01The problem
Tracking what you eat only works if logging is effortless. Dedicated apps make you search databases and tap through forms; the friction wins after a week.
I wanted the opposite: tell someone what I just ate, in plain language, from the app I already have open, and have it recorded somewhere I control.
02The solution
Achu is a conversational assistant with the persona of a personal trainer and nutritionist. I message it on Telegram; it interprets the message and appends the entry to a local CSV.
It runs unattended on a repurposed Acer Aspire at home, built with OpenClaw as the agent framework.
The current flow
This diagram shows the current logging flow.
Key decisions
- 01
Own the data.
A CSV on my own machine, not a vendor’s database. Trivial to inspect, back up or load into anything else.
- 02
Zero new habits.
Telegram is already open all day; the assistant is one more chat.
- 03
Reuse hardware.
An old laptop is more than enough for a bot that waits for messages and calls an LLM API.
03Where AI fits
AI is the core of the product: an LLM interprets free-text messages, and the OpenClaw agent turns each one into a structured entry in the CSV.
It is the smallest end-to-end agent I run: a messaging interface, an LLM doing the interpretation, a persistence layer and a deployment that has to survive unattended.
04Results
- Free-text food messages are logged from Telegram to a local CSV, end to end.
Next — not yet built
- Daily kcal chart with 7- and 30-day moving averages, computed over days with valid records and labelled with their coverage (e.g. 5/7).
- Opt-in reminders for missing logs and deviations from a self-set target, with quiet hours.
- A synthetic demo dataset for this page. No real logs will be published.
Nothing here is health advice. Achu is a logging assistant with a persona, not a credentialed trainer or nutritionist.