log2insight

A local, privacy-first "Screen Time for everything" tracker for macOS.

A small Python daemon polls what you're doing every few seconds and writes it to a local SQLite database for later analysis. Nothing leaves your machine — there is no network code, no telemetry, no cloud.

What it tracks

Signal	Source	Permission needed
Frontmost app	lsappinfo	none
Focused window title	Accessibility API (System Events)	Accessibility (+ Screen Recording on Sequoia+)
Per-process network I/O	nettop	none
System disk throughput	iostat	none
Idle / active state	ioreg HID idle time	none
Browser tab URL	AppleScript (Safari/Chrome/Edge/Brave/Arc)	Automation (per browser)
Apple Screen Time history	knowledgeC.db	Full Disk Access

A "zero-permission" core (app + network + disk + idle) works immediately; the richer signals light up as you grant permissions.

Quick start

cd log2insight

# 1. See what works right now and what needs a permission.
python3 -m log2insight doctor

# 2. Create the database.
python3 -m log2insight init

# 3. Try one foreground run (Ctrl-C to stop). Let it gather a few samples.
python3 -m log2insight run

# 4. See a summary.
python3 -m log2insight report --hours 24

Install it as a background agent that survives reboot:

python3 -m log2insight install     # writes a launchd plist and loads it
python3 -m log2insight status      # agent state + row counts
python3 -m log2insight uninstall   # stop and remove

Optionally install the console script so you can drop the python3 -m prefix:

pip install -e .
log2insight doctor

Menu bar app

A native macOS menu bar app gives you live status, one-click control, and a dashboard without the terminal. It's a thin controller — collection still runs in the headless launchd agent above; the menu bar app just views and drives it. It needs the menubar extra (rumps + PyObjC), kept out of the core so the daemon stays dependency-free:

pip install -e ".[menubar]"
log2insight menubar            # run it now (a greyscale bar-chart icon appears)
log2insight install-menubar    # …or launch it automatically at login

From the menu you can see whether collection is running and the live sample count, Start/Stop the agent, open a Dashboard (top apps, network, Apple Screen Time — a native WKWebView window), and edit Settings (poll interval and collector cadences). Saving settings writes ~/.log2insight/settings.json and restarts the agent so changes take effect.

No menu bar? log2insight dashboard renders the same dashboard into your default browser.

Granting permissions

System Settings → Privacy & Security:

• Accessibility → window titles
• Screen Recording → window titles on macOS Sequoia and later
• Full Disk Access → Apple Screen Time history (knowledgeC.db)
• Automation → browser tab URLs (prompts the first time per browser)

Grant them to whichever app runs the daemon — your terminal app for a foreground run, or the python3 binary that the launchd agent launches.

Configuration

All via environment variables (read at startup):

Variable	Default	Meaning
LOG2INSIGHT_DIR	~/.log2insight	data + log directory
LOG2INSIGHT_DB	<dir>/activity.db	database path
LOG2INSIGHT_INTERVAL	10	seconds between cycles
LOG2INSIGHT_NET_EVERY	1	collect network every N cycles
LOG2INSIGHT_DISK_EVERY	6	collect disk every N cycles
LOG2INSIGHT_KNOWLEDGE_EVERY	30	import Screen Time every N cycles
LOG2INSIGHT_IDLE_THRESHOLD	60	idle seconds before counted inactive

Data model

Everything lands in ~/.log2insight/activity.db:

• activity — one row per cycle: frontmost app, window title, browser URL, idle seconds, active flag
• net_samples — per-process bytes in/out over a ~1s window
• disk_samples — per-device throughput snapshots
• app_usage — Apple Screen Time events imported from knowledgeC.db
• meta — bookkeeping (e.g. last Screen Time import watermark)

It's plain SQLite — point any tool at it (sqlite3, Datasette, a notebook).

Daily email report (opt-in)

A separate launchd agent can email you a narrative of the previous weekday's activity — what you worked on (most-focused windows), apps by active time, browser domains, a focus/presence summary, network by process, and Apple Screen Time totals. It runs at 3pm, Monday–Friday; each run reports the most recent prior weekday (so Monday's email covers the previous Friday).

This is the only part of log2insight that sends anything off the machine, and only once you configure it. Sending uses your own Gmail account over SMTP, so you need a Gmail App Password (your normal password won't work, and you'll need 2-Step Verification on). The app password is stored in the macOS Keychain, never in a plaintext file.

# 1. One-time setup (prompts for the app password; nothing is echoed).
.venv/bin/python -m log2insight email-setup \
    --to you@example.com --from you@gmail.com

# 2. Preview without sending.
.venv/bin/python -m log2insight email-report --dry-run

# 3. Schedule it (3pm Mon–Fri).
.venv/bin/python -m log2insight install-emailer

# Stop it later:
.venv/bin/python -m log2insight uninstall-emailer

Non-secret settings (recipient, sender, SMTP host/port) live in ~/.log2insight/email.json. email-report --day YYYY-MM-DD re-sends any past day; --dry-run prints to the terminal instead of emailing.

Privacy note

This records a detailed history of everything you do, including window titles and URLs. The database is unencrypted on disk. Treat it as sensitive; it's .gitignored so you don't commit it by accident.

The collector itself makes no network connections. The only feature that sends data off the machine is the opt-in daily email report, and it emails a summary only to the address you configure, via your own Gmail account.