Cost Optimization
Strategies for reducing agent costs without sacrificing capability.
Broker Routing
The biggest cost saver is the Broker Routing engine. Instead of sending every task to an LLM, the broker tries cheaper options first:
Task → Skill Match (free) → WASM Handler ($0.001) → LLM ($0.03+)
Enable broker routing:
broker:
enabled: true
routing:
preferSkill: true
llmAsLastResort: true
Cost Savings Breakdown
| Route | Avg Cost | Speed |
|---|---|---|
| Skill match | $0.00 | <1ms |
| WASM handler | ~$0.001 | ~10ms |
| LLM fallback | ~$0.03 | ~2s |
With broker routing enabled, typical projects see 40-60% cost reduction.
Caching
Enable response caching to avoid duplicate LLM calls:
broker:
cache:
enabled: true
ttl: 3600 # cache for 1 hour
Right-Sizing Budgets
Use the dashboard to analyze actual spending patterns:
naos dashboard --open
Navigate to the Cost page to see:
- Per-agent spending over time
- Peak usage periods
- Budget utilization rates
Tips
- Start with generous budgets and tighten based on actual usage
- Use skill handlers for repetitive tasks (summarize, classify, extract)
- Enable caching for idempotent operations
- Monitor the dashboard weekly to identify optimization opportunities
- Use throttle instead of pause to maintain availability at reduced capacity