omni-content-builder

Omni Content Builder

Create, update, and manage Omni documents, dashboards, and apps programmatically via the Omni CLI — document lifecycle, drafts, workbook models, filters, controls, and dashboard content.

Tip: Use omni-model-explorer to understand available fields and omni-content-explorer to find existing dashboards to modify or learn from.

Documents are created and edited through the v2 documents API (omni documents v2-*) — an explicit envelope of queryPresentations, controls, containers, and settings, edited through a draft → publish flow. This is the only path for building, reading, or changing a document — never fall back to the v1 documents create/get commands (v1 put/update were removed in CLI 1.2.2). A few document-management operations (list, delete, move, duplicate, downloads) have no v2 form; see Commands below.

Known Issues & Safe Defaults

• Always run the full validation loop — see Validation Loops below. At minimum: validate the model, test every query via omni query run, check viz spec consistency, and verify the dashboard by reading the draft back and executing its queries before publishing.
• Never round-trip a v2-get tile back into a patch unchanged — a GET returns the inner vis config flat, but a patch only persists it nested under config. Re-sending the flat shape — even for an unrelated edit like a rename — silently drops the vis config (KPI loses its number, charts lose mark/series, markdown goes blank). Always re-author the inner visConfig nested under config. See references/documents-v2.md.
  • This applies to any write whose tile JSON came from a GET payload — not just edits. Restoring, reverting, duplicating, or moving a tile by copying it out of a v2-get/snapshot and patching it back is also a round-trip: the inner config is flat and will be dropped (renders "No chart available"). A "restore to how it was" still needs the config re-nested. And verify the specific tile you wrote — re-read it (v2-get) and run its query to confirm the config persisted; if a visual/preview tool is available, you can also request a screenshot to confirm the render — including reverts; don't assume a restore is safe.
• Patches merge by key; only containers (and the order arrays) are full replacements. To change one tile, send just that key. To delete a tile, set its key to null AND remove it from order. When you send containers, send the complete layout tree with your edit applied.
• A multi-tile v2-create only lays out the first tile — the rest are stored but render nowhere until you author the full containers tree. See references/containers.md.
• Tile "1" on create merges over a server seed tile — some seed properties (e.g. automaticVis: true) can win over what you sent. Read the document back and re-patch tile "1" if its exact fields matter.
• Every tile query must include the full collection-field set — sorts, filters, calculations, column_totals, row_totals, fill_fields, pivots, userEditedSQL (empty values are fine) alongside table, fields, limit, join_paths_from_topic_name. Omitting the schema-required ones is a 400 with per-field errors. Do not include modelId or model_extension_id — the server anchors tiles to the document's workbook model and silently rewrites any value you send.
• HARD RULE — never build a non-topic tile from handed-over SQL without an explicit user decision. When the user hands you SQL (or a metric) and existing topics don't express it, you may not silently convert it to a non-topic / userEditedSQL tile. First apply the topic-first reflex (see omni-query): map each query's intent to a topic. If none fits, stop and ask whether to model it — extend a topic or create a new one (on a branch via omni-model-builder, validated, merged only with confirmation). Only build a non-topic / raw-SQL tile after the user has explicitly chosen that path (declined modeling, or it's a genuine one-off / userEditedSQL is required). Non-topic + Access Boost is never the default for handed-over SQL — it requires that explicit decision. This is the easiest thing to get wrong when "build a dashboard" starts from raw SQL.
• Non-topic / raw-SQL tiles → ask about the audience, then counsel Access Boost. When a tile's query is non-topic — a populated userEditedSQL (raw-SQL tile) or a bare base-view query (no join_paths_from_topic_name) — it is invisible to Viewer / Restricted Querier roles by default. Before finalizing such content, ask the user whether the dashboard's audience includes Restricted Queriers or Viewers. If it does, advise (don't silently enable) Access Boost: it makes those tiles viewable on the dashboard (dashboard-only — not the underlying workbook) for chosen users/groups or the whole org — provided the org capability is enabled and the caller has Manager on the document. Because it loosens access controls, treat enabling it as a separate, explicitly-confirmed step: recommend the narrowest scope that satisfies the need (specific users/groups over org-wide) and get a clear go-ahead before it's applied. See the Raw-SQL tiles recipe in references/queryPresentations.md and omni-admin → Document Permissions (which carries the full confirmation checklist and the commands/prerequisite).
• --body silently wins over shorthand flags — if you pass --body, every promoted flag (--name, --summary, --branch-id, …) is ignored without warning. Put those fields inside the JSON body instead, or use flags alone with no --body.
• Draft commands take the document identifier first, then the draft identifier: v2-get-draft <identifier> <draftIdentifier> and v2-patch-draft-by-identifier <identifier> <draftIdentifier>.
• Classic-layout dashboards return 422 from every v2 endpoint — "Upgrade the dashboard to the advanced layout before editing it through the API." There is no API fallback; ask the user to upgrade the layout in the Omni UI, then retry.
• The workbook model ID rotates on every draft → publish cycle — each draft clones the workbook model (carrying extensions along), and publishing swaps the document to the clone. Never cache a workbook model ID; read it fresh from the draft's workbookModelId (omni documents list-drafts <identifier>) each time you need it.
• Interactive controls scope per-tile via map — a field/timeframe switcher's {"<tileKey>": false} excludes that tile, exactly like a filter's. See references/controls.md.
• Markdown tiles need automaticVis: false — otherwise the renderer auto-derives a chart and the tile is blank. See references/visConfig.md.
• Mustache: reference a filter under filters, not controls. To caption a tile with a filter's current value (e.g. a date window), use {{filters.…summary}} — not {{controls.<id>.…}} (a filter is configured as a control but mustache routes it to filters; the controls namespace is for FIELD_SELECTION/picker/Top-N controls and renders empty for a filter). The filters key is context-dependent: view.field in a markdown-viz tile (resolved against the tile's own query, so one token works per-tile), the control id in a dashboard text tile. Full namespace/token map + scenarios in references/mustache.md.
• query.filters needs the object form — the relative-date shorthand ("last 6 months") throws a 500; send {type:"date", kind:"TIME_FOR_INTERVAL_DURATION", ui_type:"PAST", left_side, right_side}. See references/documents-v2.md.
• A tile with no real visConfig renders as "Item missing" — the server seeds new tiles with visConfig.visType: null + automaticVis: false, which draws nothing. Every tile needs either an explicit visConfig (e.g. the table recipe in references/visConfig.md) or automaticVis: true to auto-derive one. Layout is separate: a multi-tile v2-create lays out all tiles you pass in order, so "Item missing" is a vis-config gap, not a containers gap.
• Chart rendering: Complex chart types may show "No chart available" if the inner config, visType, or prefersChart are misconfigured. If the user asks for a specific chart, include the complete chart-specific config from references/visConfig.md nested under visConfig.visConfig.config. Use chartType: "table" only as a deliberate table fallback, not for requested charts.
• Every query must include at least one measure — a query with only dimensions produces empty/nonsense tiles (e.g., just months with no data).
• Boolean filters may be silently dropped when a pivots array is present (reported Omni bug). If boolean filters aren't applying, remove the pivot and test again.
• Use identifier not id — get a document's identifier from the v2-create/patch responses or omni documents list records.
• Do not use omni unstable documents-import to update an existing dashboard — import creates a new document and may drop newly-added tiles. Use the draft flow on the existing document.
• Do not persist invalid query-level filters — if omni query run returns a server-side parsing error for a tile query filter, validate the unfiltered base query once. Do not save that broken filter into the tile. If a dashboard-level control can satisfy the request, use that path and verify by readback; otherwise leave the dashboard unchanged and report the blocker.
• Bound failed updates — if a patch returns a validation error, stop after one corrected retry at most. Do not try repeated filter syntaxes or endpoint loops. Because edits happen on a draft, recovery is clean: discard the draft (omni documents discard-draft <identifier>) and report what was preserved — the published document was never touched.

Prerequisites

# Verify the Omni CLI is installed — if not, ask the user to install it
# See: https://github.com/exploreomni/cli#readme
command -v omni >/dev/null || echo "ERROR: Omni CLI is not installed."
 
# Verify the CLI has the v2 documents commands — if not, ask the user to upgrade it

Auth: a profile authenticates with an API key or OAuth. If whoami (or any call) returns 401, hand off — ask the user to run ! omni config login <profile> (OAuth 2.1 browser flow; it blocks ~2 min on the browser). Don't run config login yourself in a headless/CI session (no browser → timeout); on a local interactive machine you may. See the omni-api-conventions rule for profile setup (omni config init --auth oauth) and discovering command and request-body shapes with --schema.

Discovering Commands

Tip: Use -o json to force structured output for programmatic parsing, or -o human for readable tables. The default is auto (human in a TTY, JSON when piped). --compact strips indentation for piping.

Commands

Build and edit documents with the documents v2-* commands — always. There is no situation where you reach back to the v1 documents create/get path to build, read, or change a document; the v2 draft flow covers all of it.

A handful of document-management operations have no v2 form — they aren't alternatives to the v2 build path, just the only command for that job: documents list / list-drafts (find documents and drafts), documents discard-draft (abandon a draft), documents delete / move / duplicate (lifecycle), documents get-queries (extract a tile's runnable query for validation), dashboards download / download-status, and models yaml-create / validate (model writes).

Dashboard Architecture

Omni dashboards are built from documents. A document's v2 state is an envelope of four slices:

• Tiles live in queryPresentations.data, keyed by record key ("1", "2", …); order is the tab order. Tile keys must be NUMERIC strings ("1", "42") — a descriptive key like "revenue_kpi" or "probe" is rejected with 400 … Invalid key in record. Track the tile↔meaning mapping in your build script, not in the key.
• Filters and interactive controls are one map: controls (see references/controls.md). Unlike tile keys, control ids here ARE arbitrary strings ("date_filter", "granularity") — don't carry that freedom over to tile keys, or vice-versa.
• Layout is the containers tree — a tile renders only where a container references it (see references/containers.md).
• Each document also has a workbook model (per-dashboard model customizations) — its ID is the workbookModelId on the document's draft record from documents list-drafts.

A document is edited through drafts: v2-patch-draft creates a draft and applies your patch; the published document is untouched until v2-publish-draft. v2-get returns the current draft state if a draft exists, else the published state. Drafts can also be bound to a model branch (see references/branch-bound-drafts.md).

Build queries on a topic

Build every tile's query on a topic whenever possible: set the query table to the topic's base view and pass join_paths_from_topic_name: <topic>, plus topicName: <topic> on the presentation (the presentation-level topicName is tile-specific — a standalone query has no equivalent). Joined-view fields then resolve through the topic's join map from the base view. For the full shape — how the join map reaches joined-view fields, the worked example, and verifying with omni models get-topic (base_view_name/join_via_map) — see omni-query's Build queries on a topic.

Access matters: a tile not built on a topic is not accessible to restricted queriers/viewers. A bare base-view query (or a raw-SQL userEditedSQL tile) still works — it traverses the global relationships file — but is restricted-access-invisible in a dashboard. Use it only when no topic fits and the audience isn't restricted, or enable Access Boost on the document so Viewer/Restricted Querier roles can see it (dashboard-only — not the underlying workbook). To author a raw-SQL tile and boost it end-to-end, see references/queryPresentations.md → Raw-SQL tiles; for the boost commands and the org-level prerequisite, see omni-admin → Document Permissions (add-permits with accessBoost, or update-permission-settings).

Authoring limit (distinct from the visibility above): if you are a Restricted Querier, you can't author a non-topic / raw-SQL tile at all — QUERY_TOPICS queries topics only — so every tile you build must be topic-based. The non-topic + Access-Boost path is only for a QUERY_FULL_MODEL author building for a restricted audience. (Check with whoami; see omni-admin → Model Roles & Caller Access.)

If no existing topic fits the request, don't just fall back to a base view (or raw SQL) — ask the user whether to extend an existing topic or create a new one, and build it on a branch only with their go-ahead. Don't silently convert un-modeled SQL into non-topic tiles. Use omni-query to choose/decide the topic and omni-model-builder to create or modify one (branch → validate → merge only on confirmation).

Document Management

Create Document (Name Only)

• <model-id> is the shared model; the server mints a per-document workbook model.
• The document is created and published immediately. The response returns only {identifier, name, description} — when you need the workbook model ID, open a draft and read its workbookModelId from omni documents list-drafts <identifier>.
• --folder-id omitted → the document lands in the creator's personal "My documents" (requires personal-content permission). Pass a folder ID to place it in a shared folder.

Create Document with Queries and Visualizations

Pass the full envelope via --body. Tiles are keyed — write tile "1" explicitly to replace the server's seed tile:

The rendering spec goes in visConfig.visConfig.config (visType beside it, spec nested under config). chartType and fields sit at the outer visConfig level. Misplaced spec keys are silently dropped on write — see the round-trip warning in Known Issues. See references/queryPresentations.md and references/visConfig.md for the structures and per-chart-type configs.

Key points:

• Tile keys are strings "1", "2", … and must appear in order to be tabs; tiles also need a containers entry to render on the dashboard. A multi-tile create auto-lays-out only tile "1" — author containers for the rest (references/containers.md).
• prefersChart must be true to render a chart; set automaticVis: false when you author an explicit vis config.
• The query needs the full collection-field set (see Known Issues) and no modelId.
• controls and settings slices can be included in the same create body — see Dashboard Filters & Controls.

To learn the exact structure for a chart type, build a reference dashboard in the Omni UI and read it back with omni documents v2-get <identifier> — remembering that the inner config reads back flat and must be re-nested under config before reuse.

Rename Document

--summary is written to the document's history audit trail. (There is no clearExistingDraft in the v2 flow — if a draft already exists, patch it directly with v2-patch-draft-by-identifier, or discard it first.)

Delete Document

Soft-deletes the document (moves to Trash).

Move Document

Use "null" as the folder path to move to root. --scope is optional — auto-computed from the destination folder.

Duplicate Document

Only published documents can be duplicated. Draft documents return 404.

Update Existing Dashboard

Edits go through the draft flow — the published dashboard is untouched until you publish, so validation happens before anything goes live:

1. Read the current state: omni documents v2-get <identifier> > doc.json.
2. Author the patch — patches merge by key, so send only the slices you're changing:
   • Add a tile: new key in queryPresentations.data + append it to order (+ a containers tile stack so it renders).
   • Edit a tile: send just that key — and re-author its inner vis config nested under config (never echo the flat GET shape back).
   • Delete a tile: set its key to null and remove it from order (and its stack from containers).
   • Layout: containers is a full replacement — send the whole tree with your edit applied.
3. Create the draft + apply: omni documents v2-patch-draft <identifier> --body @patch.json (or --body - < patch.json) — capture draftIdentifier from the response. Include a summary in the body for the audit trail.

   Prefer --body @file for these bodies: it needs no shell quoting, keeps the patch diffable, and validates the JSON client-side — a malformed patch fails immediately with the byte offset instead of a server-side 400.

4. Validate the draft — omni documents v2-get-draft <identifier> <draftIdentifier>, run the affected queries (see Validation Loops). Iterate with omni documents v2-patch-draft-by-identifier <identifier> <draftIdentifier> --body ….
5. Publish: omni documents v2-publish-draft <identifier>. On failure or abandonment, omni documents discard-draft <identifier> cleans up without touching the published doc.

Error map, merge-semantics details, and recipes are in references/updating-dashboards.md.

Updating a Dashboard's Model

First decide where a new field belongs. Don't default to the lowest-friction path — choose the field's right home, gated by what your access actually allows.

Step 0 — check your access. Run omni whoami whoami --model-id <sharedModelId> and read rolesByModel[<id>].permissions:

• QUERY_FULL_MODEL present (Querier / Modeler / Admin) → you can create a branch → prefer a shared-model branch for anything reusable. If UPDATE is also present (Modeler / Admin) you can merge it yourself (with the creator's OK); if it's absent (Querier), open the branch and request a merge — a Querier can branch and modify but not promote.
• QUERY_FULL_MODEL absent but you can still create content (Restricted Querier — has USE_WORKBOOKS) → you can't branch → use the workbook model (extension); the write is authorized against the shared model's permissions, which is exactly the restricted-but-workbook-capable case. (A Viewer lacks USE_WORKBOOKS and can't author a dashboard at all, so they never reach this decision — content creation presupposes USE_WORKBOOKS.) Scope ceiling for a Restricted Querier: their workbook-model writes are limited to .view extensions (new dimensions/measures on an existing view) + decoration edits — not .topic (topics/joins), the model/relationships files, or access grants (those need QUERY_FULL_MODEL on a branch). Keep the build view-scoped; planning a topic/join/grant change for a restricted querier 403s mid-build and can strand the doc. (Canonical: omni-admin → Model Roles & Caller Access.)

The permission→capability map behind these signals (and why a shared-model-scoped whoami doesn't list branch ability as its own entry) is the canonical access check in omni-admin → Model Roles & Caller Access — this skill just applies it. In short: QUERY_FULL_MODEL = the proxy for "can branch"; UPDATE = "can merge/promote to shared."

Then place the field. In order:

1. Can it be a calculation? A table calculation is scoped to a single query/tile (computed on the result set). Prefer one for logic local to one query — but lean to a model field (→ #2/#3) when (a) the query shape rules a calc out, or (b) you're building multiple queries at once and the same logic spans them and can be expressed as a dimension/measure. Window-shaped logic (running total, moving average, % change) should almost always stay a calc — it runs post-query on the result set, not in-warehouse; only reach for an in-warehouse field when the window must span rows outside the result set. (See omni-query's table-calculation guidance.)
2. Reusable elsewhere and you can branch (QUERY_FULL_MODEL)? Add it to a branch on the shared model and follow omni-model-builder to create, validate, and ship it — merge it yourself only if you have UPDATE, otherwise open a PR / request a merge.
3. One-off for this dashboard (and not a calculation) — or reusable but you can't branch (no QUERY_FULL_MODEL)? Add it to the workbook model — see Building a tile that queries a workbook-model field below.
4. Unsure (or reusable + can-branch, but the creator may not want a shared-model change)? Ask the creator where the field should live.
5. Never write to the schema model — it's auto-generated and read-only.

If the field isn't in the published shared model yet — it lives only on a model branch that hasn't merged — put the tile on a branch-bound draft. See references/branch-bound-drafts.md.

Push custom dimensions and measures to a specific dashboard by writing to its workbook model. Each workbook has its own model that extends the shared model — so the ID you write YAML to is a model ID, not a separate "workbook ID". Because every edit goes through a draft, and the field has to exist before a tile can reference it, the whole flow stays in the v2 draft path:

Step 1 — open a draft and read its workbook model ID:

Note: The workbook model (which extends the shared model) is what you pass to omni models yaml-create. Each draft has its own clone of it, and the ID changes on every draft → publish cycle — always read it fresh from list-drafts; never reuse a cached value.

Step 2 — POST YAML to the draft's workbook model with mode: "extension":

Critical: Always pass "mode": "extension" when editing an existing view in a workbook model. The default is "combined", which treats your YAML body as the complete view definition and marks every field you didn't include as ignored: true — silently breaking queries that depend on fields from the shared base view. Extension mode layers your new dimensions and measures on top of the inherited view.

fileName must be "model", "relationships", or end with .view or .topic. The yaml value is a YAML string (not a JSON object) containing the view's contents — no views: wrapper. Writing to a workbook model skips git sync entirely — authorization is still checked against the underlying shared model's permissions.

The YAML you write here is ordinary model YAML — don't improvise it. For the exact shape of dimensions, measures (incl. filtered measures like a count over a boolean flag), aggregate types, formats, and topics, follow omni-model-builder → references/modelParameters.md, and for filter conditions references/yaml-filter-syntax.md. A filtered measure's filters: value is always an operator object ({ is: true }, { greater_than: 100 }), never a bare scalar — is_flag: true and status: "complete" are rejected as invalid filter specs.

Building a tile that queries a workbook-model field

v2 tile queries carry no modelId — the server anchors each tile to the draft's workbook model, so the field just has to exist in that model before the tile references it. The order is fixed:

1. documents v2-create (or use the existing document) — provisions the workbook model.
2. documents v2-patch-draft <identifier> — open the draft (a --summary-only patch is enough to create it).
3. documents list-drafts <identifier> → the draft's workbookModelId.
4. models yaml-create <draftWorkbookModelId> with mode: "extension" → add the field (above).
5. documents v2-patch-draft-by-identifier <identifier> <draftIdentifier> adding the tile that references the field — no modelId anywhere in the tile query.
6. documents v2-publish-draft <identifier> — the field and tile go live together.

After publishing, the workbook model ID has changed — open a new draft and re-read workbookModelId from list-drafts before any further yaml-create.

Verify the Extension Worked

After writing, confirm the base view's fields are still available by querying one against the draft's workbook model:

(Standalone query run bodies still take a modelId — only tile queries inside v2 documents omit it.) If the response errors on a field that exists in the shared model (e.g. order_items.id), your write likely used combined mode and ignored the inherited fields. Re-run Step 2 with "mode": "extension".

Dashboard Filters & Controls

Filters and interactive controls share one envelope slice: controls: {data, order}. Each entry is {config, map} — config holds the filter/control definition, map optionally scopes it per tile. Include them at create time or patch them in later (controls merge by key like tiles):

• The keys in controls.data are arbitrary IDs and must match order.
• Filter shapes (date / string / number / boolean, hidden, required) and interactive controls (field/timeframe switchers) are documented with examples in references/controls.md.
• map scopes a control per tile: {"<tileKey>": false} excludes a tile, {"<tileKey>": "<fieldName>"} remaps it — for both filters and interactive switchers.
• A control renders only where a container places it (filter bar, sidebar, or in-tile) — otherwise it lands in the HIDDEN CONTROLS tray. See references/containers.md.
• Every filter MUST include fieldName with the fully qualified field name (no timeframe bracket for date filters), or it won't bind to any column.
• To learn exact shapes, build filters in the Omni UI and read them back with omni documents v2-get — the controls slice is directly reusable in a patch.

Document Settings

settings is a shallow-merged object: crossfilterEnabled (click a value in one tile to filter the others), facetFilters, refreshInterval (seconds, null disables), runQueriesOn ("current-page" / "all-pages" / null), and customText ({queryError, queryNoResults} overrides). Patch only the keys you're changing.

Layout (containers)

The containers tree decides where tiles and controls render: a reserved "filter-bar" stack, then one page container per page, each holding a 24-column grid of tile stacks with gridPosition {x,y,w,h}. Safe default on create: omit containers and let the server lay out tile "1", then author the full tree when you add more tiles. When editing, containers is a full replacement — round-trip the existing array with your change applied. Multi-page dashboards, page switchers, grouped bands, in-tile controls, and sizing rules: references/containers.md.

URL Patterns

After creating or finding content, always provide the user a direct link:

The identifier comes from the v2-create/patch responses or omni documents list; the draftIdentifier comes from the v2-patch-draft response or omni documents list-drafts. Replace {OMNI_BASE_URL} with the actual base URL from the active profile or environment, normalized without a trailing slash. Do not return the literal placeholder string unless credentials are unavailable and you explicitly say the URL is a template.

Validation Loops

Every dashboard build or update must be validated before publishing — broken tiles, bad field references, and misconfigured viz specs fail silently ("Chart unavailable" / "No data") with no API-level error. The full methodology — commands, the viz-spec consistency table, and the post-creation checklist — is in references/validation-and-testing.md. In brief:

1. Validate the model — omni models validate <modelId>; treat any is_warning: false issue as an error.
2. Test every query first — run each tile's query via omni query run before building (the single most important step). Check for no error, cache_metadata.num_rows > 0 (the row count; summary.row_count doesn't exist), and include the same filters you'll use on the dashboard.
3. Check viz-spec consistency — prefersChart: true; spec nested in visConfig.visConfig.config; a valid chartType with matching visType/configType; correct _dependentAxis; the stack/color dimension in query.pivots. See references/visConfig.md.
4. Verify the draft before publishing — read it back with v2-get-draft, confirm queryPresentations.data keys match order, run omni documents get-queries + omni query run per tile, and report each tile's status + row count. After one failed corrected patch, discard the draft and report the blocker.

Recommended Build Workflows

API-First (Full Programmatic Creation)

1. Discover fields — use omni-model-explorer to find topic + fields
2. Validate model — run omni models validate <modelId> and check for errors
3. Test each query — run every query you plan to include via omni query run (using omni-query) before building the dashboard. Include the same filters you plan to use as controls to confirm they parse correctly. This catches field name typos, missing join paths, bad filter expressions, and permission errors before they become broken tiles.
4. Validate viz specs — check each tile's chartType/visType/inner config/prefersChart against the consistency rules before assembling the payload
5. Create document — single omni documents v2-create with queryPresentations + controls + settings (and containers if multi-tile) in one body
6. Verify the dashboard — read it back with omni documents v2-get, confirm all tiles are present and placed, then run each tile's query via omni documents get-queries + omni query run to verify no broken tiles
7. Share the link — return {OMNI_BASE_URL}/dashboards/{identifier} to the user (only after verification passes)
8. Refine in UI — fine chart styling and pixel-level layout tweaks are still easiest in the Omni UI

Update Existing Dashboard

1. Find the dashboard — use omni-content-explorer or omni documents list to locate it
2. Read its current state — omni documents v2-get <identifier>
3. Author the patch — merge-by-key edits to queryPresentations/controls/settings; full containers tree if layout changes; re-author inner vis configs nested under config
4. Validate changes — run new/modified queries via omni query run; check viz specs against the consistency rules
5. Patch the draft — omni documents v2-patch-draft <identifier> --body … (with a summary)
6. Verify the draft — v2-get-draft, then get-queries + query run on modified tiles
7. Publish — omni documents v2-publish-draft <identifier>; on failure, discard-draft and report
8. Share the link — return {OMNI_BASE_URL}/dashboards/{identifier} to the user (only after verification passes)

UI-First (Hybrid Approach)

1. Prepare the Model — use omni-model-builder for shared fields, or the workbook-model flow above for dashboard-specific fields
2. Build in UI — add tiles, choose viz types, arrange the grid, set filters
3. Iterate via API — read the structure back with v2-get, update model fields, extract queries for reuse

Dashboard Downloads

Async render → poll → fetch. Poll the status field (in_progress → terminal error, or a success state carrying the result/URL). Early-exit on any terminal state — a fixed-count for … sleep loop that only breaks on success runs to the end if the filter is wrong or the render errors:

"Job failed to render." has two causes — disambiguate, don't assume. It's either (a) one tile in your dashboard throws at render (a markdown component handed an undefined value, a bad token) — a single bad tile fails the whole export — or (b) the render service is down. Tell them apart by exporting a known-good dashboard: if that renders, the fault is your content (find the bad tile — isolate by exporting a one-tile test doc, or binary-search tiles); if the known-good one also errors, the service is down — bound your attempts and report. Either way, while the render path is unavailable you can't PNG-verify markdown/CSS tiles, so fall back to draft read-back + per-tile query run + the UI. On success, fetch the image with omni dashboards download-file <dashboardId> <jobId>. Full catch-22 writeup: references/markdown-tiles.md.

Download a single tile (not the whole dashboard)

Add queryIdentifierMapKey = the tile's queryPresentations.data key (e.g. "5") to the download body; it renders only that tile. Same async flow (download → download-status → download-file).

Format matrix: pdf / png / xlsx / csv / json — json is single-tile only (requires queryIdentifierMapKey), and for a full dashboard csv returns a ZIP of one CSV per tile. pdf/png give the image; csv/xlsx/json give the tile's data. For the authoritative paperFormat values and the full PDF/PNG + data option set (paperOrientation, hideTitle, showFilters, formatting/row-limit flags, filename, filterConfig, …), run omni dashboards download --schema — don't rely on this list staying exhaustive.

Multi-page dashboards — the public download API renders ONE page

POST /v1/dashboards/{id}/download has no pages/pageId parameter. PNG is inherently one page, and a PDF of a multi-page dashboard returns only the default/active page — page selection exists only in the UI download form, not this endpoint (and not the schedules/deliveries API). There is no robust public-API way to render a chosen page today — tracked as an upstream API gap.

Docs Reference

• Documents API · Dashboard Downloads · Query API · Schedules API · Visualization Types
• Skill references: documents-v2.md · containers.md · controls.md · queryPresentations.md · visConfig.md · markdown-tiles.md · mustache.md · updating-dashboards.md · branch-bound-drafts.md · validation-and-testing.md

Related Skills

• omni-model-explorer — understand available fields
• omni-model-builder — create shared model fields
• omni-query — test queries before adding to dashboards
• omni-content-explorer — find existing dashboards to learn from
• omni-embed — embed dashboards you've built in external apps