Privacy isn't a setting — it's the architecture
LogAgent is built to understand your work without harvesting it. Capture, classification and analytics run locally and offline; whatever leaves your device is your choice alone. This page explains exactly what is recorded, what never is, and how long it stays.
Your data lives on your machine
There is no cloud account behind the tracking. Everything is computed and stored on the device you're working on.
What's captured — and what never is
LogAgent records the signals it needs to explain your day, and deliberately ignores the content behind them.
Captured — to explain your day
- Active app and window title — for categorization, not content
- Domains you visit — for web-activity classification
- Input activity as counts — keystrokes and clicks as numbers, never the keys
- Git commit metadata — counts, churn and a short message head
- AI session aggregates — tokens, cost, model and working directory
- Camera-derived metrics — posture angles, blink rate, fatigue score
Never captured
- Screenshots or screen recordings
- Keystroke content — what you actually type
- File contents or clipboard data
- Prompt and message content from AI tools
- Code diffs or full commit messages
- Passwords, form fields or private messages
- Camera video or photos
A camera that remembers nothing
Posture and fatigue monitoring uses your webcam — but the imagery never leaves the moment it's seen.
Local inference
Posture, blink, screen distance and pulse are computed by MediaPipe running in WebAssembly on your machine.
No frames, ever
Camera frames are never written to disk or sent anywhere — processing is frame-by-frame, in memory.
Only numbers persist
Just aggregate metrics — angles, blink rate, fatigue score — enter history; presence is a single boolean.
Off in one click
The camera runs only while enabled and the window is focused, and can be disabled at any time.
Measured, not read
The deepest signals — your AI sessions and your commits — are reduced to numbers, never their contents.
AI agents: measured, not read
Only tokens, cost, model and working directory are stored. Prompt and message content never enter the database; an optional 200-character preview is debug-only and off by default.
Git: metadata only
A local-only read of your .git records commit counts and churn, the first 80 characters of the message plus a hash — never diffs or file contents.
You decide what's kept
Detailed data is pruned over time while the compact, valuable summaries stay. Nothing about billing is ever deleted automatically.
| Data | Kept for | Notes |
|---|---|---|
| Raw events & timeline | 90 days | Full detail, then pruned |
| 5-minute aggregates | 1 year | Charts and trends |
| Daily totals | Forever | Compact, lossless summaries |
| Billable work segments | Forever | Never auto-deleted |
| AI turn preview | 200 chars max | Debug only, off by default |
Auto-vacuum keeps the database file compact. You can export everything or delete it at any time, and specific apps can be excluded from tracking entirely.
End-to-end encrypted — only if you ask
Sync to a paired device is off until you turn it on, and even then the server can't read your data.
What LogAgent will never do
A short, firm list — the things this architecture is specifically designed to prevent.
- Run ad networks or third-party tracking SDKs
- Sell, rent or share your data
- Read your content — messages, code, files or keystrokes
- Upload anything without your explicit opt-in
- Lock your data in — export and deletion are always available
Full legal detail
Read the complete Privacy Policy for data-controller details, your rights, and the regional regulations that apply.