I Started Biohacking With an AI Agent

I got into biohacking the stupid way: I watched Bryan Johnson’s stuff. My first thought was that this is the most expensive hobby in the world. My second thought was that even with his money I would never keep a spreadsheet like that. So I built the poor man’s version. Two watches I already owned, my Home Assistant, and an AI agent to do the boring parts.

Biohacking sounds like something from a lab but it is mostly just logging. You wear sensors, you collect numbers, you wait for a pattern to show up. The part that always kills these experiments is the logging itself.

The hardware

The sensors were already in my drawer. Galaxy Watch 5 on my wrist at night for sleep, Galaxy Watch 4 Classic during the day. Nothing fancy, no blood work, no special diet. The watches feed my phone, the phone feeds Health Connect and Google Fit, and Google Fit feeds Home Assistant. The data was flowing before the bot existed. It just sat in apps I never opened.

The agent

Hermes Health is a Telegram bot. A dedicated agent on my server, its own little brain, its own tools, none of my other stuff in its context.

Every morning at seven it does three things. It reads the overnight sensor data from Home Assistant. It checks what I actually ticked off in yesterday’s daily note. And it writes a short insight into today’s note, then messages me the summary. Weight, brushing, watch battery, whatever logged itself, plus the habits from my checklist.

The best part is the pattern catching. Three short nights in a row, then my chess rating slides, and the bot will say both sentences in one message. I would never spot that on my own.

It writes straight into my Obsidian vault. Every daily note has a Habits & Health section where the numbers land on their own. Sleep, steps, water, workout, brushing, weight, blood pressure, body fat. I still tick the checkboxes that need a human. The sensor numbers just appear.

The loop

Measure, log, read, adjust. That is the whole of biohacking, and the agent closes the loop so I only do the last step. Sensors measure. The bot logs and comments. I read one short summary at breakfast and decide what to change.

Bryan Johnson has a team for this. I have a cron job and a two line message.

The rules I gave it matter more than the code. No nagging. No guilt trips. Evenings silent unless I ask. It is allowed to ask “want me to set an earlier bedtime reminder?” but it never tells me to go to sleep. And the data stays in the vault and the chat. Nobody else sees it.

Status

It is still an experiment, three weeks in. The morning summary turned into a habit, which is more than my previous logging attempts can say. Whether the daily trend line actually changes what I do at 1 AM, no idea yet. But the measurement layer is done, and that was always the hard part.

I am not Bryan Johnson. I wear two watches and read a two line summary every morning, and honestly that might be enough.