Sign in with your phone number to sync across all platforms
A written read of everything you have logged: what actually changed, what is driving it, and the single lever worth pulling next.
Your data syncs across all platforms. Use the same phone number to link accounts.
Install BioTrakk on your device for the best experience — faster access, offline support, and home screen shortcut.
To install on iOS:
App installed! You can open BioTrakk from your home screen.
Control which proactive reminders you receive. All are enabled by default.
Used for the interface and for the assistant's replies.
Your day as one thread: log, see your targets and get insights in a single feed. Switch back any time.
Theme
Text size
How the assistant talks to you.
A recap of your day — calories, activity, streaks and insights. Sent in your own timezone.
Your data belongs to you. Export or delete at any time.
The AI assistants allowed to read your health data, and to change it where you allowed that. Take access away at any time.
Report bugs, request features, or share your thoughts.
Signed in as
Every control in the web app reports a tap by name, never by content. Users is how many different people used it in the window; a control used by nobody is listed at the bottom, and that list is the removal candidates.
Every message the app sends on its own — summaries, nudges, goal questions — and whether it got there. Nothing recorded this before, which is how every proactive message in the product came to be failing for months without anyone noticing. The message itself is never stored, only that it was attempted.
Tokens as the providers report them, priced per model. Cached is prompt text served from the cache at a tenth of the input rate — on the chat path that is most of the prompt. A model with no price on file still counts its tokens and is marked, so the total is never quietly short.
Model used for food detection and logging. User's personal setting overrides this.
Auto-fallback if the primary fails. Pick a different provider for best redundancy.
Model used for chat conversations, health questions, and data analysis.
Auto-fallback if the primary fails. Pick a different provider for best redundancy.
Weighed meals with known truth, used to measure photo estimation accuracy (npm run eval:photos).
To add one: log a meal by photo from the app as usual. The photo is kept here for a week, filled in with what the model said. Weigh what is on the plate, correct the entry to the real grams and calories in the normal editor, then choose Mark as golden truth from the entry’s menu, or confirm it below. One photo per meal.
Runs the whole golden set against each selected model and scores calorie accuracy and food recognition. A score is only worth what the set is worth — check the photo count before reading one.