# SLO (Service Level Objective)

> An SLO is a reliability target for a service, expressed as the percentage of good events over a time window — for example, 99.9% of responses under 2 seconds over 30 days.

## SLI, SLO and SLA

- **SLI:** the measured indicator (e.g., share of successful responses).
- **SLO:** the internal target for that indicator.
- **SLA:** the contractual commitment, usually looser than the SLO.

## Error budget

If your SLO is 99.9%, you can fail 0.1% of events. That's the **error budget**: while it lasts, the team ships changes; when it runs out, reliability comes first.

## SLOs for LLMs

Beyond availability and latency, LLM features need quality indicators (refusals, negative ratings) and cost per request.