activity-record

name	activity-record
description	Query and orchestrate Arkloop Activity Record local activity data. Covers browser history, search terms, screen time, bluetooth, shell commands, window focus, keyboard, mouse, clipboard, screen content (accessibility tree), microphone audio transcription, and Codex sessions.

Activity Record

Activity Record stores normalized local activity facts in:

~/.Arkloop/activity-record/activity.db

Use bounded SQL queries. Always include a time range or LIMIT.

Tables

activity_events

Column	Description
occurred_at	RFC3339 timestamp
source	Source key (see Sources below)
source_event_id	Stable dedup key
app	App or profile name
window_title	Active window title when available
url	URL when available
action	Event action
title	Short event title
text	Longer text when available
metadata_json	Source-specific JSON
ref_kind, ref_key	Optional reference pointer

source_cursors

Incremental sync state per source. Diagnostics only.

activity_refs

Optional reference files.

Sources and Actions

chrome — Chromium Browser History

Covers Chrome, Chrome Canary, Chromium, Edge, Brave. Multi-profile.

Action	Title	Metadata
visited	Page title	profile, duration_sec, foreground_sec
downloaded	Filename	profile, path, size_bytes, mime_type
searched	Search term	profile, normalized_term

safari — Safari History (macOS)

Action	Title	Metadata
visited	Page title	domain, visit_count, load_successful, score

screentime — macOS Screen Time (knowledgeC.db)

Action	Title	Metadata
app_used	App name	bundle_id, source_bundle, device_id, duration_sec
screen_on	"screen on"	duration_sec
screen_off	"screen off"	duration_sec
notification_received	App name	bundle_id
media_used	App name	bundle_id, duration_sec, url, media_url
media_playing	Track title or app	bundle_id, duration_sec, playing, artist, album, genre
web_used	Domain	bundle_id, web_domain, duration_sec
bluetooth_connected	Device name	device_name, device_type, is_apple_audio_device, duration_sec
bluetooth_disconnected	Device name	Same as above

shell — Shell Command History

Reads ~/.zsh_history and ~/.bash_history. Deduplicated by command hash.

Action	Title	Metadata
command	Command (truncated)	shell, line_num. Full command in text column.

codex — Claude Code Sessions

Action	Title	Metadata
prompted	Prompt text	session_id, cwd
received	Response text	session_id, cwd, model

window — Active Window Tracking (daemon)

Action	Title	Metadata
focused	"App - Window Title"	duration_sec, pid
idle_start	"idle"	idle_threshold_sec
idle_end	"idle"	—

keyboard — Keystroke Counting (daemon)

Action	Title	Metadata
keystroke_count	"N keystrokes in App"	count, interval_sec, app, window_title

mouse — Mouse Activity (daemon)

Action	Title	Metadata
mouse_activity	"N clicks, N scrolls in App"	clicks, scrolls, interval_sec, app, window_title

clipboard — Clipboard Changes (daemon)

Action	Title	Metadata
clipboard_changed	Content preview	length. Full text in text column when enabled.

ax — Screen Content via Accessibility Tree (daemon, macOS)

Captures visible text from the focused window by walking the macOS accessibility tree. No screen capture or OCR involved — zero WindowServer overhead.

Action	Title	Metadata
ax_snapshot	"App - Window Title"	element_count, text_length, walk_duration_ms, content_hash, pid, truncated, truncation_reason

Text content: full visible text from the focused window is stored in the text column. Content is deduplicated by SHA-256 hash — identical screens do not produce repeated events.

Browser URL: extracted via AXDocument attribute when the focused app is a browser. Available in the url column.

Excluded apps: password managers (1Password, Bitwarden, KeePassXC, LastPass, Dashlane, Keychain Access) and incognito/private browsing windows are skipped entirely.

audio — Microphone Transcription (daemon)

Transcribes microphone speech via an OpenAI-compatible API. Each segment passes an RMS gate, voice-activity detection, and music filtering before transcription, so only segments containing speech are stored.

Action	Title	Metadata
audio_transcription	Transcript preview	duration_sec, device, language, model. Full transcript in text column.

Text content: the full transcript of each speech segment is stored in the text column. Overlapping words between consecutive segments are deduplicated.

Examples

Recent activity across all sources:

SELECT occurred_at, source, action, title, url
FROM activity_events
ORDER BY occurred_at DESC
LIMIT 50;

What apps were used today (Screen Time):

SELECT title AS app, COUNT(*) AS sessions,
       SUM(json_extract(metadata_json, '$.duration_sec')) AS total_sec
FROM activity_events
WHERE source = 'screentime' AND action = 'app_used'
  AND occurred_at >= datetime('now', '-1 day')
GROUP BY title ORDER BY total_sec DESC
LIMIT 20;

Recent browser searches:

SELECT occurred_at, title AS search_term, url
FROM activity_events
WHERE source = 'chrome' AND action = 'searched'
  AND occurred_at >= datetime('now', '-7 days')
ORDER BY occurred_at DESC
LIMIT 50;

Recent shell commands:

SELECT occurred_at, text AS command
FROM activity_events
WHERE source = 'shell' AND action = 'command'
ORDER BY occurred_at DESC
LIMIT 30;

Bluetooth device connections:

SELECT occurred_at, action, title AS device,
       json_extract(metadata_json, '$.duration_sec') AS duration
FROM activity_events
WHERE source = 'screentime' AND action LIKE 'bluetooth%'
ORDER BY occurred_at DESC;

Music playing history:

SELECT occurred_at, title,
       json_extract(metadata_json, '$.artist') AS artist,
       json_extract(metadata_json, '$.album') AS album
FROM activity_events
WHERE source = 'screentime' AND action = 'media_playing'
ORDER BY occurred_at DESC
LIMIT 20;

Window focus timeline:

SELECT occurred_at, app, window_title,
       json_extract(metadata_json, '$.duration_sec') AS sec
FROM activity_events
WHERE source = 'window' AND action = 'focused'
  AND occurred_at >= datetime('now', '-1 day')
ORDER BY occurred_at DESC
LIMIT 50;

Recent screen content (what the user is reading/doing):

SELECT occurred_at, app, window_title, url,
       substr(text, 1, 500) AS screen_text,
       json_extract(metadata_json, '$.element_count') AS elements
FROM activity_events
WHERE source = 'ax'
  AND occurred_at >= datetime('now', '-1 day')
ORDER BY occurred_at DESC
LIMIT 30;

Screen content for a specific app:

SELECT occurred_at, window_title, url,
       substr(text, 1, 800) AS screen_text
FROM activity_events
WHERE source = 'ax' AND app = 'Google Chrome'
  AND occurred_at >= datetime('now', '-4 hours')
ORDER BY occurred_at DESC
LIMIT 20;

Recent audio transcripts (meetings, conversations):

SELECT occurred_at,
       substr(text, 1, 500) AS transcript,
       json_extract(metadata_json, '$.duration_sec') AS dur
FROM activity_events
WHERE source = 'audio'
  AND occurred_at >= datetime('now', '-1 day')
ORDER BY occurred_at DESC
LIMIT 30;

Memory Rules

Write only durable context to memory_write:

• Stable preferences and habits.
• Ongoing projects and decisions.
• Important events with useful future context.
• Clear changes in workflow or priorities.

Do not write Notebook entries. Do not persist raw OCR, raw transcripts, app-switch logs, secrets, one-time codes, or unverified fragments.