{
  "uid": "spring-ai-observability",
  "title": "Spring AI: tokens, latency and cost",
  "schemaVersion": 39,
  "version": 1,
  "editable": true,
  "refresh": "5s",
  "time": {
    "from": "now-15m",
    "to": "now"
  },
  "tags": [
    "spring-ai"
  ],
  "panels": [
    {
      "id": 1,
      "title": "Cost per hour by endpoint (USD, illustrative prices)",
      "type": "timeseries",
      "description": "Rate of app_ai_cost_usd_total scaled to an hourly figure. Prices come from ai.pricing in application.yml.",
      "datasource": {
        "type": "prometheus",
        "uid": "prom"
      },
      "gridPos": {
        "h": 8,
        "w": 12,
        "x": 0,
        "y": 0
      },
      "fieldConfig": {
        "defaults": {
          "unit": "currencyUSD"
        },
        "overrides": []
      },
      "targets": [
        {
          "refId": "A",
          "datasource": {
            "type": "prometheus",
            "uid": "prom"
          },
          "expr": "sum by (endpoint) (rate(app_ai_cost_usd_total[1m])) * 3600",
          "legendFormat": "{{endpoint}}",
          "editorMode": "code",
          "range": true
        }
      ]
    },
    {
      "id": 2,
      "title": "Tokens per minute by endpoint and direction",
      "type": "timeseries",
      "description": "Input and output tokens as the provider reported them.",
      "datasource": {
        "type": "prometheus",
        "uid": "prom"
      },
      "gridPos": {
        "h": 8,
        "w": 12,
        "x": 12,
        "y": 0
      },
      "fieldConfig": {
        "defaults": {
          "unit": "short"
        },
        "overrides": []
      },
      "targets": [
        {
          "refId": "A",
          "datasource": {
            "type": "prometheus",
            "uid": "prom"
          },
          "expr": "sum by (endpoint, type) (rate(app_ai_tokens_total[1m])) * 60",
          "legendFormat": "{{endpoint}} {{type}}",
          "editorMode": "code",
          "range": true
        }
      ]
    },
    {
      "id": 3,
      "title": "Cost per request by endpoint (USD)",
      "type": "bargauge",
      "description": "Cost divided by successful chat client calls. The two sides use different label names (endpoint vs app_endpoint), so label_replace renames one before the division.",
      "datasource": {
        "type": "prometheus",
        "uid": "prom"
      },
      "gridPos": {
        "h": 8,
        "w": 12,
        "x": 0,
        "y": 8
      },
      "fieldConfig": {
        "defaults": {
          "unit": "currencyUSD"
        },
        "overrides": []
      },
      "targets": [
        {
          "refId": "A",
          "datasource": {
            "type": "prometheus",
            "uid": "prom"
          },
          "expr": "sum by (endpoint) (increase(app_ai_cost_usd_total[5m])) / on (endpoint) label_replace(sum by (app_endpoint) (increase(spring_ai_chat_client_seconds_count{error=\"none\"}[5m])), \"endpoint\", \"$1\", \"app_endpoint\", \"(.*)\")",
          "legendFormat": "{{endpoint}}",
          "editorMode": "code",
          "range": true
        }
      ]
    },
    {
      "id": 4,
      "title": "Model call latency p50 / p95 (s)",
      "type": "timeseries",
      "description": "gen_ai.client.operation, one sample per HTTP call to the provider. Needs the percentiles-histogram property.",
      "datasource": {
        "type": "prometheus",
        "uid": "prom"
      },
      "gridPos": {
        "h": 8,
        "w": 12,
        "x": 12,
        "y": 8
      },
      "fieldConfig": {
        "defaults": {
          "unit": "s"
        },
        "overrides": []
      },
      "targets": [
        {
          "refId": "A",
          "datasource": {
            "type": "prometheus",
            "uid": "prom"
          },
          "expr": "histogram_quantile(0.50, sum by (le) (rate(gen_ai_client_operation_seconds_bucket{error=\"none\"}[1m])))",
          "legendFormat": "p50",
          "editorMode": "code",
          "range": true
        },
        {
          "refId": "B",
          "datasource": {
            "type": "prometheus",
            "uid": "prom"
          },
          "expr": "histogram_quantile(0.95, sum by (le) (rate(gen_ai_client_operation_seconds_bucket{error=\"none\"}[1m])))",
          "legendFormat": "p95",
          "editorMode": "code",
          "range": true
        }
      ]
    },
    {
      "id": 5,
      "title": "Failed model calls (share)",
      "type": "stat",
      "description": "Calls whose error tag is not none, over all calls.",
      "datasource": {
        "type": "prometheus",
        "uid": "prom"
      },
      "gridPos": {
        "h": 8,
        "w": 12,
        "x": 0,
        "y": 16
      },
      "fieldConfig": {
        "defaults": {
          "unit": "percentunit"
        },
        "overrides": []
      },
      "targets": [
        {
          "refId": "A",
          "datasource": {
            "type": "prometheus",
            "uid": "prom"
          },
          "expr": "sum(rate(gen_ai_client_operation_seconds_count{error!=\"none\"}[5m])) / sum(rate(gen_ai_client_operation_seconds_count[5m]))",
          "legendFormat": "failed",
          "editorMode": "code",
          "range": true
        }
      ]
    },
    {
      "id": 6,
      "title": "Tool calls per minute",
      "type": "timeseries",
      "description": "spring.ai.tool, one series per tool name.",
      "datasource": {
        "type": "prometheus",
        "uid": "prom"
      },
      "gridPos": {
        "h": 8,
        "w": 12,
        "x": 12,
        "y": 16
      },
      "fieldConfig": {
        "defaults": {
          "unit": "short"
        },
        "overrides": []
      },
      "targets": [
        {
          "refId": "A",
          "datasource": {
            "type": "prometheus",
            "uid": "prom"
          },
          "expr": "sum by (spring_ai_tool_definition_name) (rate(spring_ai_tool_seconds_count[1m])) * 60",
          "legendFormat": "{{spring_ai_tool_definition_name}}",
          "editorMode": "code",
          "range": true
        }
      ]
    },
    {
      "id": 7,
      "title": "Mean tool latency (s)",
      "type": "timeseries",
      "description": "Sum over count of the tool timer.",
      "datasource": {
        "type": "prometheus",
        "uid": "prom"
      },
      "gridPos": {
        "h": 8,
        "w": 12,
        "x": 0,
        "y": 24
      },
      "fieldConfig": {
        "defaults": {
          "unit": "s"
        },
        "overrides": []
      },
      "targets": [
        {
          "refId": "A",
          "datasource": {
            "type": "prometheus",
            "uid": "prom"
          },
          "expr": "sum by (spring_ai_tool_definition_name) (rate(spring_ai_tool_seconds_sum[1m])) / sum by (spring_ai_tool_definition_name) (rate(spring_ai_tool_seconds_count[1m]))",
          "legendFormat": "{{spring_ai_tool_definition_name}}",
          "editorMode": "code",
          "range": true
        }
      ]
    },
    {
      "id": 8,
      "title": "Calls with no cost recorded",
      "type": "timeseries",
      "description": "Calls that failed, reported no usage, or used a model with no configured price. Zero cost is never recorded silently.",
      "datasource": {
        "type": "prometheus",
        "uid": "prom"
      },
      "gridPos": {
        "h": 8,
        "w": 12,
        "x": 12,
        "y": 24
      },
      "fieldConfig": {
        "defaults": {
          "unit": "short"
        },
        "overrides": []
      },
      "targets": [
        {
          "refId": "A",
          "datasource": {
            "type": "prometheus",
            "uid": "prom"
          },
          "expr": "sum by (endpoint, reason) (increase(app_ai_unpriced_total[5m]))",
          "legendFormat": "{{endpoint}} {{reason}}",
          "editorMode": "code",
          "range": true
        }
      ]
    }
  ]
}