backup-setup

Backups (local snapshots)

The installed worker snapshots ~/.claude-mem/claude-mem.db on a schedule (VACUUM INTO, consistent while sessions run) and keeps the newest N under ~/.claude-mem/backups/auto/. Each snapshot is one standalone SQLite file. Snapshots stay on this machine.

Preserve every unrelated setting and keep ~/.claude-mem/settings.json mode 0600. Never print the file's contents: it holds credentials.

Routing

Based on $ARGUMENTS:

• setup or empty: 1. Check status, then 2. Enable snapshots
• restore: 3. Restore
• status: 1. Check status only

1. Check status

npx claude-mem backup status

• Snapshot details printed: backups run. Report the last snapshot time, the snapshot count and any last error; stop unless the user asked for a change or a restore.
• "Automatic backups are not enabled": continue to step 2 if setting up.
• A "Last restore" line reports the outcome of the most recent restore.
• "Worker is not running": start it with npx claude-mem start and retry.
• "Unknown command: backup" or "Backup status failed: HTTP 404": this claude-mem (CLI or running worker) has no backup support yet, so the worker would ignore the backup settings. Stop. Tell the user that backups are not available in their installed version and that npx claude-mem update brings them once a release includes them. Do not write the settings in step 2, and never say backups are on.
• Any other failure ("Backup status failed: HTTP" with another status, "Backup request failed: ...", "Backup status failed: invalid JSON response", "Backup status failed: unexpected response") or any output not listed here: stop. Show the user the exact error line, do not write the settings in step 2, and never say backups are on. Do not suggest an update: this is an error in a version that has backups, not a missing feature.

2. Enable snapshots

Ask (or default): interval in hours (default 24, range 1 to 168) and how many snapshots to keep (default 7, range 1 to 100). Then write the settings without echoing anything else in the file:

node - <<'EOF'
const fs = require('fs'), os = require('os'), path = require('path');
const intervalHours = 'PASTE_INTERVAL_HOURS_HERE'; // e.g. '24'
const retainCount = 'PASTE_RETAIN_COUNT_HERE';     // e.g. '7'
const dir = path.join(os.homedir(), '.claude-mem');
const file = path.join(dir, 'settings.json');
fs.mkdirSync(dir, { recursive: true });
const settings = fs.existsSync(file) ? JSON.parse(fs.readFileSync(file, 'utf8')) : {};
const target = settings.env && typeof settings.env === 'object' ? settings.env : settings;
target.CLAUDE_MEM_BACKUP_ENABLED = 'true';
target.CLAUDE_MEM_BACKUP_INTERVAL_HOURS = intervalHours;
target.CLAUDE_MEM_BACKUP_RETAIN_COUNT = retainCount;
fs.writeFileSync(file, JSON.stringify(settings, null, 2) + '\n', { mode: 0o600 });
fs.chmodSync(file, 0o600);
console.log('backups enabled: every ' + intervalHours + 'h, keeping ' + retainCount);
EOF

Restart the worker so the schedule starts, then confirm with step 1. Tell the user backups are on only when that status reports them enabled:

npx claude-mem restart
npx claude-mem backup status

The first automatic snapshot lands about five minutes after the worker starts. To take one now:

npx claude-mem backup run

3. Restore

npx claude-mem backup list
npx claude-mem restore <file>
npx claude-mem backup status

restore stages the snapshot and restarts the worker (it also starts a worker that is not running). The new worker swaps the database in before it opens it, saves the database it replaced in the same folder as claude-mem-<timestamp>-pre-restore.db (restore that file to undo; it counts toward the retain count), and records the outcome. With cloud sync on, the restored database resumes syncing from the snapshot's position and can pull back changes that reached the cloud after it. backup status then prints a "Last restore" line: on success it names the saved copy; on failure it gives the reason, and the worker kept the database it had.

4. Report

Report the snapshot count, the last snapshot time, any error, and the last restore outcome if there was one.