Hexabot, Claude Code and Railway chatbot setup

This tutorial builds a customer support AI agent with Hexabot, starting with a local development project and ending with a chat widget on KGPTalkie.com. The setup requires Node.js and the Hexabot CLI. The speaker starts the dashboard on localhost port 3000, creates an MCP token, and connects Claude Code as a coding assistant to create and test the workflow.

The chatbot uses an OpenRouter API key stored in Hexabot credentials. The speaker selects an NVIDIA model through OpenRouter, which handles inference requests. The walkthrough uses a free model endpoint and Railway account credits, with usage subject to those services’ availability and limits.

Deployment uses a GitHub repository, Railway and a Postgres database. The tutorial covers environment variables and deployment errors, then recreates the workflow on the hosted instance with a new MCP connection and credentials.

For RAG, the speaker asks Claude Code to scrape 10 website articles and enables lexical retrieval. After retrieval returns no content, ingesting articles one at a time resolves the demonstrated problem. The final steps integrate the widget into a Next.js website, fix an unclickable chat bubble, configure allowed origins, and remove localhost entries before the live deployment.