xxxily / hello-ai

Hello-AI > English | **中文文档** • 📚 **Documentation: hello-ai.anzz.top** • 🏠 **Project: github.com/xxxily/hello-ai** Overview The initial intention of this project was to help oneself and others connect to the vast world of AI. As one of the forerunners in the advent of AI, during the peak popularity of ChatGPT, this initiative provided a series of public AI services to resist being exploited and help some people connect to the world of AI. Time flies, and nowadays, AI is everywhere. The public services offered have stopped, but the original intention of this project remains: **Helping oneself and others connect to the AI world!** Therefore, this project is undergoing a **V2.0 Refactoring**. We will no longer directly provide fundamental AI services. Instead, we are shifting our focus to the expansive open-source world. This project is now an **Intelligent, Auto-updating AI Project Directory**. Through an AI agent, we will automatically collect, evaluate, categorize, and track the latest and most popular AI extension projects worldwide (covering Foundation Models, AI Infrastructure, Agent Orchestration, RAG & Data Engineering, Multimodal, etc.). **Core Features:** • 🤖 **AI-Automated Maintenance**: Project collection, tagging, and outdated cleanup are fully driven by cron jobs and LLM scripts, realizing "letting AI help humans connect to AI." • 📦 **Comprehensive Categorization**: Ensuring you won't miss out on excellent emerging forces in the open-source AI community. • 🔄 **Continuous Tracking**: Dynamically tracking the latest trends and purging dead projects promptly. Welcome to explore and discover the perfect AI tools to boost your efficiency! 🏗️ Architecture & Execution Logic This project operates entirely through the collaboration of automated scripts and Large Language Models. Below is a visual flowchart demonstrating the complete data lifecycle—from discovery to frontend rendering: Its core mechanism, data flow, and system architecture details are as follows: • Dynamic Auto-Evolving Discovery Layer • **Topic Mining:** Using a predefined seed list in , the crawler iterates over the GitHub API, prioritizing the "least recently explored" topics to search for new repositories with . • **Knowledge Base Growth:** When unseen topics are detected from newly fetched projects, the system automatically registers them into as 'Level 2' (secondary exploration targets). • **Pending Queue:** All discovered new repositories flow directly into for validation. • Local/Cloud AI Batch Evaluation Engine • **Concurrent Batch Processing:** The core script pops a configured number of items (via ) from the pending pool and creates a combined prompt for the LLM. This batch design drastically reduces API frequency limits and reuses token context. • **Dynamic Category Routing:** The system never "hard-codes" categories. Upon each evaluation, it dynamically reads the valid categories and subcategories from and instructs the AI to route projects accordingly. • **Tagging & Auditing:** The AI automatically extracts tags, generates optimal Chinese descriptions, and assigns the project to the most suitable subcategory. If an item is deemed unworthy or un-categorizable by the AI, it gets discarded into an isolation audit log ( ). • **Objective Trending List:** A daily objective calculation forces the recalculation of the top 30 highest-star, recently updated projects, automatically placing them into the category, overriding AI randomness. • Automated Frontend Rendering & View Decoupling • **Adaptive Routing Presentation:** Built with VitePress, the Navbar and Sidebar have been rewritten from static mappings. Whenever categories are added or removed from , the VitePress compiler dynamically analyzes it and renders the UI perfectly, preventing data-to-UI discrepancies. • **Smart Markdown Folding:** iterates over major categories. When generating the category's markdown page, it groups items under headers according to the assigned by the AI, ensuring an organized layout even with hundreds of projects. • Automation Pipeline • For hands-free, continuous discovery (e.g., avoiding rate-limit drops), you can utilize process daemon scripts like which leverage continuous sleep loops. This achieves a permanent closed-loop operation of: **Discover -> Buffer -> AI Evaluate -> Static Page Build**, endlessly exploring the ocean of open source code. --- 🚀 Local Deployment & Running Guide You are completely welcome to run this entire auto-expanding AI knowledge base framework locally! It is very simple to start: • Environment & Setup A Node.js environment is required (v18.x or above is recommended). • Environment Variables Configuration Copy from the template: Open and adjust the core configurations: • ** ** : Bypasses the strict rate limits applied to anonymous GitHub search API calls. • ** **: Your target LLM API Key (used for analyzing and curating projects). • *💡 Zero-Cost Prompt: If you are using a local LLM setup (e.g. Ollama via llama3), you can simply use .* • ** **: LLM endpoint (e.g. , or local ). • ** **: Standard model identity to use (e.g. ). • ** ** / ** **: Modify limits per pull from GitHub and per LLM prompt bulk execution batch. • ** **: Configure the base idle time interval between consecutive cycles (default: 60s). • Run Automation Pipelines Choose how you want to execute scripts: • **Single Manual Execution**: • **Constant Background Daemon** (Continuous fetch & evaluate): • **Interactive TUI Daemon** (Recommended for manual param selection): • **Incremental Status Check** (Background process to silently check/update GitHub star & health status for evaluated items): • **Re-Evaluate Active Projects** (Moves items back to Queue to catch up with latest sub-category mapping): • **Consume Queue & Exit** (Strictly evaluates queue without pinging GitHub API to avoid rate-limits; auto-exits when empty): 💡 Advanced CLI Flags When running or its variants, you can append the following flags: • : • :…