token-cost-monitor

# Token Cost Monitor Monitor OpenClaw API costs in real-time, set budget alerts, and optimize model spending. ## Why This Matters **Real user stories:** - One user spent $18.75 overnight on heartbeat checks asking "Is it daytime yet?" - Federico Viticci spent $3,600 in a single month - Regular users report $200/day bills from misconfigured automations ## Quick Commands **Check current session costs:** ```bash # View token usage and estimated cost openclaw /status openclaw /usage ``` **Set up cost alerts:** ```bash # Alert when daily spend exceeds $10 echo "Alert if daily_cost > $10" # Alert when monthly spend exceeds $100 echo "Alert if monthly_cost > $100" ``` ## Cost Optimization Strategies ### 1. Model Routing **Use cheaper models for simple tasks:** - Claude Haiku: $0.25/1M input tokens (fast, cheap) - Claude Sonnet: $3/1M input tokens (balanced) - Claude Opus: $15/1M input tokens (expensive, powerful) **Routing rules:** ``` Simple queries (weather, facts) → Haiku Complex analysis → Sonnet Creative writing → Opus (only if needed) ``` ### 2. Context Management **Reduce token waste:** - Clear old conversation context regularly - Don't store tool outputs in context - Use concise system prompts - Limit heartbeat frequency ### 3. Heartbeat Optimization **Common mistake:** ``` ❌ Every 30 min: "Is it daytime yet?" → $18.75/night ✅ Every 6 hours: Check emails + calendar → $2/day ``` **Best practices:** - Combine multiple checks into one heartbeat - Use longer intervals (4-6 hours) - Skip nighttime heartbeats (23:00-08:00) - Track heartbeat costs separately ## Cost Tracking Dashboard ### Daily Cost Breakdown | Category | Tokens | Cost | |----------|--------|------| | Heartbeat checks | 50,000 | $0.50 | | Web searches | 100,000 | $1.00 | | Code generation | 200,000 | $2.00 | | Document analysis | 150,000 | $1.50 | | **Total** | **500,000** | **$5.00/day** | ### Monthly Projection ``` Daily average: $5.00 Monthly projection: $150.00 Budget limit: $100.00 ⚠️ Warning: Will exceed budget by $50 ``` ## Budget Alerts Setup ### Tier 1: Light User ($10-30/month) ``` Daily limit: $1.00 Alert at: $0.80/day Actions: Reduce heartbeat frequency, use Haiku more ``` ### Tier 2: Regular User ($40-80/month) ``` Daily limit: $2.50 Alert at: $2.00/day Actions: Optimize context, review model routing ``` ### Tier 3: Power User ($100-500/month) ``` Daily limit: $10.00 Alert at: $8.00/day Actions: Audit all automations, implement strict routing ``` ## Cost Reduction Checklist ### Immediate Actions (save 30-50%) - [ ] Reduce heartbeat frequency to 4-6 hours - [ ] Route simple queries to Haiku - [ ] Clear context after completed tasks - [ ] Disable unnecessary automations ### Medium-term (save 50-70%) - [ ] Implement model routing rules - [ ] Optimize system prompts - [ ] Batch similar tasks together - [ ] Use caching for repeated queries ### Long-term (save 70-90%) - [ ] Build cost-aware agent behaviors - [ ] Implement token budgets per task - [ ] Use local models for simple tasks - [ ] Optimize tool output storage ## Monitoring Commands ### Real-time Cost Check ```bash # Get current session stats openclaw /status # Get detailed usage breakdown openclaw /usage --detailed ``` ### Historical Analysis ```bash # Last 7 days cost trend echo "Analyze spending trend for past 7 days" # Identify most expensive tasks echo "What are my top 5 most expensive task types?" ``` ### Anomaly Detection ```bash # Alert if spending is 2x normal echo "Alert if hourly_cost > 2 * average_hourly_cost" ``` ## Integration Examples ### Slack/Discord Alerts ```bash # Send daily cost summary to Slack curl -X POST https://hooks.slack.com/services/YOUR/WEBHOOK/URL \ -d '{"text": "Daily OpenClaw cost: $5.00 (Budget: $10.00)"}' ``` ### Email Reports ```bash # Send weekly cost report via email echo "Weekly cost report: $35.00 total, $5.00/day average" | mail -s "OpenClaw Cost Report" user@example.com ``` ## Cost Optimization Scripts ### Auto-pause Expensive Sessions ```bash #!/bin/bash # Pause session if daily cost exceeds limit DAILY_LIMIT=10 CURRENT_COST=$(get_current_cost) if (( $(echo "$CURRENT_COST > $DAILY_LIMIT" | bc -l) )); then echo "⚠️ Daily cost ($CURRENT_COST) exceeds limit ($DAILY_LIMIT)" echo "Pausing non-essential automations..." # Add pause logic here fi ``` ### Model Router ```bash #!/bin/bash # Route queries to appropriate model based on complexity QUERY="$1" WORD_COUNT=$(echo "$QUERY" | wc -w) if [ $WORD_COUNT -lt 20 ]; then echo "Using Haiku (simple query)" MODEL="haiku" elif [ $WORD_COUNT -lt 100 ]; then echo "Using Sonnet (medium complexity)" MODEL="sonnet" else echo "Using Opus (complex analysis)" MODEL="opus" fi ``` ## Common Cost Pitfalls ### ❌ Mistake: Infinite Loops ``` Agent keeps calling itself → 1000s of API calls → $100+ in hours ``` **Fix:** Implement loop detection and limits ### ❌ Mistake: Storing Tool Outputs ``` Every tool output saved to context → Context grows → Each call costs more ``` **Fix:** Only store essential information ### ❌ Mistake: Wrong Model for Task ``` Using Opus for weather queries → 60x more expensive than needed ``` **Fix:** Implement model routing ### ❌ Mistake: Heartbeat Misconfiguration ``` Heartbeat every 5 minutes → 288 calls/day → $50+/day ``` **Fix:** Use 4-6 hour intervals ## Pricing Reference (2026) ### Anthropic Claude | Model | Input | Output | Best For | |-------|-------|--------|----------| | Haiku | $0.25/1M | $1.25/1M | Simple queries, facts | | Sonnet | $3/1M | $15/1M | General tasks | | Opus | $15/1M | $75/1M | Complex analysis | ### OpenAI GPT | Model | Input | Output | Best For | |-------|-------|--------|----------| | GPT-4o Mini | $0.15/1M | $0.60/1M | Simple queries | | GPT-4o | $2.5/1M | $10/1M | General tasks | | o1 | $15/1M | $60/1M | Complex reasoning | ### Google Gemini | Model | Input | Output | Best For | |-------|-------|--------|----------| | Flash | $0.075/1M | $0.30/1M | Simple queries | | Pro | $1.25/1M | $5/1M | General tasks | | Ultra | $7.5/1M | $30/1M | Complex analysis | ## ROI Calculator **Calculate if automation is worth it:** ``` Manual task time: 30 minutes Your hourly rate: $50/hour Manual cost: $25 per task Automation cost: - API calls: $0.50 - Setup time (amortized): $5.00 - Total: $5.50 per task Savings: $25 - $5.50 = $19.50 per task ROI: 355% ``` ## Updates - 2026-03-27: Initial release - Pricing as of March 2026 - Based on real user cost data