Concepts
Observability
Every execution is automatically recorded. You have complete visibility into what is happening in production — with zero configuration.
What is recorded
For each execution, Reliant records:
| Field | Description |
|---|---|
| execution_id | Unique ID for tracking and debugging |
| status | success, fallback ou failed |
| attempts | How many attempts were needed |
| latency_ms | Total latency including all retries |
| tokens_used | Total tokens consumed across all attempts |
| input_prompt | The prompt sent by the user |
| output | The final output returned |
| validation_errors | Validation errors from failed attempts |
| schema_version | Schema version used in the execution |
| provider / model | Provider and model used |
Dashboard
The Dashboard displays aggregated metrics in real time:
📊
Success Rate
Percentage of executions that returned valid output without needing the fallback.
⚡
Avg Latency
Average response time including all necessary retries.
🔁
Avg Attempts
Average attempts per execution — indicates how difficult the schema is for the model.
🪙
Tokens Consumed
Total tokens used in the period — useful for cost control.
Querying executions via API
typescript
// List recent executions
const executions = await reliant.listExecutions({
limit: 20,
status: 'failed', // filter by status
schema_id: 'sch_...', // filter by schema
})
// Fetch specific execution
const execution = await reliant.getExecution('exec_...')
console.log(execution.validation_errors) // see validation errors
// Project metrics
const metrics = await reliant.getMetricsSummary(30) // last 30 days
console.log(metrics.success_rate) // ex: 98.5
console.log(metrics.avg_latency_ms) // ex: 743