Performance Issues

Typical bottlenecks

• Slow model responses
• Overlapping scheduled tasks
• Large command outputs
• Long conversation context

Optimization checklist

1. Use faster models for high-frequency interactions.

2. Reduce task frequency to avoid overlaps.

3. Keep command output focused by narrowing the command scope and avoiding large noisy logs.

4. Tune context compaction:

{
  "context": {
    "messages": {
      "maxInputTokensApprox": 128000,
      "compactRatio": 0.5
    }
  }
}

Observe performance signals

grep -Ei "timeout|slow|context|compact" .downcity/logs/$(date +%F).jsonl