omni-model-explorer

Omni Model Explorer

Explore and understand an Omni semantic model through the Omni CLI. This is the starting point — understand what exists before building, querying, or modifying anything.

Tip: Start with the Shared model — it contains the curated analytics layer.

Prerequisites

Configure the Omni CLI:

# 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."

# Show available profiles and select the appropriate one
omni config show
# If multiple profiles exist, ask the user which to use, then switch:
omni config

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 request-body shapes with --schema.

API keys: Settings > API Keys (Organization Admin) or User Profile > Manage Account > Generate Token (Personal Access Token).

Discovering Commands

When unsure what operations or flags are available:

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).

Core Workflow

Explore top-down: List models → Pick a model → List topics → Inspect a topic → Explore views and fields.

Step 1: List Available Models

Returns models with id, name, connectionId, and modelKind (SCHEMA or SHARED). Use the SHARED model — it contains the curated semantic layer.

To also see active branches on each model:

Each model in the response will include a branches array. Each branch has an id (UUID) and name — use the id as the branchId parameter in other API calls.

Step 2: List Topics in a Model

Topics are entry points for querying. Each topic defines a base view and the set of joined views available.

Returns topic names, base views, labels, and descriptions.

Step 3: Inspect a Topic

Get full detail including all views, dimensions, measures, relationships, and AI context:

The response includes:

• base_view_name — the primary table
• views[] — all accessible views, each with dimensions[] and measures[]
• relationships[] — how views join together
• default_filters — filters applied by default
• ai_context — instructions for Blobby (Omni's AI)
• sample_queries and AI field-selection metadata when configured

Step 4: Read the Model YAML

For the full semantic model definition:

The mode parameter: combined (default) merges schema + shared model; extension shows only shared model customizations.

The files map is keyed by each file's full stored path (e.g. MARTS/order_items.view), and --filename is a regex on read. Reuse that exact key — including any folder prefix — when editing with omni-model-builder; a shortened name creates a duplicate instead of editing the original.

Model Architecture

Omni has three layers:

1. Schema Model — auto-generated from your database (read-only)
2. Shared Model — analytics engineer customizations (dimensions, measures, labels, topics, AI context)
3. Workbook Model — per-dashboard customizations (ad-hoc, not shared)

When exploring, use the combined view to see everything available.

Key Concepts

Views correspond to database tables. Each has dimensions (groupable fields) and measures (aggregations).

Topics join views together into queryable units — curated starting points for analysis. A topic has a base view, joined views, default filters, and AI context.

Relationships define joins: join_from_view, join_to_view, on_sql, relationship_type (one_to_one, many_to_one, one_to_many, many_to_many), and join_type (always_left, inner, full_outer).

Field naming: view_name.field_name with bracket notation for date granularity: orders.created_at[week].

Exploration Patterns

"What data do we have about X?" — List topics → inspect the most relevant one → review views and fields.

"How do these tables relate?" — Inspect the topic's relationships[] — check join_from_view, join_to_view, on_sql, and relationship_type.

"What measures are available for Y?" — Inspect the topic containing view Y → review the measures[] array with aggregate_type and sql definitions.

Fallback: Expected View Missing from yaml-get

Use this pattern only when normal exploration comes up short — the user names a specific view and it's absent from the yaml-get or get-topic response, or a relationship references a view that doesn't appear. If yaml-get returned what you expected, skip this section.

Why it happens: yaml-get only returns views from currently-loaded schemas. If a schema is offloaded or inactive, its views won't show up. The get-schemas call surfaces all schemas the connection knows about — including offloaded and inactive ones — so it's the right next step before telling the user "not found."

Two-step recovery:

If the schema isn't in the list at all, this isn't a lazy-load issue — the connection likely doesn't have access or the schema isn't synced. Check with a Connection Admin.

Rules for --includeschemas:

• Accepts exactly one schema name per call — commas are rejected by the API. Load schemas one at a time if you need multiple.
• When set, the response contains only views belonging to that schema. Relationships are preserved even when they reference views in other schemas.
• To scope to a branch, add --branchid <id> to yaml-get or --branch-id <id> to get-schemas (the flag names differ per command — this matches the API's underlying casing).

Calculation Fields

Calculation fields in the model use a different format than regular dimensions/measures. The field key is calc_name and the expression property is sql_expression — not name/sql.

AI Context Inspection

When the user asks what Blobby knows about a topic, inspect the topic and report the actual AI configuration — do not infer it from field names:

Read ai_context, sample_queries, and AI field-selection values. Depending on CLI/API shape, get-topic may wrap these under a top-level topic object; if a top-level ai_context is null, check topic.ai_context before concluding none exists. If get-topic does not expose the full ai_fields list, read the topic YAML and report the configured ai_fields there. Include configured sample query names/prompts and the fields they exercise when present.

Field Impact Analysis

Assess the blast radius of a field migration or removal before pushing changes to dbt:

1. Create a model branch where the field is actually absent before running validator checks:

Then use the right setup for the change being tested:

• Database column deletion/rename: refresh the schema on the branch after the warehouse change is present, then validate the branch.
• Model-only field removal/rename: write the modified YAML to the branch with omni models yaml-create, using the exact fileName key returned by yaml-get.

yaml-create accepts the update as a JSON body, not separate --filename or --branchid flags:

Use the response and yaml-get --branchid readback to verify the file was written to the branch. For model-only field removal impact, remove the field's own definition from the branch YAML, but leave existing dependent field references in place unless the user's planned change also removes them. Those unresolved references are what omni models validate uses to reveal dependent measures, dimensions, topics, and joins that would break.

Do not reuse an existing branch unless yaml-get --branchid <branchId> proves the target field is absent or renamed there. A branch with a matching name but unchanged YAML is not a valid blast-radius branch.

2. Validate the branch setup before interpreting content results:

Verify the field definition precisely, not with a whole-file substring search. For example, sale_price may still appear in sql: ${sale_price} after the dimensions: sale_price: {} definition has been removed; that is a valid impact-test branch and should be reported as a dependent reference. If the original field definition still appears in branch YAML, say the branch setup is invalid and do not claim validator results represent removal impact.

3. Run the content validator against the verified branch:

This returns all dashboards and tiles with broken references to the removed field.

4. Search model YAML for additional references (run in parallel with step 3):

Search the response for the field name to find references in other views, topics, and calculated fields.

5. Report: Combine branch validation, content-validator results (broken dashboards/tiles), and YAML search results (model references) into a structured blast-radius report. Separate direct field references, cascading references through dependent fields, raw SQL column references, and saved-content breakage.

Do NOT paginate documents and check queries individually — the content validator does this for you in one call.

Docs Reference

• Models API · Topics API · Modeling Overview · Views · Topics · Dimensions · Measures
• List model schemas · Get model YAML

Related Skills

• omni-model-builder — create or modify views, topics, and fields
• omni-query — run queries against discovered fields
• omni-ai-optimizer — inspect and improve AI context on topics