Update & Maintain¶

Pemeliharaan harian dan jangka panjang.

Daily checklist¶

Cek setiap hari (atau setup monitoring):

Service status: sudo systemctl is-active kai-bot
Disk space: df -h (alert kalo >80%)
Memory: free -h
Recent errors: sudo journalctl -u kai-bot -p err --since today
Bot responsive (chat di Telegram)

Cron script untuk auto-check:

cat > ~/agent/daily-check.sh << 'EOF'
#!/bin/bash
# Run via cron daily, alert via Telegram if issues

source ~/agent/.env

ISSUES=()

# Service check
if ! systemctl is-active --quiet kai-bot; then
    ISSUES+=("Service DOWN")
fi

# Disk check
DISK_PCT=$(df / | tail -1 | awk '{print $5}' | tr -d '%')
if [ "$DISK_PCT" -gt 80 ]; then
    ISSUES+=("Disk ${DISK_PCT}% full")
fi

# Memory check
MEM_PCT=$(free | grep Mem | awk '{print int($3/$2 * 100)}')
if [ "$MEM_PCT" -gt 80 ]; then
    ISSUES+=("RAM ${MEM_PCT}% used")
fi

# Recent errors
ERROR_COUNT=$(journalctl -u kai-bot -p err --since "24 hours ago" | wc -l)
if [ "$ERROR_COUNT" -gt 10 ]; then
    ISSUES+=("$ERROR_COUNT errors last 24h")
fi

# Alert if any
if [ ${#ISSUES[@]} -gt 0 ]; then
    MSG="Daily check issues:%0A"
    for issue in "${ISSUES[@]}"; do
        MSG="${MSG}- ${issue}%0A"
    done
    curl -s -X POST "https://api.telegram.org/bot${TELEGRAM_BOT_TOKEN}/sendMessage" \
        -d "chat_id=${OWNER_TELEGRAM_ID}" \
        -d "text=${MSG}"
fi
EOF

chmod +x ~/agent/daily-check.sh
(crontab -l 2>/dev/null; echo "0 9 * * * /home/ubuntu/agent/daily-check.sh") | crontab -

Weekly maintenance¶

Setiap minggu:

1. System update¶

sudo apt update && sudo apt upgrade -y
sudo apt autoremove -y

2. Cleanup cache¶

sudo apt-get clean
sudo journalctl --vacuum-time=7d
rm -rf ~/.cache/{pip,npm}

3. Disk audit¶

# Big folders di /home
du -h ~ --max-depth=2 | sort -rh | head -10

# Big folders di /
sudo du -h / --max-depth=2 2>/dev/null | sort -rh | head -10

4. Log rotation manual¶

Kalo log file lo bengkak (custom logs, bukan journal):

ls -lh ~/agent/*.log
# kalo gede:
mv ~/agent/agent.log ~/agent/logs-archive/agent-$(date +%Y%m%d).log
gzip ~/agent/logs-archive/agent-$(date +%Y%m%d).log

5. Backup verify¶

# Test backup readable
aws s3 ls s3://my-backups/ | tail -5

# Test restore (dry run)
aws s3 cp s3://my-backups/agent-latest.tar.gz /tmp/test-restore.tar.gz
tar -tzf /tmp/test-restore.tar.gz | head -5  # list files tanpa extract
rm /tmp/test-restore.tar.gz

Monthly maintenance¶

1. Review memory.json¶

cat ~/agent/data/memory.json | jq '.notes' | less

Hapus yang stale, redundant, atau prompt injection. Edit file langsung.

2. Archive history¶

History per chat bisa numpuk. Archive yang lama:

cd ~/agent/data/history
# Move file yang udah > 30 hari ke archive
find . -maxdepth 1 -name "*.json" -mtime +30 -exec mv {} archive/ \;

3. Audit credentials¶

ls -la ~/agent/credentials/

Token mana yang masih active?
Mana yang udah ga dipake?
Mana yang perlu rotation (sudah > 90 hari)?

Rotate token yang dipake: 1. Generate new token di provider 2. Update ~/agent/credentials/<platform>.env 3. Test agent jalan 4. Revoke old token

4. SOUL.md review¶

Buka ~/agent/SOUL.md, baca pelan-pelan:

Apakah identity masih akurat?
Apakah capabilities up-to-date dengan tool yang ada?
Apakah ada pattern aneh dari bot bulan terakhir yang harus di-fix di SOUL.md?
Hapus section yang ga relevan, tambah yang baru.

Setelah edit, reset history biar pattern lama ga affect:

/reset

di Telegram.

5. Cost audit¶

Kalo lo pakai paid services:

LLM API: berapa token dipake bulan ini? Cek dashboard provider.
VPS: kalo bukan free tier, berapa charge bulan ini?
Database / storage: berapa GB used?

Optimize kalo over-budget: - Switch LLM model ke yang murah (gpt-4o-mini, claude-haiku, llama via Groq) - Reduce history window di system prompt - Cleanup old data

Quarterly (3 bulan sekali)¶

1. Provider review¶

Apakah lo masih pakai provider yang paling cost-effective?

AWS Free Tier abis di bulan 12 → migrate ke Oracle / Hetzner
Provider X naik harga → consider alternative

2. Security audit¶

# Check SSH attempts
sudo journalctl _COMM=sshd --since "30 days ago" | grep "Failed password" | wc -l

# Cek active user
who
last -n 20

# Cek listening ports
sudo netstat -tlnp
# Hanya ada port yang lo expect (22 SSH + bot tertentu)?

# Cek file permission credential
ls -la ~/agent/credentials/
# Semua 600?

# Cek SOUL.md
ls -la ~/agent/SOUL.md
# 644 atau 600?

3. Backup restore test¶

Backup ga useful kalo restore-nya ga jalan. Test:

# Download backup terbaru
aws s3 cp s3://my-backups/agent-latest.tar.gz /tmp/

# Extract di tmp folder
mkdir -p /tmp/restore-test
tar -xzf /tmp/agent-latest.tar.gz -C /tmp/restore-test

# Verify isi
ls -la /tmp/restore-test/data/
cat /tmp/restore-test/data/memory.json | jq '.notes | length'

# Cleanup
rm -rf /tmp/restore-test /tmp/agent-latest.tar.gz

Kalo ada error, fix backup pipeline.

4. Disaster recovery drill¶

Skenario: VPS lo di-suspend, harus migrate ke VPS baru dalam 1 jam. Apakah lo bisa?

Step: 1. Sign up VPS baru 2. SSH ke VPS baru 3. Clone repo: git clone https://github.com/user/agent.git 4. Setup env: python3 -m venv venv && source venv/bin/activate && pip install -r requirements.txt 5. Restore data: aws s3 cp s3://my-backups/agent-latest.tar.gz . && tar -xzf agent-latest.tar.gz 6. Setup systemd 7. Start service 8. Update DNS / Telegram webhook kalo perlu

Kalo lo ga bisa dalam 1 jam, improve documentation / scripts.

Yearly¶

1. Major upgrades¶

Ubuntu LTS upgrade (22.04 → 24.04 → 26.04 saat available)
Python major version (3.12 → 3.13)
Library major versions

Test dulu di staging / cloned VPS.

2. SOUL.md philosophical review¶

SOUL.md udah 1 tahun aktif. Pertanyaan:

Apakah filosofi inti masih sesuai? (mungkin lo udah mature, butuh adjust)
Apakah identity masih relevan? (mungkin lo mau rename, ganti tone)
Apakah boundaries cukup ketat / cukup longgar?
Apakah autonomy levels masih sesuai?

Rewrite kalo perlu. Bot lo mungkin udah evolve dari original use case.

3. Stack evaluation¶

Apakah LLM model lo masih SOTA?
Apakah Python library lo masih maintained?
Apakah Telegram masih platform yang lo pake? (mungkin Discord lebih cocok sekarang)

Switching cost vs benefit. Kalo benefit > cost, plan migration.

Update best practices¶

Always backup before update¶

# Sebelum git pull / pip upgrade / etc
tar -czf ~/agent-pre-update-$(date +%Y%m%d).tar.gz ~/agent

Update sequence¶

# 1. Save state
sudo systemctl stop kai-bot

# 2. Backup
tar -czf ~/agent-backup-$(date +%Y%m%d).tar.gz ~/agent/data ~/agent/credentials

# 3. Update code
cd ~/agent
git pull

# 4. Update deps
source venv/bin/activate
pip install -r requirements.txt --upgrade

# 5. Run tests (kalo ada)
# pytest

# 6. Start service
sudo systemctl start kai-bot

# 7. Verify
sleep 5
sudo systemctl is-active kai-bot
sudo journalctl -u kai-bot -n 20 --no-pager

Rollback if needed¶

# Stop current
sudo systemctl stop kai-bot

# Restore from backup
cd ~
tar -xzf agent-backup-20260514.tar.gz

# Revert code
cd ~/agent
git log --oneline -5
git checkout <previous-commit>

# Start
sudo systemctl start kai-bot

Anti-patterns¶

❌ Update tanpa backup¶

git pull  # ← bisa break, ga ada rollback

Always backup first.

❌ Update tanpa test¶

git pull && sudo systemctl restart kai-bot
# bot crash, lo baru tau setelah user complain

Test manual dulu (run python main.py foreground), terus systemd.

❌ Lupa monitoring setelah update¶

Bot bisa silent fail (jalan tapi ga respond bener). Check beberapa kali setelah update.

❌ Edit SOUL.md tanpa version¶

nano ~/agent/SOUL.md
# kalo broken, ga ada way to rollback

Pakai git untuk SOUL.md:

cd ~/agent
git add SOUL.md
git commit -m "soul: tweak communication tone"

Rollback gampang:

git log SOUL.md
git checkout <commit> -- SOUL.md

Monitoring dashboard sederhana¶

Bikin command yang summary status:

cat > ~/agent/status.sh << 'EOF'
#!/bin/bash
echo "=== Kai Agent Status ==="
echo ""
echo "Service:"
systemctl is-active kai-bot
systemctl is-enabled kai-bot
echo ""
echo "Memory:"
free -h | head -2
echo ""
echo "Disk:"
df -h / | tail -1
echo ""
echo "Recent errors (24h):"
journalctl -u kai-bot -p err --since "24 hours ago" --no-pager | wc -l
echo ""
echo "Uptime:"
uptime
EOF

chmod +x ~/agent/status.sh
~/agent/status.sh

Atau make available via Telegram command:

async def status_cmd(update, ctx):
    if not is_owner(update.effective_user.id):
        return
    result = await run_shell("/home/ubuntu/agent/status.sh")
    await update.message.reply_text(f"{result[:3000]}")