How coder-teddy/ai-support-agent works
Project Type and Purpose
• This is a single‐service, full-stack “AI Support Agent” CLI/daemon built in Node.js.
• Its purpose is to scrape website content, convert it into vector embeddings via OpenAI, store/search in a vector DB (ChromaDB), and answer user queries semantically.
Core Components
a. Web Scraper Module
• Implements Axios + Cheerio (in index.js)
• Fetches HTML from target URLs
b. Text Processor & Chunker
• Splits raw HTML/text into manageable chunks (textChunks() logic)
c. Embedding Service
• Calls OpenAI API (text-embedding-3-small) to convert chunks → vectors
d. Vector Database
• ChromaDB running in Docker (defined in docker-compose.yml)
e. Query Handler
• Accepts user input (CLI or API hook)
• Embeds query, performs semantic search over ChromaDB
f. Response Generator
• Ranks top N results, synthesizes answer in natural language
g. External Services
• Target website(s) being scraped
• OpenAI API endpoint
Relationships & Data Flow
1. Start: User invokes node index.js → triggers Scraper
2. Scraper fetches HTML from External Website(s)
3. Processor extracts text, splits into chunks
4. Embedding Service sends each chunk to OpenAI API → receives embeddings
5. Embeddings are written to ChromaDB (Docker container)
6. On user query: Query Handler embeds query via OpenAI → gets query vector
7. Vector DB performs similarity search → returns closest documents
8. Response Generator collates and returns best answer
Architectural Patterns & Principles
• Single-process pipeline with modular stages (Scraper → Embedder → DB → Search → Response)
• Separation of concerns: extraction, transformation, storage, retrieval
• Infrastructure as code: Docker Compose for vector DB
• External API integration: OpenAI for embeddings
Additional Notes
• If expanding into REST or chat widgets, overlay an “API Gateway” fronting the Query Handler.