LOCAL OBSERVATIONS
Observations that open directly to the evidence
Kept notices which words keep coming back, which went quiet, and which two entries belong together — and every observation opens to the exact entries behind it.
How it works
Unlike generic mood scoreboards or sentiment charts, Kept's Insights view is built on local document frequency and retrieval. It identifies meaningful patterns in your writing without cloud AI or external analytics.
Technical implementation
Four Computed Observations
The engine evaluates topic × mood contrast, persisting themes, quiet subjects, and paired entries once you reach seven entries.
Every Claim Opens to Evidence
Tap any observation card to immediately view and read the specific dated journal entries that produced that insight.
Local Document Frequency (TF-IDF)
Recurring terms and sparklines are calculated deterministically on your device using statistical document frequency without machine learning or cloud calls.
Echo: A Year Ago Today
The Today view highlights what you wrote exactly one year ago today (±3 days), providing a quiet temporal anchor.
Technical specifications
| Attribute | Specification |
|---|---|
| Observation Types | Topic×Mood, Persistence, Quiet, Paired |
| Threshold | Gated until 7 entries recorded |
| Algorithm | On-device TF-IDF document frequency |
| Evidence Link | 100% verifiable (Opens to source entries) |
| Network Requirement | Zero (Operates 100% offline) |
Frequently asked questions
Does Kept give advice or diagnose mental health in Insights?
No. Kept records, organises, and retrieves what you wrote. It does not provide psychological advice or diagnoses.
How does Echo choose which entry to display?
Echo retrieves entries written 365 days ago (within a 3-day window) from your local SQLite database.
Explore other architecture deep dives
Ready to start with Kept?