I’ve been experimenting with messaging-native AI agents and built an early prototype of an AI nutrition coach that works through iMessage (WhatsApp support coming soon). You send it a meal photo or a voice note, and it estimates calories/macros and logs everything automatically. It also builds a daily plan and sends weekly summaries based on your actual behavior.
The goal is to explore whether messaging can reduce the friction and drop-off that happens in traditional food-tracking apps. Early testers seem to stick with this format much more consistently.
Some technical details for context:
Vision & reasoning powered by a combination of Gemini and Perplexity
Agent runtime using the Vercel AI SDK
Gateway layer with Vercel’s GW
Image → nutrition estimation via serverless functions
iMessage integration through a local relay setup
WhatsApp Cloud API integration in progress
Still very early, and I’m trying to understand if a chat-first UX can make nutrition tracking feel effortless instead of a chore. Happy to hear feedback on the approach, concerns I should think about, or any similar systems worth studying.
Comments URL: https://news.ycombinator.com/item?id=45953276
Points: 1
# Comments: 0
Source: habitsdm.com
