Personal systems
Abby: a health coach agent that reads my bloodwork and checks in twice a day
Every lab report, blood panel and wearable export goes into one place. Abby, a coaching agent with her own goals file, diet plan and honesty rules, reads them, tracks what moved, keeps the plan current and checks in every morning and evening. She is polite about it. She is also relentless.
Most people’s health data is a drawer of PDFs, a wearable app they check for a week, and a vague memory of what the doctor said last year. Mine was the same. So I built the drawer a brain.
The record
Every lab report, blood panel, scan summary and consultation note goes into one structured store. The agent reads the PDFs, pulls every marker into a table with its reference range and date, and flags anything that moved between tests. Wearable exports (sleep, resting heart rate, steps, training load) land in the same place, so “how did I sleep the week before that blood test” is a question I can actually answer.
The point is not the data. The point is that nothing is lost, and the next question always has the last answer sitting behind it.
The plans
From the record, the agent maintains three living documents:
- A nutrition plan built around what the markers say, not a generic diet. It updates when the data does.
- A supplement plan with what, why, how much, and which marker each one is meant to move. If a marker has not moved after two tests, the supplement gets questioned.
- An annual testing calendar: which panels, when, and which ones are due because a previous result said “recheck in six months”.
The daily check-ins
This is the bit that makes it work. Abby is one of the agents in the Tommy framework, so she shares the same Telegram and the same brain. Every morning she sends the plan for the day: the supplement I keep skipping, the test that is due, the fact that I trained hard three days running and slept badly twice. Every evening she asks how it went and logs the answer.
She works from a small set of files she keeps up to date herself: goals, a daily plan, a diet plan, a coaching log, a check-in history and a list of insights from the data. One file is just about honesty: she is told to say when I am kidding myself, and to show the number that proves it.
It is not a fitness app that wants me to open it. It is an assistant that already knows, and says so.
What broke
Lab reports are the worst documents on earth. Every provider formats them differently, units change between labs, and one report put the reference ranges in a footnote. The parser now asks a second model to check the first model’s extraction against the original PDF before anything is written to the record, which caught a decimal point that would have made a normal result look alarming.
What I would tell you
Do the boring bit first: get every document into one place with a date on it. Everything interesting is built on that, and none of it works without it.