Four primary signals — throughput, availability, latency, and errors — monitored across every agent, model call, tool, and integration in your AI stack. Real-time dashboards, alert thresholds, and copy-paste telemetry snippets.
| Signal | Definition | Alert condition | Current | Status |
|---|---|---|---|---|
| Throughput | Requests, agent runs, tokens/min | >30% above forecast or unexpected drop | 2,840 req/min | NORMAL |
| Availability | Successful agent requests / total | Alert below 99.9% | 99.94% | ABOVE SLA |
| Latency P95 | End-to-end response time | P95 > target for 3 consecutive windows | 284ms (target: 250ms) | REVIEW |
| Error rate | Failed requests, model failures, timeouts | >1% or 2× baseline spike | 0.06% | NORMAL |
| TTFT | Time to first token from model | >500ms avg for 5 min window | 94ms avg | NORMAL |
Copy one snippet into your service to start sending golden signals to Argovaa. Production code is locked — sign in to access your API key and copy-ready snippets.
<script> window.ArgovaaConfig = { apiKey: "YOUR_PUBLIC_CLIENT_KEY", service: "customer-agent-app", environment: "production", dashboardId: "golden-signals" }; </script> <script src="https://cdn.argovaa.com/telemetry/argovaa-golden-signals.js" async></script> // Automatically tracks: throughput, latency, errors, availability // on every agent call. No additional code needed.
import { ArgovaaTelemetry } from "@argovaa/telemetry"; const argovaa = new ArgovaaTelemetry({ apiKey: process.env.ARGOVAA_API_KEY, service: "agent-runtime-api", environment: "production" }); export async function runAgent(req, res) { const trace = argovaa.startTrace("agent_request", { tenantId: req.headers["x-tenant-id"], agentName: "support-agent" }); try { const result = await agent.execute(req.body); trace.recordThroughput({ requests: 1, tokens: result.usage.totalTokens }); trace.recordLatency({ totalMs: result.latencyMs, ttftMs: result.ttftMs }); trace.recordAvailability(true); res.json(result); } catch (error) { trace.recordError({ type: error.name, message: error.message }); trace.recordAvailability(false); res.status(500).json({ error: "Agent request failed" }); } finally { trace.end(); } }
from argovaa.telemetry import ArgovaaTelemetry argovaa = ArgovaaTelemetry( api_key=os.environ["ARGOVAA_API_KEY"], service="agent-runtime-api", environment="production" ) def run_agent(request): with argovaa.trace("agent_request", agent_name="finance-agent") as trace: try: response = agent.execute(request.json) trace.throughput(requests=1, tokens=response.usage.total_tokens) trace.latency(total_ms=response.latency_ms, ttft_ms=response.ttft_ms) trace.availability(success=True) return response.to_json() except Exception as exc: trace.error(type=exc.__class__.__name__, message=str(exc)) trace.availability(success=False) raise