Developing In Lightdash

Developing in Lightdash

For CLI users and writeback sandboxes working with YAML and dbt files.

Build and deploy Lightdash analytics projects. This skill covers the semantic layer (metrics, dimensions, joins) and content (charts, dashboards).

When to Use

• Working with Lightdash YAML files (charts, dashboards, models as code)
• Using the lightdash CLI (deploy, upload, download, preview, lint, warehouse-catalog, sql)
• Defining metrics, dimensions, joins, or tables in dbt or pure Lightdash projects
• Creating or editing charts and dashboards as code
• Downloading, uploading, or locally developing data apps (enterprise)
• Creating, downloading, editing, or uploading custom chart types built in Chart Studio (enterprise)
• Creating, editing, migrating, downloading, or uploading organization Data App themes

Don't use for: Developing the Lightdash application itself (use the codebase CLAUDE.md), general dbt work without Lightdash metadata, or raw SQL unrelated to Lightdash models.

What You Can Do

Common Mistakes

Before You Start

Every lightdash download or lightdash upload follows the Content Workflow: files go in a scratch directory, not the user's repository, and are not committed unless the user asks.

When a task uses lightdash download or lightdash upload, especially for bulk edits, spaces and access, scheduled content, AI agents, data apps, custom chart types, organization themes, external connections, users, groups, or custom roles, read and follow Content as Code first. Project and organization content require separate commands, and a default download is not a complete snapshot.

For any task that creates, edits, migrates, downloads, uploads, or tests an organization Data App theme, always read and follow Data App Themes before touching themes/. Theme packages are strict multi-file resources, organization upload has no theme-only mode, and lightdash lint does not validate them.

Check Your Target Project

Always verify which project you're deploying to. Deploying to the wrong project can overwrite production content.

lightdash config get-project        # Show current project
lightdash config list-projects      # List available projects
lightdash config set-project --name "My Project"  # Switch project

Detect Your Project Type

The YAML syntax differs significantly between project types.

ls dbt_project.yml 2>/dev/null && echo "dbt project" || echo "Not dbt"
ls lightdash.config.yml 2>/dev/null && echo "Pure Lightdash" || echo "Not pure Lightdash"

dbt Fusion / dbt 1.10+: Lightdash metadata must be nested under config: meta: instead of meta:. The properties are identical — only the nesting changes. Example:

models:
  - name: orders
    config:
      meta:
        metrics:
          total_revenue:
            type: sum
            sql: "${TABLE}.amount"

Syntax Comparison

dbt YAML (metadata under meta:):

models:
  - name: orders
    meta:
      metrics:
        total_revenue:
          type: sum
          sql: "${TABLE}.amount"
    columns:
      - name: status
        meta:
          dimension:
            type: string

Pure Lightdash YAML (top-level):

type: model
name: orders
sql_from: 'DB.SCHEMA.ORDERS'

metrics:
  total_revenue:
    type: sum
    sql: ${TABLE}.amount

dimensions:
  - name: status
    sql: ${TABLE}.STATUS
    type: string

Setting Up Warehouse Connection

If the project needs a different warehouse connection (e.g., switching from Postgres to BigQuery), update it from your profiles.yml:

lightdash set-warehouse --project-dir ./dbt --profiles-dir ./profiles --assume-yes

This reads credentials from profiles.yml, updates the warehouse connection on the currently selected project, and triggers a recompile. Run this before lightdash deploy.

To target a specific project:

lightdash set-warehouse --project-dir ./dbt --profiles-dir ./profiles --project <uuid> --assume-yes

Core Workflows

Verify Filter Values Before Using Them

CRITICAL: Never guess filter values. Case mismatches (e.g., 'Payment' vs 'payment') cause charts to silently return no data.

Filters are case-sensitive by default. The case_sensitive key can override this in order of priority:

• Dimension metadata
• Model/explore metadata
• lightdash.config.yml defaults.case_sensitive

Before writing any string filter, query actual values from the warehouse:

lightdash sql "SELECT DISTINCT category FROM payments LIMIT 50" -o category_values.csv

Read the CSV and use the exact values in your filter YAML. This applies to all equals/notEquals filters with string values — in charts and dashboards.

Editing Metrics & Dimensions

1. Find the model YAML file (dbt: models/*.yml, pure Lightdash: lightdash/models/*.yml)
2. Edit metrics/dimensions using the appropriate syntax for your project type
3. Validate: lightdash lint (pure Lightdash) or dbt compile (dbt projects)
4. Deploy: lightdash deploy

See Metrics Reference and Dimensions Reference for configuration options.

Creating a Pure Lightdash Project from a Warehouse Catalog

When the prepared project has no usable dbt project and the task is to bootstrap a semantic layer from warehouse metadata, always read and follow Creating from a Warehouse Catalog before inspecting data or writing YAML. This applies whether warehouse access comes from the selected Lightdash project or an already-authenticated warehouse CLI such as Snowflake CLI or bq. Do not use that workflow when an existing dbt semantic layer can be extended.

Content Workflow

This is the one loop for charts, dashboards, data apps, custom chart types, and other content. The sections below add only what is specific to each resource.

Creating content. Charts and dashboards can be created in the Lightdash UI, by a Lightdash AI agent, through the Lightdash MCP server, or as YAML with the CLI. All of these are equally valid; follow the user's preference.

Lightdash is the source of truth. Downloaded files are temporary working copies, not files to commit. To edit content as files, or to create it as YAML:

1. Confirm the target: lightdash config get-project
2. Download into a scratch directory outside the user's repository, never into the dbt project:

   LD_DIR="$(mktemp -d)/lightdash"
   lightdash download --dashboards executive-summary --path "$LD_DIR"   # or --charts <slug>, --apps <ref>, ...
   

   To create content, skip the download and write new YAML in $LD_DIR.
3. Edit the files in $LD_DIR, working through the checklist below
4. Lint: lightdash lint --path "$LD_DIR"
5. Upload with the same selectors: lightdash upload --dashboards executive-summary --path "$LD_DIR"
6. Delete $LD_DIR. Do not copy the files into the repository, git add them, or commit them.

For bulk or cross-scope changes (spaces and access, users, groups, roles, scheduled content, AI agents), follow the Agent-Safe Workflow in Content as Code instead.

When content belongs in git. Only when the user asks for it or the repository already tracks content (for example, git ls-files lightdash/ returns content files). That covers version-controlled templates, CI/CD deploys of content, and strict change management with code review. See Edit dashboards with agents. This applies only to content: dbt models, metrics, dimensions, and other semantic-layer YAML do live in the repository and ship with lightdash deploy.

Chart and Dashboard Checklist

• Filter values: if you added or changed a string filter, query the real values first (see Verify Filter Values).
• Dashboard tile titles: a saved_chart tile's title and chartName are independent overrides and do NOT auto-update when a chart is renamed. If you change a chart's name or purpose, download every dashboard with a tile whose chartSlug matches, update both properties, and upload those dashboards too:

  tiles:
    - type: saved_chart
      properties:
        chartSlug: total-revenue-kpi
        title: "Gross Profit"        # ← Update this when chart name/purpose changes
        chartName: "Gross Profit"    # ← Update this too
  

• Chart config: see Chart Types and the Dashboard Reference.

Working with Data Apps (Enterprise)

Data apps are multi-file React bundles under apps/<app-folder>/ with a lightdash-app.yml manifest — not single YAML files. Full flag semantics and manifest details: Content as Code.

Apps follow the Content Workflow: work in a scratch directory and do not commit the bundle unless the user asks. The commands below are the app-specific flags.

Download one app — --apps <ref> alone is the complete command (ref = slug, app URL, or UUID; also finds apps not added to any space):

lightdash download --apps revenue-explorer --path "$LD_DIR"

Never add --include-apps to "scope" the download — it always requests ALL project apps (see Common Mistakes).

Download all apps: lightdash download --include-apps (capped at 50; raise with --apps-limit <n>). Add --apps-only to skip charts, dashboards, and spaces.

Upload:

lightdash upload --apps revenue-explorer --path "$LD_DIR"   # one app (slug = folder name, URL, or UUID)
lightdash upload --include-apps --path "$LD_DIR"            # every app folder on disk

• --app-space <spaceRef> — space (slug or UUID) for apps this upload creates; existing apps keep their space.
• --create-new — create a fresh app (new slug) instead of updating the app referenced in lightdash-app.yml.
• --allow-custom-dependencies — required for non-interactive uploads of apps that declare custom npm dependencies.

Develop locally with the lightdash apps subcommand group:

lightdash apps create "Revenue Explorer" --path "$LD_DIR"   # scaffold a new app under $LD_DIR/apps/
lightdash apps preview                     # run the app locally against your real Lightdash instance, authenticated as you
lightdash apps validate                    # check source, manifest, dependencies, and semantic-layer references

Every created or downloaded app bundle ships its own skills inside the app folder:

• developing-data-apps-locally — the edit → validate → upload loop, local preview, SDK-only data access, and dependency boundaries
• lightdash-data-app — the @lightdash/query-sdk reference for the app's source code

When editing files inside an app folder, read those bundled skills first. They are version-matched to the app and authoritative for local development — this skill only covers moving apps between disk and Lightdash.

Working with Custom Chart Types (Enterprise)

A custom chart type is a reusable visualization built in Chart Studio: one React component that Lightdash hands query results and settings to, offered in the explorer's chart type picker for every chart in the project. On disk it is a multi-file bundle under chart-types/<slug>/ with a lightdash-app.yml manifest whose vizSchema declares the fields and options the component reads. It is not a data app (it runs no query of its own) and not the legacy Vega-Lite custom chart.

Chart types follow the Content Workflow: work in a scratch directory and do not commit the bundle unless the user asks.

Create one locally:

lightdash apps create "Radial Gauge" --chart-type --path "$LD_DIR"   # scaffolds $LD_DIR/chart-types/radial-gauge/

Download — --chart-types <ref> alone is the complete command (ref = slug, URL, or UUID):

lightdash download --chart-types radial-gauge --path "$LD_DIR"
lightdash download --include-chart-types   # every chart type in the project (capped at 50; raise with --chart-types-limit <n>)
lightdash download --chart-types-only      # chart types only, skipping charts, dashboards, and spaces

Downloading a chart that renders with a custom chart type also downloads that chart type. In chart YAML the binding is chartConfig.type: data_app_viz with config.dataAppVizSlug, fieldMapping, and optionValues.

Upload:

lightdash upload --chart-types radial-gauge --path "$LD_DIR"   # one chart type (slug = folder name, URL, or UUID)
lightdash upload --include-chart-types --path "$LD_DIR"        # every chart type folder on disk

A chart YAML that binds to a chart type by dataAppVizSlug fails to upload unless that chart type exists in the target project. Upload the chart type first, or pass --chart-types <ref> in the same run (chart types upload before charts).

Edit inside the folder. Change into chart-types/<slug>/ and read its AGENTS.md plus the reusable-visualization and developing-chart-types-locally skills before touching src/:

• reusable-visualization — the component contract: useVizContext() is the only channel to the host, and the component never queries or fetches anything itself
• developing-chart-types-locally — keeping vizSchema in lockstep with the component, lightdash apps validate --build, the fixture preview, and the upload-and-verify loop

Those files are version-matched to the chart type and authoritative for editing it. This skill only covers moving chart types between disk and Lightdash.

Working with Organization Data App Themes

Organization Data App themes are multi-file packages under themes/<slug>/ and participate automatically in lightdash download --organization and lightdash upload --organization. They do not have a standalone command group or theme-specific selectors.

Before creating, editing, migrating, synchronizing, or testing a theme, read and follow Data App Themes for the manifest contract, asset rules, synchronization behavior, and generation boundaries.

Testing with Preview

For larger changes, test in isolation:

lightdash preview --name "my-feature"
# Make changes and iterate
lightdash stop-preview --name "my-feature"

CLI Quick Reference

See CLI Reference for full command documentation.

Semantic Layer

The semantic layer defines your data model. See individual references for full configuration:

• Tables Reference — queryable entities, labels, joins
• Metrics Reference — aggregated calculations (count, sum, average, min, max, number, etc.)
• Dimensions Reference — attributes for grouping/filtering (string, number, boolean, date, timestamp)
• Joins Reference — cross-table relationships
• User Attributes Reference — SQL variables, row-level security, access control

Chart Types

All charts share a common base structure:

chartConfig:
  config: {}        # Type-specific — see individual references
  type: <type>
contentType: chart              # Required: chart, dashboard, or sql_chart
dashboardSlug: my-dashboard  # Optional: scopes chart to dashboard (won't appear in space)
metricQuery:
  dimensions:
    - my_explore_category
  exploreName: my_explore     # Required: which explore to query
  filters: {}
  limit: 500
  metrics:
    - my_explore_total_sales
  sorts: []
name: "Chart Name"
slug: unique-chart-slug
spaceSlug: target-space
tableConfig:
  columnOrder: []
tableName: my_explore           # Required: top-level explore/table name
version: 1

Key ordering: All YAML keys must be sorted alphabetically at every nesting level. The CLI writes files with sortKeys: true and warns on upload if keys are unsorted. When writing or editing YAML by hand, keep keys in alphabetical order to avoid warnings and noisy diffs.

Chart scoping: Use spaceSlug only for shared charts. Add dashboardSlug to scope a chart to a specific dashboard (it won't appear in the space).

Nested spaces: Spaces can be nested. In YAML, spaceSlug uses parent/child syntax to address a sub-space — the / denotes hierarchy. Examples:

spaceSlug: sales              # Top-level space "sales"
spaceSlug: sales/maps         # Sub-space "maps" inside "sales"
spaceSlug: sales/eu/forecasts # Deeper nesting works the same way

Each path segment must be the slug of an existing (or to-be-created) space at that level. A bare slug like sales-maps is a flat top-level space, NOT a sub-space — the slash is the only thing that creates the hierarchy.

Choosing the Right Chart Type

A SQL chart (contentType: sql_chart) is saved from raw warehouse SQL instead of an explore. See SQL Chart Reference.

Dashboards

Dashboards arrange charts and content in a grid layout. See Dashboard Reference for YAML structure, tile types, tabs, and filters.

Exploring the Warehouse

Use lightdash sql to explore data when building models:

# Preview table structure
lightdash sql "SELECT * FROM orders LIMIT 5" -o preview.csv

# Check distinct values for a dimension
lightdash sql "SELECT DISTINCT status FROM orders" -o statuses.csv

# Test metric calculations
lightdash sql "SELECT SUM(amount) FROM orders" -o test.csv

Workflow Patterns

See Workflows Reference for detailed examples and CI/CD configurations.

Resources

Semantic Layer

• Creating from a Warehouse Catalog
• Dimensions Reference
• Metrics Reference
• Tables Reference
• Joins Reference
• User Attributes Reference

Charts

• Cartesian Chart Reference - Bar, line, area, scatter
• Pie Chart Reference
• Table Chart Reference
• Big Number Reference
• Funnel Chart Reference
• Gauge Chart Reference
• Treemap Chart Reference
• Map Chart Reference
• Sankey Chart Reference
• Custom Viz Reference - Legacy Vega-Lite custom charts
• Period over Period Reference - PoP comparisons (YoY, MoM, etc.)
• SQL Chart Reference - Charts saved from raw warehouse SQL

Dashboards & Workflows

• Dashboard Reference
• Dashboard Best Practices
• Data App Themes Reference
• Content as Code Reference
• CLI Reference
• Workflows Reference

External

• Lightdash Docs