Skip to main content
Your agent gets these tools in Editor mode. It reads your current semantic layer, proposes a batch of changes, and submits them for review — nothing it does applies until a human approves it.

How an edit flows

1

Read the current model

The agent loads the relevant taxonomy, schema, and existing dimensions with sl_get_taxonomy_context, sl_get_semantic_schema, and sl_get_semantic_dimensions.
2

Propose a changeset

It composes a changeset — a batch of typed edit operations such as adding an entity, changing a join, or adding a dimension — and submits it with sl_commit_edits. The call returns a changeset_id and processes in the background.
3

Jedify validates

Every operation runs through Jedify’s guardrails without writing anything. Validation issues are recorded against the changeset.
4

Changes are staged for review

If all operations pass, Jedify stages the result as review-pending edit sessions — one per entity the changeset touches.
5

Track and approve

The agent polls sl_get_edit_state with the changeset_id to follow progress and surface any issues. A human then reviews and approves the staged sessions. Nothing takes effect until approval.
Editor mode never changes your semantic layer directly. It only proposes edits; a person approves them before they apply.

Tools

Read the model

Propose and track changes

What a changeset can contain

A changeset is a batch of edit operations. Each operation references the entities it affects by label. Relationships specify a join type (INNER, LEFT, RIGHT, FULL, CROSS, OUTER) and their join conditions (equality or BETWEEN). Semantic dimensions can be declared inline on an entity, by column.