Erik Bjäreholt is a Swedish developer who builds open-source infrastructure for personal data. He founded ActivityWatch, the most popular free time-tracker with 12,000+ GitHub stars, and gptme, a terminal-based AI agent with 4.2k stars. His work sits at the intersection of quantified self, privacy-first software, and AI tooling.
The Problem with Data Silos
Erik’s core belief: “A lot of data is collected about us daily, but most of it does not end up in our hands.” This frustration led him to build ActivityWatch in 2016. Commercial tools like RescueTime send your activity data to their servers. ActivityWatch keeps everything local.
The architecture is simple. A core server runs on localhost:5600. Watchers collect data — active windows, browser tabs, editor activity. Everything stays on your machine. You own your data completely.
ActivityWatch: Privacy-First Time Tracking
ActivityWatch tracks what you’re doing automatically:
- Window watcher: Active application and title
- Browser extension: Current tab URL (Chrome, Firefox)
- Editor plugins: What files you’re coding
- AFK detection: When you step away
No accounts. No cloud sync. No surveillance capitalism.
The tool gained traction among developers and privacy advocates who wanted RescueTime-level insights without the data export. It runs on Windows, macOS, Linux, and Android.
gptme: AI Agent in Your Terminal
Erik’s newer project gptme takes a different angle on personal AI. It’s a command-line agent that can:
- Write and edit code
- Execute shell commands
- Browse the web
- Process images
- Work with local files
The key difference from web-based AI assistants: gptme runs locally with full access to your system. It’s your AI, on your machine, using your tools.
# Example usage
gptme "create a Python script that analyzes my git commit frequency"
The agent thinks, writes code, executes it, debugs errors — all in a terminal conversation. It supports multiple LLM backends including OpenAI, Anthropic, and local models.
Quantified Self Philosophy
Erik maintains detailed notes on quantified self methodology. His approach distinguishes between:
Automatic tracking (low effort, high consistency):
- Computer activity via ActivityWatch
- Location via Google Timeline
- Fitness via Mi Band
- Sleep via wearables
Manual logging (high effort, unique insights):
- Mood tracking 3x daily
- Productivity self-assessment
- Drug/supplement intake
- Custom spreadsheets
He found that mood and productivity are “highly correlated” in his data — unsurprising, but validating through numbers matters.
The Privacy Tradeoff
Erik draws a clear line: “I don’t use things like keyloggers but plan to use keycounters.” Full keystroke logging captures too much sensitive data. Character counts give productivity signals without the privacy risk.
His rule: the benefit of data must outweigh the risk of exposure. ActivityWatch doesn’t sync by default because local-first is safer than convenient.
Building for the Long Term
From the ActivityWatch site: “We don’t work on ActivityWatch full-time, so development may slow at times. But we think that ActivityWatch is important software, and we’re building for the long term.”
This is sustainable open-source philosophy. No VC funding, no growth pressure. Just slow, steady improvement on tools that matter to the people who use them.
Key Takeaways
- Own your data: If a service won’t export your data, you don’t really have it
- Local-first wins: Privacy and performance improve when computation stays on your machine
- Automate what you can: Manual logging creates unique insights but demands discipline
- Build tools you need: ActivityWatch exists because Erik wanted it to exist
- Sustainability over speed: Slow development beats abandoned vaporware
Links
- ActivityWatch — open-source time tracker
- gptme — terminal AI agent
- Erik’s website
- Quantified Self wiki
- @ErikBjare on Twitter