Configure semantic entity search

Where: Entities → select an entity → Edit → embedding configuration. The instance must have the required embedding service/provider configuration.

Embedding configuration chooses whether semantic search is enabled and which properties define similarity.

Choose meaningful content#

Enable embeddings only when the task benefits from semantic similarity. Select All properties or Selected properties according to the form. The primary key is excluded from embedding content. For a product description model, choose the descriptive fields that actually explain the product rather than identifiers or unrelated operational values.

If selecting properties explicitly, choose at least one. Save the configuration before requesting an embedding/re-embedding job. The editor prevents starting a job while there are unsaved configuration changes or an active job where applicable.

Inspect the job#

Review status and progress rather than assuming the enable switch has already processed records. A queued/running job is different from completed embeddings; a failed job needs its error investigated. Check the configured runtime and selected properties when work cannot start.

Query and verify#

Use the search/assistant operation appropriate to semantic similarity. Inspect returned records and compare their actual property content with the question. Similarity ranking is not a guarantee of an exact key match or a factual answer.

Use ordinary property filters for exact IDs, status equality, dates, and other precise conditions. Marking a property Searchable and enabling embeddings are separate settings with different purposes.

When changing embedded content, decide whether to re-embed existing records and verify the new job. Keep representative questions and known matching records for regression checks.