Skip to main content
Help & Documentation

Edit a dataset's semantic model

Attach a semantic model to a dataset in Prepare — from a YAML file or a connected warehouse — to apply field descriptions, synonyms, and metrics, while your own edits always win.

3 min readUpdated October 10, 2026

Every dataset in Prepare has a Semantic model tab, beside Table view and Data Dictionary. A semantic model layers field titles, descriptions, synonyms, and reusable metrics onto the dataset's columns, so the AI surfaces and anyone browsing the data read the same governed meaning. Your own edits always win over anything the model sets.

Open the tab

  1. Open the Prepare view of your dataset.
  2. Switch to the "Semantic model" tab.
  3. If the dataset isn't saved yet, save it first — a semantic model attaches to a saved dataset.

The model is written as a YAML file (the OSI format). The tab opens on an example you can replace, or on the model already attached. Your edits are checked as you type: the editor marks the line of a parse error, and once the file is valid it says so and offers to save.

Attach a model from a file

  1. Type or paste the model into the editor, or click "Upload file…" and choose a .yaml, .yml, or .txt file.
  2. Click "Validate" to check it against the dataset without saving — the report lists what it matched and any warnings.
  3. Click "Save" to attach it. "Discard changes" clears an unsaved draft.
Note: Relationships (joins) defined in the file aren't applied yet — only field descriptions, synonyms, and metrics reach the dataset.

Pull one from the warehouse

When a dataset is backed by a Snowflake or Databricks connection, Golden can offer the warehouse semantic view or metric view that already describes the table. If one is found, a banner asks whether to use it — pick it and click Apply, or Not now to dismiss it. Warehouse column comments and tags also flow in from the connection's settings.

A pulled model is shown as read-only YAML, with its warehouse source and when it was last updated. Click "Re-sync" to pull the latest definitions again. "Edit as YAML" turns it into an editable file — which stops re-syncing, since the model is then your own. Pulling and re-syncing need permission to manage the connection; without it, those actions are withheld and the tab says so.

Note: A dataset imported directly as a Snowflake semantic view or Databricks metric view already is a model, so its Semantic model tab explains that and shows no editor — see Work with semantic views.

Remove a model

"Remove" clears the model's descriptions, synonyms, and metrics from the dataset. Your own edits are kept.