Skip to main content

​
Overview

The Analytics Page provides visual insights into agent costs and performance. Track spending trends, identify expensive operations, and optimize your agent fleet.

​
Time window selector

At the top, choose time range:
  • Last 24 hours
  • Last 7 days
  • Last 30 days
  • All time
All charts and stats update based on selected window.

​
Cost dashboard

​
30-day consumption chart

Bar chart showing daily costs:
  • X-axis: Dates (last 30 days)
  • Y-axis: Cost (USD)
  • Bars: Daily spending
  • Hover: See exact cost for each day
Visual features:
  • Color coding:
    • 🟢 Green - Below average
    • 🟡 Yellow - Average
    • 🔴 Red - Above average
  • Trend line: Overlaid to show direction
  • Annotations: Mark significant events (deployments, incidents)
Insights:
  • Spot cost spikes
  • Identify trends (increasing/decreasing)
  • Correlate with deployments or changes
  • Weekly patterns (weekdays vs weekends)

​
Total consumption

Large stat card showing:
  • Total Cost - Sum for selected time window
  • Comparison - vs previous period
    • ⬆️ 15% increase
    • ⬇️ 10% decrease
    • ➡️ No change (< 2%)
  • Projection - Estimated monthly cost (if trend continues)
Example: 
Total Consumption (Last 30 days)
$567.89
⬆️ 15% vs previous 30 days
Projected monthly: ~$580

​
Daily breakdown table

Table showing each day with:
  • Date
  • Runs - Count of runs
  • Actions - Count of actions
  • Cost - Total USD
  • Avg Cost per Run - Mean cost
  • Avg Cost per Action - Mean cost
Sorting:
  • Click any column header to sort
  • Default: Date (descending)
Export:
  • Download as CSV for analysis in Excel/BI tools

​
Performance dashboard

​
Average metrics

Three stat cards: 1. Avg Cost per Action 
$0.0123
⬇️ 5% vs previous period
Lower is better - shows efficiency improving. 2. Avg Duration per Action 
523ms
➡️ No change vs previous period
Lower is better - shows performance. 3. Actions per Run 
37.2
⬆️ 8% vs previous period
Context-dependent - more actions = more work done OR inefficiency.

​
Cost by action type

Pie chart showing distribution:
  • Each slice = one action type
  • Size = proportion of total cost
  • Hover for exact amount
  • Click to filter ledger by that action
Example breakdown:
  • 🔵 call_llm: $234.56 (41%)
  • 🟢 search_web: $123.45 (22%)
  • 🟡 fetch_data: $89.12 (16%)
  • 🟠 analyze_text: $67.89 (12%)
  • Others: $52.87 (9%)
Insights:
  • Identify top cost drivers
  • Focus optimization on largest slices
  • Track changes over time

​
Cost by agent

Bar chart (horizontal) showing:
  • Agent names (Y-axis)
  • Cost (X-axis)
  • Sorted by cost (descending)
  • Color-coded by relative cost
Use cases:
  • Identify most expensive agents
  • Compare agent efficiency
  • Budget allocation by agent

​
Actions over time

Line chart showing:
  • X-axis: Time (hours, days, or months based on window)
  • Y-axis: Action count
  • Multiple lines: Different action types
Insights:
  • Usage patterns (peak times)
  • Growth trends
  • Action mix changes

​
Cost optimization recommendations

AI-powered suggestions based on your data:

​
High-cost actions

💡 Recommendation: Optimize LLM calls

Finding: "call_llm" accounts for 45% of costs ($234.56)

Suggestions:
• Reduce max_tokens from 1000 to 500 (potential savings: ~$100/month)
• Use gpt-4o-mini for simple queries (potential savings: ~$80/month)
• Implement response caching (potential savings: ~$50/month)

Total potential savings: $230/month (40%)

​
Inefficient patterns

⚠️ Warning: Redundant API calls detected

Finding: Agent "trading-bot" calls "fetch_price" avg 15 times per run

Suggestion: Cache price data for 5 minutes to reduce API calls

Potential savings: $45/month

​
Rate limiting opportunities

💡 Recommendation: Add rate limiting

Finding: "search_web" called 500+ times/day

Suggestion: Rate limit to 100 calls/hour to prevent runaway costs

Risk reduction: Prevents potential $1000+ cost incidents

​
Filters and segmentation

​
Filter by agent

Select specific agents to analyze:
  • Compare costs across agents
  • Focus on one agent’s performance
  • Identify underutilized agents

​
Filter by action type

Select specific action types:
  • Analyze LLM costs separately
  • Track API call costs
  • Compare tool usage

​
Group by

Change aggregation:
  • By day - Daily breakdown
  • By hour - Hourly breakdown (for 24h window)
  • By week - Weekly breakdown (for 30d+ window)
  • By month - Monthly breakdown (for all time)

​
Budget Forecasting & Alerts

Agent Sentinel provides GCP-style predictive analytics to help you stay within budget and optimize spending.

​
Budget forecast widget

Set a budget amount to see real-time forecasting: Metrics displayed:
  • Current Spend - What you’ve spent so far this period
  • Burn Rate - Spending velocity relative to time elapsed
    • 1.0 = on track
    • > 1.0 = overspending
    • < 1.0 = underspending
  • Projected Total - Estimated end-of-period spend
  • Days Remaining - Time left in budget period
Alert severity levels:
  • 🟢 Info - On track (< 80% projected)
  • 🟡 Warning - Approaching limit (80-100% projected)
  • 🟠 Critical - Will exceed budget (100%+ projected)
  • 🔴 Exceeded - Already over budget
Example: 
Current Spend: $234.56 (46.9% of budget)
Burn Rate: 1.15x (overspending)
Projected Total: $575.23 of $500.00 budget
Days Remaining: 18 days

​
Smart recommendations

The forecast widget provides actionable recommendations:
  • “⚠️ Budget will be exceeded by $75.23 (15.0%)”
  • ”🔥 High burn rate (1.15x) - reduce agent frequency or switch to cheaper models”
  • ”💡 18 days remaining - time to optimize or increase budget”
  • ”💰 Switch expensive actions to cheaper models (GPT-4 → GPT-3.5 saves 90%)”
  • “⚙️ Implement rate limiting or approval gates for high-cost actions”

​
Budget exhaustion warning

When projected to exceed budget, see:
  • Projected date exceeded - When budget will run out
  • Days until exceeded - Countdown to budget exhaustion
  • Visual alert - Warning banner with urgency indicator

​
Spend trend chart

Interactive chart showing:
  • Historical spend - Actual daily costs (solid line)
  • Projected spend - Forecasted future costs (dashed line)
  • Budget line - Budget limit reference (dotted line)
Features:
  • Hover for exact daily amounts
  • See cumulative spend over time
  • Visual indication when projected to exceed budget
  • Automatic period selection (defaults to current month)

​
Budget alerts

Real-time alerts appear when:
  • Budget reaches 80% threshold
  • Projected to exceed budget
  • Budget already exceeded
  • Burn rate exceeds 1.2x
Alert badges show in the UI with:
  • Severity level (warning/critical/exceeded)
  • Current vs projected spend
  • Days until exhaustion
  • Quick recommendations

​
How forecasting works

The platform uses linear projection based on your current burn rate: 
daily_rate = current_spend / days_elapsed
projected_spend = daily_rate × total_days_in_period
Confidence increases with more data:
  • First few days: ~50% confidence
  • Mid-period: ~70% confidence
  • Late period: ~90% confidence

​
Setting budget periods

Budgets can be configured for:
  • Current month (default)
  • Custom date range
  • Rolling 30 days
API example: 
GET /api/v1/budgets/forecast?budget=500.0&period_start=2025-01-01&period_end=2025-01-31

​
Budget alert emails

Configure email alerts (coming soon):
  • Threshold alerts (at 80%, 90%, 100%)
  • Daily budget summaries
  • Weekly forecast updates
  • Exceeded budget notifications

​
Common workflows

​
Monthly cost review

  1. Select “Last 30 days” time window
  2. Review total consumption
  3. Check 30-day chart for spikes
  4. Investigate anomalies:
    • Click spike day in chart
    • Filter Activity Ledger by that date
    • Review expensive actions
  5. Check cost by action type pie chart
  6. Read optimization recommendations
  7. Implement suggested optimizations

​
Agent cost comparison

  1. View “Cost by Agent” chart
  2. Identify top 3 most expensive agents
  3. For each agent:
    • Click bar to navigate to agent details
    • Review “Top Actions”
    • Check “Cost Breakdown”
    • Look for optimization opportunities
  4. Compare with similar agents:
    • Why is Agent A 3x more expensive than Agent B?
    • Different usage patterns?
    • Different action mix?
    • Bugs or inefficiencies?

​
Track optimization impact

  1. Note current “Avg Cost per Action” before optimization
  2. Implement optimization (e.g., use cheaper model)
  3. Wait 24 hours
  4. Return to Analytics page
  5. Compare “Avg Cost per Action”
  6. Calculate savings: 
    Daily savings = (Old avg - New avg) × Daily actions
Monthly projection = Daily savings × 30
  7. Document successful optimizations

​
Budget planning

  1. Select “Last 30 days”
  2. Note “Total Consumption”
  3. Check “Projected monthly” estimate
  4. Review “Cost by Agent” to allocate budgets
  5. Set per-agent budgets in Settings → Policies
  6. Monitor adherence in future periods

​
Identify usage patterns

  1. View “Actions over time” chart
  2. Look for patterns:
    • Peak usage times (optimize for these)
    • Weekend vs weekday differences
    • Growth trends (plan capacity)
    • Seasonal variations
  3. Adjust:
    • Scale infrastructure for peaks
    • Budget for expected growth
    • Implement rate limiting during peaks

​
Export and reporting

​
Export charts

Click Export on any chart:
  • PNG image - For presentations
  • CSV data - For analysis
  • PDF report - For stakeholders

​
Scheduled reports (future)

Coming soon: Automated reports:
  • Daily cost summary (email)
  • Weekly trends analysis (email + Slack)
  • Monthly executive summary (PDF)

​
Best practices

Review analytics weekly: Set a recurring calendar reminder to review analytics every Monday.
Act on recommendations: The AI suggestions are based on real data - implementing them can save 20-40% on costs.
Spikes need investigation: Any day with > 2x average cost should be investigated immediately to prevent runaway costs.
Compare periods: Always use the period comparison (vs previous 30 days) to understand if you’re improving or regressing.
Focus on top 20%: Usually 20% of actions account for 80% of costs - optimize these first.

​
Keyboard shortcuts

  • 1 - Switch to 24h view
  • 7 - Switch to 7d view
  • 3 - Switch to 30d view
  • a - Switch to all time view
  • e - Export current view

​
Troubleshooting

​
”Charts not loading”

  • Check time window selector
  • Verify you have runs in selected period
  • Refresh page
  • Try shorter time window

​
”No data for selected period”

  • Extend time window (try “All time”)
  • Check if any runs exist (Runs page)
  • Verify agent is sending data

​
”Cost seems wrong”

  • Check if multiple agents are included
  • Verify time window is correct
  • Review Activity Ledger for source data
  • Check for currency/timezone mismatches

​
”Recommendations not showing”

  • Recommendations appear after 7+ days of data
  • Need minimum 100 actions for analysis
  • Some patterns need 30 days to detect

​
See also