My fork of ActivityWatch is ready
Hey everyone, it's been a week since I posted my idea for restructuring ActivityWatch, and today's finally the day: I've published my fork of the project, as the maintainers asked me to.
When I wrote the first post, I figured I'd just modernize the UI and leave everything else as-is. Then I started digging into the code, and one thing led to another: week after week the project changed way more than I'd planned. What came out of it isn't "ActivityWatch with a new skin" anymore — it's basically a different app, built on the same foundations.
Under the hood, the entire Python stack has been removed — including system notifications, which now run entirely through native Rust/Tauri. The frontend was rewritten from scratch, completely decoupled from Bootstrap. Auto-updates were implemented from the ground up, so you never have to manually download the latest version again. And configurable privacy filters, so sensitive data gets dropped or masked before it's even written to disk.
The base watchers existed in the original project already — I'm not taking credit for that, it'd be dishonest. What I did was rewrite and simplify the whole chain that manages them, so everything else could be built on top without fighting legacy code at every step.
Same story for custom watchers: the ability to track your own data existed conceptually before too, but it meant writing a full program that followed an internal protocol. What I changed is the experience: now you just write any script (PowerShell, Python, JS, .cmd/.bat, or a ready-made executable) and the app handles the rest — it detects the right interpreter on its own, walks you through creating it step by step, shows a live log so you immediately see if something's off, and offers 8 ready-made display templates to show your data on the Home page without writing a line of HTML.
And here's the part I'm most proud of: I integrated a real AI agent, not some gimmick bolted on for show. You can open a chat with it and ask literally anything about your data — "how much did I work on this project yesterday," "which apps do I use most in the morning," "summarize my week" — and it queries your actual data on its own and answers you. The same agent also helps auto-categorize the apps you use, instead of doing it by hand one by one.
A bunch of other things I hadn't mentioned yet:
- Periodic screenshots: a feature that simply didn't exist before, now built in — the app captures screenshots at regular intervals and links them directly to the exact moment in your day they were taken.
- App categorization rewritten from scratch: organizing your apps into categories is much, much easier than before.
- "Always active" pattern: same deal, handling it is now way more straightforward.
- App icons: didn't exist before, now every app shows its real icon instead of a generic colored block.
- Colors extracted from icons: no more random greens pulled from a name hash — Timeline colors now come from the app's actual icon, much more consistent and recognizable at a glance.
- Notification system rewritten from scratch: you can now create custom notifications — whenever you want, for whatever you want (category, app, project, idle time, VPN session) — without touching a line of code.
- New Home modules, freely reorderable.
- Projects section with start/pause timer, hour budgets, and overrun alerts.
- Redesigned "Raw data" page, with a dedicated Timeline and a much more readable events table.
Here's the project's website, where you can take a look and download the Windows installer: https://codex-cpp.github.io/TrackFlow/
