docs/guides/fleet monitoring

Fleet Monitoring

Track robot health, model performance, and anomalies across your entire fleet in real time.

Dashboard Overview

The Artemis dashboard provides a live view of your fleet. The main monitoring panels are:

Fleet Map

Visual overview of all robots, their online/offline status, and active model versions.

Live Telemetry

Real-time inference latency, throughput, and error rates per robot.

Model Health

Accuracy drift detection — compares current inference confidence vs. baseline.

Resource Usage

CPU, GPU, RAM, and disk utilisation across the fleet.

Alert Feed

Chronological log of all triggered alerts with severity and resolution status.

Key Metrics

Artemis tracks the following metrics for each deployed model:

Metric	Description	Alert threshold
inference_latency_p50	Median inference time per request	> 50 ms
inference_latency_p99	99th percentile inference time	> 100 ms
error_rate	Fraction of failed inference calls	> 1%
confidence_score	Average model output confidence	< 0.6
throughput	Inference requests per second	< 5 req/s
cpu_usage	Agent CPU utilisation	> 90%
memory_usage	Agent memory utilisation	> 85%

Monitoring via CLI

Stream live metrics for a robot directly in your terminal:

bash

kairo monitor --robot rob_a1b2c3d4 --follow

# Robot: my-arm-01  |  Model: arm-model-v1
# ─────────────────────────────────────────
# Latency p50:   11 ms    Latency p99:  24 ms
# Error rate:    0.2%     Throughput:   18 req/s
# Confidence:    0.94     CPU:          34%
# ─────────────────────────────────────────
# [12:01:03] inference ok  14ms  conf=0.96
# [12:01:04] inference ok  11ms  conf=0.94
# [12:01:05] inference ok  13ms  conf=0.91

Configuring Alerts

Define custom alert rules in your workspace settings or via the CLI. Alerts can notify via email, Slack, webhook, or PagerDuty:

bash

kairo alerts create \
  --robot rob_a1b2c3d4 \
  --metric inference_latency_p99 \
  --condition "gt 80ms" \
  --channel slack \
  --webhook https://hooks.slack.com/services/xxx

List and manage existing alerts:

bash

kairo alerts list --robot rob_a1b2c3d4
kairo alerts delete --alert alt_abc123

Anomaly Detection

Artemis continuously compares live inference behaviour against a baseline captured at deployment time. When the distribution of model outputs shifts significantly — indicating sensor drift, hardware changes, or an out-of-distribution environment — an anomaly alert fires.

Enable anomaly detection for a deployment:

bash

kairo deploy push \
  --model mdl_abc123 \
  --robot rob_a1b2c3d4 \
  --anomaly-detection enabled \
  --anomaly-sensitivity medium   # low | medium | high

✦

When an anomaly alert fires, consider whether you need to retrain with more recent data (if the environment has genuinely changed) or investigate the hardware (if the robot behaves unexpectedly).

Viewing Logs

bash

# Stream agent logs
kairo logs --robot rob_a1b2c3d4 --follow

# Download logs for a time window
kairo logs --robot rob_a1b2c3d4 \
  --from "2026-03-20T10:00:00Z" \
  --to   "2026-03-20T11:00:00Z" \
  --output ./robot-logs.txt

←

Deploying to Production

API Reference

→

Was this page helpful?