How I Optimized My AI Agent Setup for Daily Use

The Problem: A Messy Config Nobody Touched

Saya punya VPS yang jalan 24/7 — Hermes, gateway utama untuk semua operasi. Cada hari dipakai untuk chat, automation, monitoring, dan ngurusin infrastructure.

Masalahnya: config belum pernah dioptimalkan. Pakai fallback provider yang broken, logging verbose banget, session nggak pernah diprune, memory disabled, dan default model bukan yang terbaik untuk daily use.

Ini dokumentasi proses saya audit dan optimize setup Hermes dari awal — langkah per langkah, biar kamu bisa ikutin atau adaptasi.

Step 1: Audit — Cek Semua Provider

Sebelum rubah apapun, saya cek satu per satu provider mana yang actually work.

curl -X POST "https://YOUR_PROVIDER_ENDPOINT/v1/messages"
-H "x-api-key: YOUR_API_KEY"
-H "anthropic-version: 2023-06-01"
-d '{"model":"your-best-model","messages":{"role":"user","content":"ping"},"max_tokens":5}'

curl -X POST "https://YOUR_FALLBACK_ENDPOINT/v1/chat/completions"
-H "Authorization: Bearer YOUR_API_KEY"
-d '{"model":"your-fallback-model","messages":{"role":"user","content":"ping"},"max_tokens":5}'

Hasil audit:

Lesson: Jangan trust fallback yang nggak kamu test langsung.

Step 2: Audit Config Sekarang

cat ~/.hermes/config.yaml

Yang saya dapat:

model:
  default: your-fallback-model  # ← Salah, harusnya best model
  provider: custom:your-fallback-provider
  cache_refresh_interval: 0
fallback_providers:
  - provider: custom:your-fallback-provider
    model: your-cheap-fallback
  - provider: custom:broken-provider  # ← Sudah broke, jadiin beban
    model: your-opus-model

Masalah:

1. Wrong primary model — pakai fallback padahal best Model lebih baik
2. Broken fallback — 401 error, nggak pernah di-check
3. No TTS config — pakai default yang lebih lambat
4. Display default — nggak ada streaming, timestamps, atau compact mode
5. Compression off — context panjang = mahal dan lambat
6. Session retention forever — database bengkak
7. Memory disabled — hilang konteks antar session
8. Logging verbose — journal penuh noise

Step 3: Optimize

3.1 Model — Pakai yang Terbaik

model:
  default: your-best-model
  provider: custom:your-best-provider
  cache_refresh_interval: 0
fallback_providers:
  - provider: custom:your-fallback-provider
    model: your-cheap-fallback

Kenapa best model? Reasoning yang lebih dalam. Untuk daily use — drafting, analysis, coding, strategy — ini worth every cent.

Remove broken provider? Ya. Provider yang return 401 nggak berguna — justru bikin system lambat karena coba-coba sebelum timeout.

3.2 TTS — Pakai Edge

tts:
  provider: edge

Edge TTS — fast, free, dan suara cukup natural. Nggak perlu API key.

3.3 Display — Buat CLI Bersih

display:
  show_reasoning: true
  streaming: true
  compact: true
  interface: cli
  timestamps: true

3.4 Compression — Aggressive Context Management

compression:
  enabled: true
  threshold: 0.35

Threshold 35% — kalau conversation mencapai 35% dari limit, compression kick in. Mencegah bill shock.

3.5 Privacy — Redact PII

privacy:
  redact_pii: true

3.6 Sessions — Auto Prune

sessions:
  auto_prune: true
  retention_days: 30

3.7 Memory — Cross-Session Context

memory:
  memory_enabled: true
  user_profile_enabled: true

3.8 Telegram — Streaming

platforms:
  telegram:
    streaming: true

3.9 Logging — Kurangi Noise

logging:
  level: WARNING

Step 4: Apply dan Restart

# Backup dulu
cp ~/.hermes/config.yaml ~/.hermes/config.yaml.bak-$(date +%Y-%m-%d)

# Restart gateway
kill PID_GATEWAY
systemctl --user start hermes-gateway

# Verify
curl http://localhost:8642/api/status

Results

Config Lengkap

model:
  default: your-best-model
  provider: custom:your-best-provider
  cache_refresh_interval: 0
fallback_providers:
  - provider: custom:your-fallback-provider
    model: your-cheap-fallback

tts:
  provider: edge

display:
  show_reasoning: true
  streaming: true
  compact: true
  interface: cli
  timestamps: true

compression:
  enabled: true
  threshold: 0.35

privacy:
  redact_pii: true

sessions:
  auto_prune: true
  retention_days: 30

memory:
  memory_enabled: true
  user_profile_enabled: true

platforms:
  telegram:
    streaming: true

logging:
  level: WARNING

Tips Tambahan

Test provider baru sebelum ditambahkan ke fallback:

curl -X POST "https://YOUR_PROVIDER_ENDPOINT/v1/chat/completions" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -d '{"model":"test-model","messages":[{"role":"user","content":"ping"}],"max_tokens":5}'

Backup sebelum ubah:

cp ~/.hermes/config.yaml ~/.hermes/config.yaml.bak-$(date +%Y-%m-%d)

Kesimpulan

Yang paling impactfull:

1. Best reasoning model sebagai primary — quality jauh lebih baik
2. Memory enabled — conversation context carry over
3. Auto-prune sessions — database nggak bengkak
4. Streaming + compact CLI — lebih nyaman dipakai daily

test everything, backup everything, monitor everything.