The Avenex website implemented a free AI assistant using Cloudflare's infrastructure to answer visitor questions based on site content. The system utilizes Workers for request processing and Workers AI for model execution, avoiding the need for dedicated servers.
The knowledge base consists of 14 Markdown documents totaling approximately 12,900 characters. Currently, the entire base is transmitted with each query as context. The system is designed to transition to a RAG (Retrieval-Augmented Generation) approach using Vectorize as the database grows beyond 14,000 characters.
The development team tested various models from the Workers AI catalog, which includes 70 options as of October 13, 2026. They selected the GLM-4.7-Flash model for its performance in generating concise answers while adhering to the provided source materials.
Cloudflare's free tier provides the necessary resources, including limits on request volume and processing time. The assistant is programmed to refuse off-topic requests and acknowledge when information is missing from the provided documentation.
