing to suitable LLMs based on task complexity.
OthersPolymarket CLIPWPolymarket's Rust command-line tool for predicting markets supports viewing market data, placing trades, and tracking betting portfolios, providing an efficient terminal interface for quantitative trading in the prediction market.
OthersGoogle ADK SamplesPWThe official Google Agent Development Kit (ADK) example collection includes reference implementations of various AI Agent application patterns, covering scenarios such as multi-agent collaboration, tool invocation, and RAG. 8.9k+ Stars。
OthersHands-on Learning Large Models (Dive into LLMs)PWThe "Hands-on Large Model Learning" series of programming practice tutorials starts from scratch to help you understand and implement the core technologies of large language models. 34k+ Stars, the top choice for learning Chinese LLMs.
OthersMicrosoft PresidioPWMicrosoft's open-source sensitive data detection and desensitization framework supports PII identification, masking, and anonymization in text, images, and structured data, with customizable NLP pipelines. 7.7k+ Stars。
OthersRTK - Agent Token compressed proxyPWA CLI proxy tool written in Rust can reduce the token consumption of LLM encoding agents by 60-90%. Zero dependency on single binary files, automatically compresses redundant output from development commands like git diff and npm install, supports 12+ mainstream encoding agents.
OthersMicrosandboxPW🧱 A secure locally programmable sandbox designed for AI Agents. Based on microVM technology, it provides an isolated execution environment with millisecond-level startup, supports multilingual code execution, and can be controlled via API programming, suitable for production-level agent deployment.
OthersHuggingFace ML-InternPWHuggingFace's open-source ML Engineer Agent can automatically read papers, train models, and publish them to Hub. Let AI handle the entire process from paper to model for you.
OthersManifest - Smart Model RouterPWAn open-source AI model intelligent router that automatically sends simple requests to cheaper models and complex requests to strong models, saving up to 70% on API costs. Supports one-click Docker deployment.
OthersPDF InspectorPWFirecrawl's high-performance Rust PDF parsing library intelligently detects scans versus text PDFs, completes classification and text extraction within 200ms, supports Python/Node.js binding.
OthersT3 CodePWPing.gg Lightweight Web GUI for unified management of Codex and Claude Code encoding agents, 10K+ Star, desktop cross-platform support.
OthersOpen Generative AIPWAn open-source, uncensored AI image/video generation studio integrating 200+ models (Flux, Midjourney, Kling, Sora, etc.), with MIT protocol self-hosting.
OthersMemoriPWAgent-native memory infrastructure: a persistent memory layer independent of LLMs, converting agent execution and conversations into structured persistent states, supporting Python and TypeScript SDKs
OthersmemsearchPWA cross-platform AI coding agent semantic memory system supporting multi-platform memory sharing such as Claude Code, OpenClaw, Codex, and a lightweight memory solution based on Markdown
AI and AgentsGeneral Commercial Agreement (UCP)PWStandardized commercial interaction protocol specifications and documentation between AI Agents
OthersStitch Skills (Google Labs)PWGoogle's official Agent Skills library, compatible with Stitch MCP Server, supporting mainstream encoding agents such as Gemini CLI / Claude Code / Cursor
AI and AgentsHermes Agent - NousResearch super-intelligent agentPWNousResearch's open-source AI agent framework has 110,000+ stars, supports autonomous growth and tool usage, and is currently one of the hottest open-source agent projects.
AI and AgentsGenericAgent - Self-evolving agentPWAutomatically grow skill trees from 3.3K rows of seed code, achieving complete system control and reducing token consumption by 6 times. A new paradigm for self-evolving AI Agents.
OthersProject LyraPWNVIDIA's open-source generative 3D world model can generate interactive 3D scenes from text or images, suitable for gaming, simulation, and robot training.
OthersSGLangPWHigh-performance LLM and multimodal model inference service framework, supporting structured generation, RadixAttention prefix caching, and various quantization methods, with throughput leading vLLMs.