5bv57zcm44-max/noxus-ai-open-whatsapp

This project is a self-hostable platform that turns a WhatsApp account into an AI-powered customer service channel. Instead of just providing an HTTP endpoint to send messages, NOXUS AI Open WhatsApp also ships an AI assistant that replies on the user's behalf, an admin dashboard for running the whole thing, and a way to plug in the operator's own back-end APIs so the assistant can answer with up-to-date information. The AI part is built around what the README calls grounding: the assistant only answers from the knowledge file the operator writes and from the live API responses fetched at reply time. If neither source covers the question, the assistant says the team will follow up rather than making something up. Replies are generated by Google's Gemini model, and the assistant automatically mirrors the customer's language, with English and Arabic explicitly tested and full right-to-left support in the dashboard. The headline feature is what the README calls API slash commands. The operator registers each back-end API in the dashboard with a slug, a name, a description, a URL, headers, and an HTTP method. They then write their reply data, which is essentially a knowledge file in markdown, and reference an API by writing its slug, for example /menu for a restaurant's live menu or /stock for a stock check. When a customer message arrives, NOXUS scans the reply data, fetches every referenced API in parallel with a six-second timeout and a 4 KB cap, and injects the responses into the prompt sent to Gemini alongside each API's description, so the assistant can answer with current data. On top of the AI layer there is a full WhatsApp REST API covering sessions, messages, media, groups, channels, and labels, plus a React dashboard for managing sessions, AI settings, webhooks, infrastructure, and an audit log. There is a per-chat cooldown to stop reply storms, group replies are off by default, and every interaction is logged under data/ai for review. The stack is NestJS 11 on Node 22 with React 19, ships as Docker images, and uses SQLite or Postgres for storage. The README links a hosted version at noxusai.online for users who do not want to self-host. The project is MIT licensed.