Inwire Observe User Guide
Monitor model, agent, and platform behavior through traces, metrics, alerts, costs, drift, errors, events, topology, and runbooks.
Who This Guide Is For
- SREs
- MLOps operators
- AI reliability teams
- Support engineers
Where To Go
| Page |
Use It For |
/inwire-observe |
Observe dashboard. |
/inwire-observe/metrics |
Metrics exploration. |
/inwire-observe/traces |
Trace inspection. |
/inwire-observe/alerts |
Alert rules and incidents. |
/inwire-observe/errors |
Error analysis. |
/inwire-observe/drift |
Drift monitoring. |
/inwire-observe/costs |
Cost analysis. |
/inwire-observe/topology |
Service topology. |
/inwire-observe/runbooks |
Operational runbooks. |
/inwire-observe/live-tail |
Live log tailing. |
Core Concepts
| Concept |
Meaning |
| Trace |
A request-level timeline across services and model calls. |
| Metric |
A numeric signal such as latency, throughput, error rate, token cost, or GPU usage. |
| Drift |
A change in inputs, embeddings, outputs, or behavior over time. |
| Alert |
A rule that triggers when a signal crosses a threshold. |
| Runbook |
A documented response procedure for known incidents. |
Common Workflows
Investigate an incident
- Open Alerts or Errors.
- Identify impacted deployment, agent, or service.
- Open traces and metrics for the same time window.
- Check topology and dependency status.
- Run the relevant runbook.
- Record outcome and follow-up.
Monitor drift
- Open Drift.
- Select model, pipeline, or embedding source.
- Review drift metric and sample changes.
- Compare with recent releases or data changes.
- Trigger evaluation, retraining, or rollback if needed.
Best Practices
- Define SLOs for production-facing models and agents.
- Connect alerts to ownership and runbooks.
- Review cost and latency together for LLM workloads.
- Use traces to diagnose cross-service failures before changing model settings.