Observability

Bastio provides complete observability for your AI stack. Every request is captured as a trace, conversations are grouped into sessions, and analytics dashboards show you patterns across your usage.

Traces

Every request that flows through Bastio is captured as a trace with full context.

What Gets Captured

Each trace includes:

Example Trace

{
  "trace_id": "tr_7kL9mN2pQr",
  "status": "success",
  "latency_ms": 847,
  "tokens": {
    "input": 1234,
    "output": 567
  },
  "cost_usd": 0.0023,
  "session_id": "sess_abc123",
  "model": "gpt-4o",
  "security": {
    "risk_score": 0.1,
    "threats": [],
    "blocked": false
  },
  "created_at": "2025-01-13T10:30:00Z"
}

Viewing Traces

Access traces in your dashboard under Observe > Traces:

• Real-time updates: Dashboard polls every 5 seconds for new traces
• Filtering: Filter by status, model, time range, or session
• Search: Find specific traces by ID
• Export: Download traces as CSV for external analysis

Sessions

Sessions group related requests into conversations for easier analysis.

How Sessions Work

Bastio supports two methods for session management:

1. Auto-Sessions (Default)

• Requests grouped by end_user_id or device fingerprint
• 30-minute inactivity timeout creates new session
• Zero configuration required

2. Explicit Sessions

• Pass X-Session-ID header to group requests
• Same session ID always groups together regardless of timing
• Perfect for multi-turn conversations

Session View

Click into any session to see:

• Timeline: All traces in chronological order
• Metrics: Total tokens, cost, and average latency
• Duration: Time from first to last request
• Security Events: Any threats or blocks during the session

Analytics Dashboard

The analytics dashboard shows trends and patterns across your usage.

Available Charts

• Request Volume: Requests per hour, day, or week
• Cost Breakdown: Spending by model and over time
• Latency Distribution: P50, P95, P99 response times
• Security Metrics: Threats detected, requests blocked, average risk scores
• Session Statistics: Average session length, turns per session

Time Ranges

Select from different time windows based on your plan:

CSV Export

Export any chart data to CSV for:

• External reporting tools
• Custom analysis
• Compliance documentation
• Cost allocation

Evaluation

Run quality tests on your AI outputs to measure performance.

What You Can Test

• Accuracy: Does the response answer the question correctly?
• Relevance: Is the response on-topic?
• Safety: Does the response avoid harmful content?
• Custom Metrics: Define your own scoring criteria

Running Evaluations

1. Select sessions or individual traces to evaluate
2. Choose evaluation criteria
3. Run automated scoring
4. Review results and trends over time

Evaluation helps you:

• Compare model performance on similar inputs
• Track quality metrics over time
• Identify problematic responses for review
• Build confidence in your AI outputs

Pricing and Limits

Observability features are included in all plans:

Next Steps

• Sessions Documentation - Detailed session configuration and code examples
• API Reference - Programmatic access to traces and analytics
• Security Features - Understanding threat detection in traces