Kibana Dashboards and Visualizations

Overview

The Kibana dashboards and visualizations APIs provide a declarative, Git-friendly format for defining dashboards and visualizations. Definitions are minimal, diffable, and suitable for version control and LLM-assisted generation.

Key Benefits:

• Minimal payloads (no implementation details or derivable properties)
• Easy to diff in Git
• Consistent patterns for GitOps workflows
• Designed for LLM one-shot generation
• Robust validation via OpenAPI spec

Version Requirement: Kibana 9.4+ (SNAPSHOT)

Important Caveats

ES|QL Visualizations: ES|QL-based visualizations cannot be created via /api/visualizations. They must be created as inline panels within dashboards using the Dashboard API.

Inline vs Saved Object References: When embedding visualization panels in dashboards, prefer inline definitions over ref_id references. Inline definitions are more reliable and self-contained.

Quick Start

Environment Configuration

Kibana connection is configured via environment variables. Run node scripts/kibana-dashboards.js test to verify the connection. If the test fails, suggest these setup options to the user, then stop. Do not try to explore further until a successful connection test.

Option 1: Elastic Cloud (recommended for production)

export KIBANA_CLOUD_ID="deployment-name:base64encodedcloudid"
export KIBANA_API_KEY="base64encodedapikey"

Option 2: Direct URL with API Key

export KIBANA_URL="https://your-kibana:5601"
export KIBANA_API_KEY="base64encodedapikey"

Option 3: Basic Authentication

export KIBANA_URL="https://your-kibana:5601"
export KIBANA_USERNAME="elastic"
export KIBANA_PASSWORD="changeme"

Option 4: Local Development with start-local

Use start-local to spin up Elasticsearch/Kibana locally, then source the generated .env:

curl -fsSL https://elastic.co/start-local | sh
source elastic-start-local/.env
export KIBANA_URL="$KB_LOCAL_URL"
export KIBANA_USERNAME="elastic"
export KIBANA_PASSWORD="$ES_LOCAL_PASSWORD"

Then run node scripts/kibana-dashboards.js test to verify the connection.

Optional: Skip TLS verification (development only)

export KIBANA_INSECURE="true"

Basic Workflow

# Test connection and API availability
node scripts/kibana-dashboards.js test

# Dashboard operations
node scripts/kibana-dashboards.js dashboard get <id>
echo '<json>' | node scripts/kibana-dashboards.js dashboard create -
echo '<json>' | node scripts/kibana-dashboards.js dashboard update <id> -
node scripts/kibana-dashboards.js dashboard delete <id>
echo '<json>' | node scripts/kibana-dashboards.js dashboard upsert <id> -

# Visualization operations (standalone saved objects)
node scripts/kibana-dashboards.js vis list
node scripts/kibana-dashboards.js vis get <id>
echo '<json>' | node scripts/kibana-dashboards.js vis create -
echo '<json>' | node scripts/kibana-dashboards.js vis update <id> -
node scripts/kibana-dashboards.js vis delete <id>
echo '<json>' | node scripts/kibana-dashboards.js vis upsert <id> -

Dashboards API

Dashboard Definition Structure

The API expects a flat request body with title and panels at the root level. The response wraps these in a data envelope alongside id, meta, and spaces.

{
  "title": "My Dashboard",
  "panels": [ ... ],
  "time_range": {
    "from": "now-24h",
    "to": "now"
  }
}

Note: Dashboard IDs are auto-generated by the API. The script also accepts the legacy wrapped format { id?, data: { title, panels }, spaces? } and unwraps it automatically.

Dashboard with Inline Visualization Panels (Recommended)

Use inline definitions (properties directly in config) for self-contained, portable dashboards:

{
  "title": "My Dashboard",
  "panels": [
    {
      "type": "vis",
      "id": "metric-panel",
      "grid": { "x": 0, "y": 0, "w": 12, "h": 6 },
      "config": {
        "title": "",
        "type": "metric",
        "data_source": { "type": "esql", "query": "FROM logs | STATS total = COUNT(*)" },
        "metrics": [{ "type": "primary", "column": "total", "label": "Total Count" }]
      }
    },
    {
      "type": "vis",
      "id": "chart-panel",
      "grid": { "x": 12, "y": 0, "w": 36, "h": 8 },
      "config": {
        "title": "Events Over Time",
        "type": "xy",
        "axis": {
          "x": { "scale": "temporal", "domain": { "type": "fit", "rounding": false } }
        },
        "layers": [
          {
            "type": "area",
            "data_source": {
              "type": "esql",
              "query": "FROM logs | WHERE @timestamp <= ?_tend AND @timestamp > ?_tstart | STATS count = COUNT(*) BY BUCKET(@timestamp, 75, ?_tstart, ?_tend)"
            },
            "x": { "column": "BUCKET(@timestamp, 75, ?_tstart, ?_tend)", "label": "@timestamp" },
            "y": [{ "column": "count" }]
          }
        ]
      }
    }
  ],
  "time_range": { "from": "now-24h", "to": "now" }
}

Dashboard Grid System

Dashboards use a 48-column, infinite-row grid. On 16:9 screens, approximately 20-24 rows are visible without scrolling. Design for density—place primary KPIs and key trends above the fold.

Target: 8-12 panels above the fold. Use descriptive panel titles on the charts themselves instead of adding markdown headers.

Grid Packing Rules:

• Eliminate Dead Space: Always calculate the bottom edge (y + h) of every panel. When starting a new row or placing a panel below another, its y coordinate must exactly match the y + h of the panel immediately above it.
• Align Row Heights: If multiple panels are placed side-by-side in a row (e.g., sharing the same y coordinate), they should generally have the exact same height (h). If they do not, you must fill the resulting empty vertical space before placing the next full-width panel.

Panel Schema

{
  "type": "vis",
  "id": "unique-panel-id",
  "grid": { "x": 0, "y": 0, "w": 24, "h": 15 },
  "config": { ... }
}

Visualizations API

Supported Chart Types

Note: To create donut charts, use pie with styling.donut_hole set to "s", "m", or "l" (small, medium, large hole). Use "none" for a solid pie. Example: "styling": { "donut_hole": "m" }.

Dataset Types

There are three dataset types supported in the Visualizations API. Each uses different patterns for specifying metrics and dimensions.

Data View Dataset

Use data_view_reference with aggregation operations. Kibana performs the aggregations automatically.

{
  "data_source": {
    "type": "data_view_reference",
    "ref_id": "90943e30-9a47-11e8-b64d-95841ca0b247"
  }
}

Available operations: count, average, sum, max, min, unique_count, median, standard_deviation, percentile, percentile_rank, last_value, date_histogram, terms. See Chart Types Reference for details.

ES|QL Dataset

Use esql with a query string. Reference the output columns using { column: 'column_name' }.

{
  "data_source": {
    "type": "esql",
    "query": "FROM logs | STATS count = COUNT(), avg_bytes = AVG(bytes) BY host"
  }
}

ES|QL Column Reference Pattern:

{ "column": "count" }

Key Difference: With ES|QL, you write the aggregation in the query itself, then reference the resulting columns. With data view, you specify the aggregation operation and Kibana performs it.

Important: ES|QL visualizations cannot be created via /api/visualizations. They must be created as inline panels in dashboards via the Dashboard API.

Index Dataset

Use index for ad-hoc index patterns without a saved data view:

{
  "data_source": {
    "type": "data_view_spec",
    "index_pattern": "logs-*",
    "time_field": "@timestamp"
  }
}

Examples

For detailed schemas and all chart type options, see Chart Types Reference.

Metric (Data View):

{
  "type": "metric",
  "data_source": { "type": "data_view_reference", "ref_id": "90943e30-9a47-11e8-b64d-95841ca0b247" },
  "metrics": [{ "type": "primary", "operation": "count", "label": "Total Requests" }]
}

Metric (ES|QL):

{
  "type": "metric",
  "data_source": { "type": "esql", "query": "FROM logs | STATS count = COUNT()" },
  "metrics": [{ "type": "primary", "column": "count", "label": "Total Requests" }]
}

XY Bar Chart (Data View):

{
  "title": "Top Hosts",
  "type": "xy",
  "axis": { "x": { "title": { "visible": false } }, "y": { "title": { "visible": false } } },
  "layers": [
    {
      "type": "bar_horizontal",
      "data_source": { "type": "data_view_reference", "ref_id": "90943e30-9a47-11e8-b64d-95841ca0b247" },
      "x": { "operation": "terms", "fields": ["host.keyword"], "limit": 10 },
      "y": [{ "operation": "count" }]
    }
  ]
}

XY Time Series (ES|QL):

{
  "title": "Requests Over Time",
  "type": "xy",
  "axis": {
    "x": { "title": { "visible": false }, "scale": "temporal", "domain": { "type": "fit", "rounding": false } },
    "y": { "title": { "visible": false } }
  },
  "layers": [
    {
      "type": "line",
      "data_source": {
        "type": "esql",
        "query": "FROM logs | WHERE @timestamp <= ?_tend AND @timestamp > ?_tstart | STATS count = COUNT() BY BUCKET(@timestamp, 75, ?_tstart, ?_tend)"
      },
      "x": { "column": "BUCKET(@timestamp, 75, ?_tstart, ?_tend)", "label": "@timestamp" },
      "y": [{ "column": "count" }]
    }
  ]
}

Tip: Always hide axis titles when the panel title is descriptive. Use bar_horizontal for categorical data with long labels. Use axis for axis configuration.

Full Documentation

• Dashboard API Reference — Dashboard endpoints and schemas
• Visualizations API Reference — Visualization endpoints
• Chart Types Reference — Detailed schemas for each chart type
• Example Definitions — Ready-to-use definitions

Key Example Files

See assets/ for ready-to-use definitions: demo-dashboard.json, dashboard-with-visualizations.json, metric-esql.json, bar-chart-esql.json, line-chart-timeseries.json.

Common Issues

Guidelines

1. Design for density — Operational dashboards must show 8-12 panels above the fold (within the first 24 rows). Use compact panel heights: metrics MUST be h=4 to h=6, and charts MUST be h=8 to h=12.
2. Never use Markdown for titles/headers — Do NOT add markdown panels to act as dashboard titles or section dividers. This wastes critical vertical space. Use descriptive panel titles on the charts themselves.
3. Prioritize above the fold — Primary KPIs and key trends must be placed at y=0. Deep-dives and data tables should be placed below the charts.
4. Use descriptive chart titles, hide axis titles — Write titles that explain what the chart shows (e.g., "Requests by Response Code"). A good panel title makes axis titles redundant. Always set axis.x.title.visible: false and axis.y.title.visible: false.
5. Choose the right dataset type — Use data_view_reference for simple aggregations, esql for complex queries
6. Inline definitions — Prefer inline properties in config over config.ref_id for portable dashboards
7. Test connection first — Run node scripts/kibana-dashboards.js test before creating resources
8. Get existing examples — Use vis get <id> to see the exact schema for different chart types (the CLI subcommand is vis)
9. Avoid redundant metric labels — For ES|QL metrics, avoid using both a panel title and an inner metric label, as it wastes space. Set the panel title to "" and configure the human-readable label by aliasing the ES|QL column name using backticks (e.g., STATS `Total Requests` = COUNT() and "column": "Total Requests").
10. Format numbers with units — Use the format property on metrics and y-axis columns to display proper units instead of raw numbers. Types: bytes, bits, number, percent, duration, custom. Example: "format": { "type": "bytes", "decimals": 0 }. See Chart Types Reference for the full format table.

Schema Differences: Data View vs ES|QL

Key Pattern: ES|QL uses { column: 'column_name' } to reference columns from the query result. The aggregation happens in the ES|QL query itself. Use data_source for all data source configuration.

Data source types: Use data_view_reference (with ref_id) for saved data views, data_view_spec (with index_pattern) for ad-hoc index patterns, and esql for ES|QL queries.

ES|QL: Time Bucketing

Use BUCKET(@timestamp, n, ?_tstart, ?_tend) for time series charts. The numeric argument is the target number of buckets. Kibana injects ?_tstart/?_tend automatically. Do not reassign the result — use the full expression BUCKET(@timestamp, 75, ?_tstart, ?_tend) as both the BY clause and the column reference. Set "label" to provide a friendly display name:

"x": { "column": "BUCKET(@timestamp, 75, ?_tstart, ?_tend)", "label": "@timestamp" }

Important: To get a proper multilevel time axis (e.g., "9th / April 2026 / 10th") instead of raw timestamp labels, you must set "scale": "temporal" on the x-axis:

"axis": {
  "x": { "scale": "temporal", "domain": { "type": "fit", "rounding": false } }
}

Without "scale": "temporal", Kibana treats the bucket column as categorical text and renders unsorted, verbose timestamp strings.

FROM logs | WHERE @timestamp <= ?_tend AND @timestamp > ?_tstart | STATS count = COUNT(*) BY BUCKET(@timestamp, 75, ?_tstart, ?_tend)

Note: BUCKET(@timestamp, n, ?_tstart, ?_tend) requires a WHERE clause with ?_tstart/?_tend bounds (Kibana injects these). Alternatively, use BUCKET(@timestamp, 1 hour) with a fixed duration — this does not require parameters but won't auto-scale.

ES|QL: Extracting Date Parts

Use DATE_EXTRACT(part, date) with ES|QL part names (not SQL keywords). The part string must be double-quoted. Common parts: "hour_of_day", "day_of_week", "day_of_month", "month_of_year", "year", "day_of_year".

FROM logs | STATS count = COUNT() BY hour = DATE_EXTRACT("hour_of_day", @timestamp), day = DATE_EXTRACT("day_of_week", @timestamp)

ES|QL: Creating Static/Constant Values

ES|QL does not support static_value operations. Instead, create constant columns using EVAL:

FROM logs | STATS count = COUNT() | EVAL max_value = 20000, goal = 15000

Then reference with { "column": "max_value" }. For dynamic reference values, use aggregation functions like PERCENTILE() or MAX() in the query.

Design Principles

The APIs follow these principles:

1. Minimal definitions — Only required properties; defaults are injected
2. No implementation details — No internal state or machine IDs
3. Flat structure — Shallow nesting for easy diffing
4. Semantic names — Clear, readable property names
5. Git-friendly — Easy to track changes in version control
6. LLM-optimized — Compact format suitable for one-shot generation