"description": "Live first-party CLI account telemetry. A last-good snapshot remains visible for up to 20 minutes through a transient fetch miss; its age is shown separately. Not exposed means access is healthy but that credential scope cannot read quota; unavailable means access is unhealthy or the snapshot is stale."
"description": "Live first-party CLI account telemetry. A last-good snapshot remains visible for up to 20 minutes through a transient fetch miss; its age is shown separately. Not exposed means access is healthy but that credential scope cannot read quota; unavailable means access is unhealthy or the snapshot is stale."
"description": "Live first-party CLI account telemetry. A last-good snapshot remains visible for up to 20 minutes through a transient fetch miss; its age is shown separately. Not exposed means access is healthy but that credential scope cannot read quota; unavailable means access is unhealthy or the snapshot is stale."
"description": "Live first-party CLI account telemetry. A last-good snapshot remains visible for up to 20 minutes through a transient fetch miss; its age is shown separately. Not exposed means access is healthy but that credential scope cannot read quota; unavailable means access is unhealthy or the snapshot is stale."
"description": "Live first-party CLI account telemetry. A last-good snapshot remains visible for up to 20 minutes through a transient fetch miss; its age is shown separately. Not exposed means access is healthy but that credential scope cannot read quota; unavailable means access is unhealthy or the snapshot is stale."
"expr": "(sum(increase(switchyard_prompt_tokens_total{model=~\"route/claude/.+\"}[24h])) or on() vector(0)) + (sum(increase(switchyard_cached_tokens_total{model=~\"route/claude/.+\"}[24h])) or on() vector(0)) + (sum(increase(switchyard_cache_creation_tokens_total{model=~\"route/claude/.+\"}[24h])) or on() vector(0)) + (sum(increase(switchyard_completion_tokens_total{model=~\"route/claude/.+\"}[24h])) or on() vector(0))",
"description": "Prompt, cache-read, cache-creation, and completion tokens from Claude calls routed through Switchyard in the last 24 hours. This measures consumption even when Anthropic does not expose the subscription allowance."
"expr": "(sum(increase(switchyard_prompt_tokens_total{model=~\"route/claude/.+\"}[7d])) or on() vector(0)) + (sum(increase(switchyard_cached_tokens_total{model=~\"route/claude/.+\"}[7d])) or on() vector(0)) + (sum(increase(switchyard_cache_creation_tokens_total{model=~\"route/claude/.+\"}[7d])) or on() vector(0)) + (sum(increase(switchyard_completion_tokens_total{model=~\"route/claude/.+\"}[7d])) or on() vector(0))",
"description": "Prompt, cache-read, cache-creation, and completion tokens from Claude calls routed through Switchyard in the last seven days. This is measured locally and does not require a broader Anthropic OAuth scope."
"description": "Live first-party CLI account telemetry. A last-good snapshot remains visible for up to 20 minutes through a transient fetch miss; its age is shown separately. Not exposed means access is healthy but that credential scope cannot read quota; unavailable means access is unhealthy or the snapshot is stale."
"description": "Live first-party CLI account telemetry. A last-good snapshot remains visible for up to 20 minutes through a transient fetch miss; its age is shown separately. Not exposed means access is healthy but that credential scope cannot read quota; unavailable means access is unhealthy or the snapshot is stale."
"description": "Successful client-facing Switchyard responses in the selected range."
},
{
"id": 19,
"type": "stat",
"title": "Classifier Fail-Open (Range)",
"datasource": {
"type": "prometheus",
"uid": "atlas-vm"
},
"gridPos": {
"h": 4,
"w": 6,
"x": 12,
"y": 24
},
"targets": [
{
"expr": "sum(increase(switchyard_classifier_fail_open_total[$__range])) or on() vector(0)",
"refId": "A"
}
],
"fieldConfig": {
"defaults": {
"color": {
"mode": "thresholds"
},
"mappings": [],
"thresholds": {
"mode": "absolute",
"steps": [
{
"color": "dark-green",
"value": null
},
{
"color": "dark-yellow",
"value": 1
},
{
"color": "dark-red",
"value": 5
}
]
},
"unit": "none",
"custom": {
"displayMode": "auto"
},
"decimals": 0
},
"overrides": []
},
"options": {
"colorMode": "value",
"graphMode": "area",
"justifyMode": "center",
"reduceOptions": {
"calcs": [
"lastNotNull"
],
"fields": "",
"values": false
},
"textMode": "value"
},
"description": "Local classifier failures that safely fell back to the conservative hosted route."
},
{
"id": 20,
"type": "stat",
"title": "Upstream Errors (Range)",
"datasource": {
"type": "prometheus",
"uid": "atlas-vm"
},
"gridPos": {
"h": 4,
"w": 6,
"x": 18,
"y": 24
},
"targets": [
{
"expr": "sum(increase(switchyard_errors_total[$__range])) or on() vector(0)",
"refId": "A"
}
],
"fieldConfig": {
"defaults": {
"color": {
"mode": "thresholds"
},
"mappings": [],
"thresholds": {
"mode": "absolute",
"steps": [
{
"color": "dark-green",
"value": null
},
{
"color": "dark-yellow",
"value": 1
},
{
"color": "dark-red",
"value": 5
}
]
},
"unit": "none",
"custom": {
"displayMode": "auto"
},
"decimals": 0
},
"overrides": []
},
"options": {
"colorMode": "value",
"graphMode": "area",
"justifyMode": "center",
"reduceOptions": {
"calcs": [
"lastNotNull"
],
"fields": "",
"values": false
},
"textMode": "value"
},
"description": "Hosted model attempts that returned errors in the selected range."
},
{
"id": 21,
"type": "timeseries",
"title": "Local Classifier Calls",
"datasource": {
"type": "prometheus",
"uid": "atlas-vm"
},
"gridPos": {
"h": 8,
"w": 8,
"x": 0,
"y": 28
},
"targets": [
{
"expr": "sum by (outcome) (rate(switchyard_llm_calls_total{selected_model=~\"qwen.*\"}[5m]))",
"refId": "A",
"legendFormat": "{{outcome}}"
}
],
"fieldConfig": {
"defaults": {
"unit": "reqps"
},
"overrides": []
},
"options": {
"legend": {
"displayMode": "table",
"placement": "bottom"
},
"tooltip": {
"mode": "multi"
}
},
"description": "Local Qwen routing-classifier activity, split by successful and failed calls."
},
{
"id": 22,
"type": "timeseries",
"title": "Routing Overhead p95",
"datasource": {
"type": "prometheus",
"uid": "atlas-vm"
},
"gridPos": {
"h": 8,
"w": 8,
"x": 8,
"y": 28
},
"targets": [
{
"expr": "histogram_quantile(0.95, sum by (le, algorithm) (rate(switchyard_routing_overhead_ms_bucket[5m])))",
"refId": "A",
"legendFormat": "{{algorithm}}"
}
],
"fieldConfig": {
"defaults": {
"unit": "ms"
},
"overrides": []
},
"options": {
"legend": {
"displayMode": "table",
"placement": "bottom"
},
"tooltip": {
"mode": "multi"
}
},
"description": "95th percentile time Switchyard spends selecting a model before the upstream call."
},
{
"id": 23,
"type": "bargauge",
"title": "Traffic Lanes (Range)",
"datasource": {
"type": "prometheus",
"uid": "atlas-vm"
},
"gridPos": {
"h": 8,
"w": 8,
"x": 16,
"y": 28
},
"targets": [
{
"expr": "sort_desc(sum by (lane) (label_replace(increase(switchyard_requests_total{model=~\"(route|worker)/.*\"}[$__range]), \"lane\", \"$1\", \"model\", \"^(route|worker)/.*\")))",
"refId": "A",
"legendFormat": "{{lane}}",
"instant": true
}
],
"fieldConfig": {
"defaults": {
"unit": "short",
"min": 0,
"max": null,
"thresholds": {
"mode": "absolute",
"steps": [
{
"color": "dark-green",
"value": null
},
{
"color": "dark-yellow",
"value": 50
},
{
"color": "dark-orange",
"value": 70
},
{
"color": "dark-red",
"value": 85
}
]
}
},
"overrides": []
},
"options": {
"displayMode": "basic",
"orientation": "horizontal",
"reduceOptions": {
"calcs": [
"lastNotNull"
],
"fields": "",
"values": false
}
},
"description": "Request volume split between interactive route traffic and durable worker traffic.",
"transformations": [
{
"id": "sortBy",
"options": {
"fields": [
"Value"
],
"order": "desc"
}
}
]
},
{
"id": 24,
"type": "text",
"title": "Cost Semantics",
"gridPos": {
"h": 8,
"w": 12,
"x": 0,
"y": 36
},
"datasource": null,
"options": {
"mode": "markdown",
"content": "Codex and Claude currently run through first-party subscription OAuth lanes, so providers expose quota utilization rather than per-call dollar invoices. This dashboard does not invent API costs from tokens. If a metered API-key lane is added, its provider billing metric should be displayed separately from subscription usage."
}
},
{
"id": 25,
"type": "timeseries",
"title": "Hermes Workload CPU (Attribution Proxy)",
"datasource": {
"type": "prometheus",
"uid": "atlas-vm"
},
"gridPos": {
"h": 8,
"w": 12,
"x": 12,
"y": 36
},
"targets": [
{
"expr": "sum by (pod, container) (rate(container_cpu_usage_seconds_total{namespace=\"hermes\",pod=~\"hermes-(agent|chat-tenant|switchyard|model-gate).*\",container!=\"\",image!=\"\"}[5m]))",
"refId": "A",
"legendFormat": "{{pod}} \u00b7 {{container}}"
}
],
"fieldConfig": {
"defaults": {
"unit": "cores"
},
"overrides": []
},
"options": {
"legend": {
"displayMode": "table",
"placement": "right"
},
"tooltip": {
"mode": "multi"
}
},
"description": "Compute use by Hermes pod/container. Switchyard currently exposes model and worker-vs-route attribution, but not tenant-slot token labels; CPU is clearly marked as a proxy rather than token usage."
