Daemon & Autostart

On-wake trigger vs daemon — which do you need?

On-wake trigger (built-in, zero setup) — consolidation runs in the background at the start of any claude-review command if the time or volume trigger is met. This is right for 99% of users:

• Works on any machine, no process to manage
• Pauses when you're not using the tool (closes laptop, goes on holiday)
• Catches up automatically next time you run a review
• No installation, no system permissions

Background daemon (optional) — a persistent process that checks every 5 minutes and runs consolidation independently of your usage. Worth it only in specific cases:

Use case	Right choice
Laptop / normal developer machine	On-wake trigger
CI pipeline	On-wake trigger (fires during each review run)
Shared CI server with infrequent claude-review invocations	Daemon
Always-on workstation where you want insights updated overnight	Daemon

If you find yourself wondering "should I run memory start?" — the answer is almost certainly no. The on-wake trigger handles it.

What the daemon does

The daemon wakes up every 5 minutes, checks if consolidation should run (30-minute time trigger or 10-finding volume trigger), and if so calls the Consolidation Agent. This agent:

1. Reads all findings stored since the last consolidation
2. Groups them by file, severity, and description similarity
3. Calls the Anthropic API with metadata only (no source code) to identify patterns
4. Writes structured consolidation records to the consolidations table

The consolidation results power the claude-review insights command and the memory context injected into future reviews.

Daemon commands

claude-review memory start    # start the daemon
claude-review memory stop     # stop the daemon
claude-review memory status   # show running status and stats

PID file

The daemon writes its PID to ~/.claude-review/memory.pid. If the process dies unexpectedly and leaves a stale PID file:

rm ~/.claude-review/memory.pid
claude-review memory start

Autostart on macOS (launchd)

claude-review memory install

This creates a launchd plist at ~/Library/LaunchAgents/com.critbot.claude-review-memory.plist and loads it immediately.

To verify:

launchctl list | grep claude-review

To unload manually:

launchctl unload ~/Library/LaunchAgents/com.critbot.claude-review-memory.plist

Autostart on Linux (systemd)

claude-review memory install

This creates a systemd user service at ~/.config/systemd/user/claude-review-memory.service.

To enable and start:

systemctl --user daemon-reload
systemctl --user enable claude-review-memory
systemctl --user start claude-review-memory

To check status:

systemctl --user status claude-review-memory
journalctl --user -u claude-review-memory -f

Consolidation triggers

The daemon consolidates when either condition is met:

Trigger	Value
Time elapsed since last consolidation (with ≥1 new finding)	30 minutes
New findings stored since last run	10 findings

On the very first run with any findings stored, consolidation runs immediately regardless of elapsed time.

These are not currently configurable, but tuning them will be added in a future release.

Resource usage

The daemon is extremely lightweight:

• Memory: ~5–10 MB RSS
• CPU: Negligible except during consolidation (a few seconds every 6 hours)
• Network: Only during consolidation — sends a small metadata payload to the Anthropic API
• Disk: SQLite DB grows at ~50–200 bytes per finding; a year of active usage is typically under 5 MB

Running without the daemon

The daemon is optional. Without it:

• --memory flag still stores and retrieves findings
• No consolidation runs automatically
• claude-review insights returns limited results (no cross-PR patterns)

To manually trigger a consolidation:

# Not yet exposed as a direct command — start the daemon to trigger consolidation,
# then stop it once it completes
claude-review memory start
# Wait ~30 seconds for first consolidation
claude-review memory stop

Manual consolidation triggering will be a proper command in a future release.