Why should you care?

Hospital SLA requires 99.9% triage API uptime; finance needs alerts when ERP agent error rate exceeds 2%.

See it live — copy this example

Paste into an ASP.NET Core 8+ / AgentNest project, then run with dotnet run (set your API keys in user-secrets).

// AgentNest.Api/Health/AiModelHealthCheck.cs
public sealed class AiModelHealthCheck(IChatClient chat) : IHealthCheck
{
    public async Task<HealthCheckResult> CheckHealthAsync(HealthCheckContext ctx, CancellationToken ct)
    {
        var sw = Stopwatch.StartNew();
        try
        {
            await chat.GetResponseAsync(["ping"], new ChatOptions { MaxOutputTokens = 5 }, ct);
            return HealthCheckResult.Healthy($"ok in {sw.ElapsedMilliseconds}ms");
        }
        catch (Exception ex) { return HealthCheckResult.Unhealthy(ex.Message); }
    }
}

What happened?

• AiModelHealthCheck pings the model with a tiny prompt.
• Kubernetes liveness uses it; Azure Monitor alerts on Unhealthy status.

Practice next

1. Register AddHealthChecks().AddCheck("ai_model").
2. Map /health/ready for orchestrators.
3. Create alert when unhealthy > 3 minutes.
4. Add dependency health for vector store and Redis.
5. Track copilot satisfaction thumbs up/down metric.

Remember

Monitor model reachability with lightweight health checks. Alert on latency, errors, and token spend per tenant. Separate liveness from deep RAG pipeline checks.

Hospital SLA pager

Azure OpenAI regional blip breaks triage endpoint.

Outcome: AiModelHealthCheck fails; failover route activates secondary region deployment.