Every improvement, feature, and fix — documented as it ships.
Opening the map used to load every location you've ever saved, which could take a while and pull in a lot of data. The map now starts on today's locations and loads quickly — you can still switch to a week, month, year, or all time whenever you want the bigger picture.
Rows you created on the web or another device wouldn't appear in the terminal client (TUI) while you had it open — the grid only ever refreshed when you started editing a cell or restarted the app, so new entries from elsewhere seemed to never arrive. The TUI now checks the server for changes in the background and quietly refreshes when you're sat at the top of the list, so rows added on another device show up on their own within a few seconds. It holds off while you're mid-edit or scrolled down reading older entries, then catches up the moment you come back to the top. Claude's local cache, which the TUI also keeps in sync, had a related issue where its sync marker could skip rows created on other devices; it now tracks the server's own timestamps so those entries are no longer missed.
Adding a row while your connection briefly stalls could queue it for sync without you noticing. If the app then refreshed the list from the server before that row finished syncing, it would disappear from the view until you reopened the app. Rows with edits still waiting to sync are now kept on screen through a refresh, so what you just saved stays visible.
On networks that connect but don't actually reach us — airplane or hotel WiFi that only allows a few apps, captive portals — your device reports itself as online, so the app used to keep trying to reach the server and never switched to offline mode. We now actively probe reachability with a short timeout, so the app drops into offline mode promptly: your changes are saved locally and queued to sync, and the offline banner appears as it should.
Fixed a problem where rows added while offline (for example, on a plane) wouldn't appear in the list — and stayed hidden even after reopening the app — until you reconnected. The app now rebuilds the rows view from your local offline copy when there's no connection, so newly added rows are visible immediately. Your changes were always saved safely and synced on reconnect; this fixes only the display.
You can now create, edit and delete rows (and GPS points) with no connection. Changes are saved to a local store and queued; the moment you're back online they sync automatically, and edits made on your other devices flow back in. Notes search also works offline now, instantly, against a local mirror of your data. Conflicts use last-write-wins, and the full edit history remains available for recovery.
The installed app now launches without a connection. Previously, opening rows.life on an iPhone home screen while offline showed a blank screen — even though your data was cached locally, the app shell itself never loaded. The service worker now caches the app shell document and serves it for offline launches, so you can open the app and browse your cached entries with no signal.
Excel export still defaults to exporting everything — a full backup — but you can now narrow it down. In Settings → Export, with Excel selected, tick Filter what to export to choose a date range and a category (optionally a single subcategory). Picking a parent category includes all of its subcategories. Handy for pulling just a client's hours or a month's entries into a report without hand-trimming the spreadsheet afterwards. SQLite export is unchanged and always exports the complete database.
The map could stall when opening a range with many thousands of GPS points (e.g. after importing a long track). It now opens smoothly: the "furthest apart" stat skips a brute-force pass, marker pins are built and inserted in bulk, and pin details are rendered only when you actually click a pin.
The Jump to heatmap on the map could show a non-zero count for a week or month that had no points to display — clicking the cell opened an empty map. This happened when a row had GPS points added manually (or imported) whose own timestamps fell outside the row's day: the heatmap counted those points under the point's timestamp, while the filter that loads the map fetches rows by the row's start time, so the two disagreed. The heatmap now buckets every point by its parent row's start time, matching what the map actually renders when you click the cell.
When you share a photo library with someone — partner, family — and you both have the same iPhone model, rows import-gps used to lump everyone's photos under the same device label ("Apple iPhone 14 Pro"), so --device couldn't separate them.
The importer now also reads the EXIF HostComputer field, which is where Apple's Camera app writes the device's user-set name (the one in Settings → General → About → Name). When present, it gets appended to the device label, so two same-model phones show up as separate entries in the device summary and you can filter them apart:
Devices found:
1,243 Apple iPhone 14 Pro — Pablo's iPhone
872 Apple iPhone 14 Pro — María's iPhone
48 Apple iPhone 14 ProThen --device "Pablo" (case-insensitive substring) imports only yours.
There's also a new --note-device flag. Used together with --create-missing, it appends a line to each new entry's notes saying which device(s) the photos came from — e.g. "Pictures taken with Apple iPhone 14 Pro — Pablo's iPhone". Handy when you import a partner's photos and want the entry itself to remember whose camera it was, without rummaging through metadata later. The preview output shows the device suffix too, so you can see what the notes will say before committing the import.
Two caveats on the HostComputer side. It only works for photos, not videos — Apple's .mov metadata doesn't carry the device name. And it only helps if the phones have non-default names: if both are still called "iPhone", there's nothing to distinguish. Renaming a phone affects future photos only; historical ones keep whatever was set when they were shot. Photos that came through metadata-stripping services (WhatsApp, Telegram) won't carry HostComputer either.
Named locations (your @home, @office, …) are the one place coordinates were historically stored unencrypted on the server, while everything else — notes, GPS points, category names — has always been encrypted in your browser before it leaves your device. We've been migrating those anchor coordinates under the same zero-knowledge encryption; each client quietly re-encrypts them the next time you log in.
This release closes the remaining gap: once a location's coordinates have been encrypted, the API stops returning the old plaintext copy entirely — it's dead weight the apps already ignore. Plaintext is now sent only for locations that haven't been re-encrypted yet, and solely so the migration can finish the job. We also fixed an edge case in the desktop/TUI client where a half-written record could be mistaken for "already encrypted" and skipped, leaving its plaintext stranded.
No action needed on your part, and nothing changes in how locations look or behave.
On phones, picking a specific scope (Day, Week, Month, Year) made the date stepper push the toolbar wider than the screen — the next-day arrow and the Area button got clipped off the right edge, so there was no easy way to draw an area filter. The toolbar now stacks cleanly: the scope pills and the date stepper each take their own full-width row, the date label truncates instead of wrapping, and the Area button (with its label) stays fully visible next to the category filter.
The new Filter by area rectangle didn't work on touch devices — drag-to-draw conflicts with single-finger pan, so no rectangle would ever land. Touch now uses a two-tap flow instead: tap the Area button, tap one corner of the area you want, tap the opposite corner, done. A small red dot marks the first corner so you can see what's coming, and pan/zoom stay enabled between taps so you can reposition the map.
Two related fixes on phones:
Desktop drag-to-draw is unchanged.
A new Area button in the map toolbar lets you drag a rectangle on the map. Once drawn, only locations inside that rectangle show up — date scope and category filters still apply on top, so you can ask things like "how many days did I spend in London in 2023?" with one selection.
A Locations panel appears alongside the map listing the distinct days you've been in the selected area, with the per-day pin count. Click a day to expand its entries; click an entry to pan the map to that point. The headline number — N days in area — is the simplest honest answer to "how often have I been there?"
On mobile, the panel becomes a second tab next to the map (only visible while a filter is active). The mobile toolbar also got a small layout fix: km apart no longer gets clipped at the right edge of the screen.
Press Esc while drawing to cancel; click the Filtered by area chip in the toolbar to clear an active filter.
Settings has a new New rows tab that turns previously hardcoded behavior into preferences that sync across devices.
The default still applies — when you add a row at the bottom of the sheet and a long gap has passed since your last entry, the new row jumps to the current time instead of continuing from where the previous one ended (introduced in PR #12 with a hardcoded 5-hour threshold). You can now:
New: opt in to set end time to current time on new rows. Off by default. When on, every new row is created with both start and end already filled in — useful if you mostly track tasks you've just finished.
Everything is stored per-account and applies on web today, with TUI and mobile to follow.
When you have Set end time to current time on new rows turned on, pressing Enter repeatedly on a blank row at the bottom of the sheet used to create a stack of zero-duration rows with timestamps marching into the future and the focused row jumping a couple of positions up between presses.
Now: if the bottom row has no category, no subcategory and no notes, pressing Enter is a no-op — the cursor stays put on the row that's still waiting to be filled in. Once you've filled in any of those three fields, Enter creates a new row again as before, and the new row's end time is clamped so it never sits before its start.
Adding the new New rows tab pushed the Settings tab bar past the edge of the screen on phones — the rightmost tabs (and the page content) ended up cut off. Same thing was already happening to the Admin Analytics page once it grew to seven tabs.
On narrow viewports, tabs now wrap onto a second row with a pill-style active highlight, so every section is reachable without horizontal scrolling. Desktop layout is unchanged.
The map toolbar now shows a small X km apart stat right next to the pin count — the great-circle distance between the two furthest GPS points in whatever date range you have selected. Switch from Year to Month to Week and watch it shrink as the trips fall out of view.
It's the simplest honest answer to "how far did I get this year?" Picking diameter rather than path-length sidesteps the trap of pretending we know the route between sparsely-tagged points: a single Madrid pin and a single Tokyo pin from the same trip would otherwise read as a 10,800 km straight-line jump even though we have no idea what was traveled in between. X km apart just states the geographic fact about two points the user actually tagged, with no claim about routes.
Hidden when there are fewer than two points or the spread is under 1 km — a stay-at-home week stays clean, a trip across Europe lights up.
The in-app messages system now supports single-choice questions, not just announcements with a CTA button. When a message has options attached, the popup renders them as chips — picking one records the response and dismisses the message in a single click.
First use of it: a one-question survey for users with 5+ days of tracking, asking whether they'd prefer the End column to fill automatically with the current time when creating a new row.
The map's date filter has been rebuilt around a single question: which week should I be looking at? The old All time dropdown is gone; in its place there's a scope bar with All · Year · Month · Week · Day pills, a prev/next stepper for the current scope, and a new heatmap popover for jumping across years.
Scope pills + stepper. Pick the granularity once; the ◀ label ▶ arrows step ±1 of whatever scope you're in — next week, previous month, next year. One click, no re-selection.
Heatmap popover for big jumps. Click the label to open a 12-month × 5-week grid for the selected year. Each cell is colored by how many GPS entries landed in that week — so a week spent in Madrid lights up dark, a quiet week at home stays pale. The exact count is printed inside every cell (847, 2.4k, 16k) so you can triple-check what you're looking at.
Day scope swaps the heatmap for a regular month calendar — click any day to filter to it, including a Today shortcut.
Quieter data loading. Navigating years inside the popover is staged — the map only reloads when you actually apply a selection, so scrubbing back through 2014 → 2015 → 2016 doesn't trigger a cascade of fetches. Each year's heatmap density is fetched once and cached.
The heatmap counts individual GPS points (not rows), so it matches exactly what the map shows when you zoom in — a row with a full day of auto-tracking contributes all of its points to the week, not just one.
Ctrl+S to save and close (previously Ctrl+Shift+Enter, which caused issues with some terminal emulators). Enter continues to insert newlines for multi-line notes.Two quality-of-life improvements for the terminal client:
Adding a location to an entry that already had one resulted in duplicate GPS points with no way to remove them. The agent and MCP integration now support a delete_locations tool that removes all GPS points from an entry.
To fix a wrong location: ask the agent to delete the location, then add the correct one.
You can now save frequently-used places like @Home, @Office, or @Gym and reference them with a simple @ mention in your notes. Selecting a location automatically attaches GPS coordinates to the entry.
Three ways to create named locations:
@ autocomplete everywhere:
@ in the notes field (rows view, edit dialog, or mobile) and a dropdown appears with your saved locations
The rows agent and MCP integration can now edit existing entries and add GPS locations — two capabilities users had been asking for.
The delete row shortcut on macOS is now ⌘+Backspace, matching the standard macOS convention. Previously it required ⌘+Delete (forward delete), which many Mac keyboards don't have as a dedicated key. Windows and Linux shortcuts remain unchanged (Ctrl+Delete).
You can now delete time entries through the MCP integration and the rows agent. Claude can find and soft-delete entries on your behalf — deleted entries can be recovered within 30 days.
Use search or filter tools to find the entry first, then ask Claude to delete it.
New users now see a Take the tour button when they first log in, instead of a blank screen. The tour loads real demo data (2 years, 4,260 entries from a fictional Czech programmer) and walks you through every feature in 12 steps:
The tour works on both desktop and mobile, with adapted layouts for each. On mobile, stats panels slide in above the tour card so you can see both. Navigate with Next/Back buttons or arrow keys.
A tour icon in the sidebar lets you restart it anytime — it pulses red if you have fewer than 15 days logged, gently reminding you to explore.
Introducing rows agent — a conversational AI assistant that lives in your terminal and turns your rows.life data into a queryable personal memory. This is an early proof of concept that connects directly to the Anthropic API.
What it can do:
Smart enough to reason: With Claude Sonnet, the agent can distinguish between talking about Madrid in your notes vs. actually being there, cross-referencing GPS data with note content to give precise answers.
How to try it (requires your own Anthropic API key):
$env:ANTHROPIC_API_KEY="sk-ant-api03--your-key"; $env:ROWS_AGENT_MODEL="claude-sonnet-4-6"; rows agent
This is a proof of concept — the final product won't require you to bring your own API key. We're using it to learn what makes a great AI-powered life memory interface.
s from the table view to enter Stats mode with four tabs: Year in Review (bar chart of hours by category with drill-down into subcategories), Categories (browse all categories with all-time stats and subcategory breakdown), Overall (year-over-year comparison bar chart with summary table), and Trends (line chart with configurable date range, granularity, and category picker dialog). Navigate years with arrow keys, switch tabs with 1/2/3/4, drill down with Enter, and press c to select trend categories.
Delete or Backspace on any row and confirm with y — any other key cancels. The status bar shows the keybinding hint and a red confirmation prompt so you won't delete anything by accident.m on any row to see its location points on a true-color Mapbox map rendered with half-block pixel art. Pins are color-coded by category, and pressing n/p cycles through markers with a label showing time and category. Press a to toggle between the selected entry's locations and all your locations (fetched live from the API). Rows with GPS data show a ● indicator in the table so you can spot them at a glance. Pan with arrow keys, zoom with +/-, and switch between six map styles with s.
rows import-gps, this means every photo you took during an entry automatically becomes a pin on the map — so a 3-hour bike ride doesn't collapse to a single dot, it becomes a trail of everywhere you stopped to take a picture.The rows import-gps command now uses real distance calculations (haversine formula) to deduplicate GPS coordinates. Photos taken within 50 meters of each other on the same entry collapse into a single point — so ten selfies at the same restaurant become one pin, not ten. You can safely re-run the import on the same photo directory: existing coordinates are checked by distance, and only genuinely new locations are added. This also means importing overlapping photo libraries (e.g. iCloud copies + local backups) won't create duplicate pins.
The --create-missing flag is now much smarter. Instead of creating one row per day, it clusters photos by time: if there's more than a 1-hour gap between consecutive photos, it starts a new entry. Each entry gets the proper start and end times from its photo cluster, all GPS points (deduplicated within 50m), and a place name automatically resolved via OpenStreetMap reverse geocoding — so instead of a blank row, you get something like "Calle Mayor, Salamanca" as the note. The entries are created under the 'Location' category.
You can now enrich your entries by adding GPS locations extracted from your photo collections! Since photos and videos contain GPS coordinates in their EXIF metadata, the new rows import-gps command reads them and adds the locations to your rows.
For example:
$ rows import-gps C:\Users\pablo\Pictures\2021\ --device "Apple iPhone 7" --device "Apple iPhone 13" --previewThis scans all photos and videos in the directory, filters by the specified devices (your phones, not your family's), and shows a preview of which entries would be enriched — without changing anything. The command displays a summary of all camera devices found, so you can decide which ones to include.
Remove --preview to actually apply the changes. Photos are strictly matched to rows by time window: a photo only enriches a row if it was taken during that row's time range. Existing locations are never overwritten.
Use --create-missing to also create new 'Location' entries for dates where you have GPS-tagged photos but no matching row. Without it, only existing rows are enriched.
Supports JPEG, HEIC, PNG, TIFF, MOV, and MP4 files.
