Dashboards show you what happened. They do not tell you what it means. You can stare at a week of charts and miss the fact that your coding output drops 40% every Wednesday afternoon, or that your best sleep weeks coincide with no Spotify listening after 10 PM.
The AI weekly digest finds these patterns for you.
What Gets Analyzed
Every week, a Supabase edge function aggregates your data across all sources:
- App usage — total screen time, productive vs. distracted ratio, top apps, category breakdown
- Coding — hours coded, languages used, projects worked on, session patterns
- Music — listening hours, top genres, listening-while-coding correlation
- Health — average steps, sleep duration, heart rate trends, workout frequency
- GitHub — commits, PRs opened/merged, review activity
- Locations — home vs. office time, travel days
This aggregated snapshot — numbers, not raw events — gets sent to an LLM via OpenRouter. The prompt asks for three things: patterns worth noting, anomalies compared to the previous 4-week average, and one specific actionable recommendation.
What You Get Back
The digest is a structured markdown document with sections for each data domain. But the most valuable part is the cross-domain insights — things that require combining data sources:
- "Your coding sessions on days you hit 8,000+ steps averaged 4.2 hours vs. 2.8 hours on sedentary days"
- "You listened to 40% more music this week but your focus time was unchanged — music may be compensating for a noisier environment"
- "Three of your five longest coding sessions this week happened after 10 PM. Your sleep quality those nights averaged 62% vs. your usual 78%"
No human would cross-reference step count with coding duration. The AI does it because all the data is in the same place.
The Insights Page
Digests appear on the Insights page in the dashboard, sorted by week. Each one is rendered as styled markdown with the xeve design system — orange section headers, monospace typography, and clean hierarchy. You can also trigger a manual generation from the Recaps page if you do not want to wait for the weekly schedule.
Cost and Privacy
Each digest costs about $0.02-0.05 in LLM API usage. The function sends aggregated statistics — "4.2 hours of coding, 7,200 steps, 6.8 hours sleep" — not raw event logs or window titles. Your actual browsing history, app names, and file paths never leave xeve's database.
The model sees numbers and categories, not content. It cannot determine what you were working on, what you were reading, or what specific apps you used. It only sees the productivity classification and duration.