Analytics
Dashboards
Build custom dashboards to visualize AI application metrics, performance, and quality
Dashboards
Brokle dashboards provide visual insights into your AI application. Use built-in dashboards or create custom views for your specific needs.
Built-in Dashboards
Overview Dashboard
The main dashboard showing key metrics at a glance:
┌─────────────────────────────────────────────────────────────────┐
│ Overview Dashboard │
├──────────────────┬──────────────────┬──────────────────┬────────┤
│ Total Traces │ Token Usage │ Total Cost │ Errors │
│ 45,230 │ 12.5M │ $1,234 │ 0.8% │
│ ↑ 12% │ ↑ 8% │ ↓ 5% │ ↓ │
├──────────────────┴──────────────────┴──────────────────┴────────┤
│ │
│ ████████████████████████ │
│ ████████████████████ Traces over Time │
│ ██████████████████████████ │
│ ████████████████████ │
│ │
├─────────────────────────────┬────────────────────────────────────┤
│ Model Usage │ Top Features │
│ GPT-4o: 60% │ Chat: 45% │
│ GPT-4o-mini: 35% │ Search: 30% │
│ Claude: 5% │ Summary: 25% │
└─────────────────────────────┴────────────────────────────────────┘Performance Dashboard
Detailed latency and reliability metrics:
┌─────────────────────────────────────────────────────────────────┐
│ Performance Dashboard │
├─────────────────────────────────────────────────────────────────┤
│ Latency (P50/P95/P99) │
│ ├─ Overall: 1.2s / 2.8s / 4.5s │
│ ├─ GPT-4o: 1.5s / 3.2s / 5.0s │
│ └─ GPT-4o-mini: 0.8s / 1.5s / 2.2s │
│ │
│ Time to First Token │
│ ├─ P50: 180ms │
│ └─ P95: 450ms │
│ │
│ Error Breakdown │
│ ├─ Rate Limits: 0.5% │
│ ├─ Timeouts: 0.2% │
│ └─ API Errors: 0.1% │
└─────────────────────────────────────────────────────────────────┘Quality Dashboard
Evaluation scores and feedback metrics:
┌─────────────────────────────────────────────────────────────────┐
│ Quality Dashboard │
├─────────────────────────────────────────────────────────────────┤
│ Evaluation Scores (7-day avg) │
│ ├─ Relevance: ████████░░ 0.85 │
│ ├─ Helpfulness: ███████░░░ 0.78 │
│ └─ Accuracy: █████████░ 0.92 │
│ │
│ User Feedback │
│ ├─ Satisfaction: 89% 👍 │
│ ├─ Response rate: 4.2% │
│ └─ Top issues: "Too verbose", "Missing details" │
│ │
│ Quality Trend │
│ [Chart showing score changes over time] │
└─────────────────────────────────────────────────────────────────┘Creating Custom Dashboards
Navigate to Dashboard Builder
Go to Analytics → Dashboards → Create Dashboard
Add Widgets
Choose from available widget types:
- Metric Card: Single key metric
- Line Chart: Trends over time
- Bar Chart: Comparisons
- Pie Chart: Distribution
- Table: Detailed data
- Heatmap: Time-based patterns
Configure Data Sources
Select metrics and dimensions for each widget:
widget:
type: line_chart
metric: trace_count
group_by: model
time_range: 7d
filters:
- environment = "production"Save and Share
Name your dashboard and set sharing permissions.
Widget Types
Metric Cards
Display key numbers prominently:
┌────────────────────┐
│ Today's Cost │
│ $45.23 │
│ ↑ 12% vs avg │
└────────────────────┘Configuration:
widget:
type: metric
metric: cost
aggregation: sum
comparison: "vs_avg_7d"Time Series Charts
Track trends over time:
# Via API
chart_data = client.analytics.get_time_series(
metrics=["trace_count", "token_count"],
start_time=datetime.now() - timedelta(days=7),
group_by="day"
)Comparison Charts
Compare dimensions side-by-side:
widget:
type: bar_chart
metric: latency_p95
group_by: model
filters:
- time_range = "24h"Distribution Charts
Show breakdowns:
widget:
type: pie_chart
metric: cost
group_by: modelData Tables
Detailed tabular data:
widget:
type: table
columns:
- trace_id
- model
- latency_ms
- cost
- quality_score
sort_by: cost
sort_order: desc
limit: 100Heatmaps
Visualize patterns across time:
widget:
type: heatmap
metric: trace_count
x_axis: hour_of_day # 0-23
y_axis: day_of_week # Mon-SunFilters and Drill-Down
Global Filters
Apply filters to all widgets:
┌─────────────────────────────────────────────────────────────────┐
│ Filters: [Environment: Production ▼] [Model: All ▼] [7 days ▼] │
├─────────────────────────────────────────────────────────────────┤
│ │
│ [Dashboard content filtered accordingly] │
│ │
└─────────────────────────────────────────────────────────────────┘Widget-Level Filters
Configure filters per widget:
widget:
type: line_chart
metric: cost
filters:
- model = "gpt-4o"
- environment = "production"
- quality_score < 0.7Click-to-Drill
Click on chart elements to drill down:
Click on "GPT-4o" bar
→ Opens filtered view showing only GPT-4o traces
→ Adds filter: model = "gpt-4o"Dashboard Templates
Engineering Template
name: Engineering Dashboard
widgets:
- type: metric
metric: error_rate
position: [0, 0]
- type: line_chart
metric: latency_p95
group_by: model
position: [1, 0]
- type: table
columns: [trace_id, error_type, timestamp]
filters: [status = "error"]
position: [0, 1]Product Template
name: Product Dashboard
widgets:
- type: metric
metric: daily_active_users
position: [0, 0]
- type: pie_chart
metric: trace_count
group_by: feature
position: [1, 0]
- type: line_chart
metric: satisfaction_rate
position: [0, 1]Finance Template
name: Finance Dashboard
widgets:
- type: metric
metric: monthly_cost
position: [0, 0]
- type: bar_chart
metric: cost
group_by: team
position: [1, 0]
- type: line_chart
metric: cost_per_user
position: [0, 1]Sharing & Access
Share Options
| Option | Description |
|---|---|
| Private | Only you can view |
| Team | All team members can view |
| Project | All project members can view |
| Public Link | Anyone with link can view (read-only) |
Embedding
Embed dashboards in other tools:
<!-- Embed via iframe -->
<iframe
src="https://app.brokle.com/embed/dashboard/abc123?token=xyz"
width="100%"
height="600"
frameborder="0"
></iframe>Scheduled Reports
Send dashboard snapshots via email:
# Configure scheduled report
client.dashboards.schedule_report(
dashboard_id="dash_123",
schedule="weekly",
recipients=["team@company.com"],
format="pdf"
)Real-Time Updates
Live Mode
Enable real-time updates for active monitoring:
┌─────────────────────────────────────────────────────────────────┐
│ 🔴 LIVE MODE - Refreshing every 30 seconds │
├─────────────────────────────────────────────────────────────────┤
│ │
│ [Widgets auto-refresh with latest data] │
│ │
└─────────────────────────────────────────────────────────────────┘Configure refresh intervals:
- 30 seconds (for incident response)
- 1 minute
- 5 minutes
- Manual refresh only
Streaming Metrics
For critical monitoring, use streaming:
# Stream real-time metrics
async for metric in client.analytics.stream_metrics(
metrics=["trace_count", "error_rate"],
interval_seconds=10
):
print(f"Traces: {metric.trace_count}, Errors: {metric.error_rate}%")Best Practices
1. Purpose-Driven Dashboards
Create focused dashboards for specific audiences:
# Bad: One dashboard for everything
# Good: Separate dashboards per audience
engineering_dashboard:
focus: performance, errors, latency
product_dashboard:
focus: usage, features, engagement
executive_dashboard:
focus: costs, ROI, trends2. Use Consistent Time Ranges
Align time ranges across related widgets for accurate comparison:
# All widgets should use the same time range
dashboard:
default_time_range: "7d"
widgets:
- time_range: inherit # Uses dashboard default
- time_range: inherit3. Add Context
Include comparison metrics for context:
widget:
type: metric
metric: daily_cost
comparison: "vs_same_day_last_week" # Shows +15% context4. Set Up Alerts from Dashboards
Link alerts to dashboard thresholds:
# Create alert from dashboard metric
client.alerts.create_from_widget(
dashboard_id="dash_123",
widget_id="widget_456",
threshold=100,
operator=">"
)Dashboards are automatically saved. Your layout and filters are preserved between sessions.
Next Steps
- Cost Tracking - Deep dive into cost analytics
- Alerts - Set up monitoring alerts
- Tracing - Investigate individual traces