The engine has no UI. This is the whole product surface: tools your agent calls over MCP. The modeling core is below.
Abridged from the engine's tool definitions, 2026-07-26. Full reference ships with launch. ratify: docs
Create and reshape tables. A table is born with named dimensions, not rows and columns to defend.
ops · create · update · duplicate
Write the default formula for a variable, once, and every cell in every dimension fills itself. Dry-run first: validation and warnings come back with nothing saved.
modes · dry_run · write
Scope a formula to a slice of the model (a quarter, a team), or define how values roll up across a dimension.
Write a value at a named address. hires[month=Oct, team=Eng] = 3. No coordinates, ever.
Add, rename, and manage the axes of the model: their items, ranges, and settings.
surfaces · items · ranges · settings
Scenarios are real objects, not copied files. Create one, snapshot it, merge it back.
ops · create · snapshot · rename · merge · edit_variant
Every write lands in the changelog with its chain. Undo reverts it. This is why a wrong assumption costs one call, not an afternoon of archaeology.
ops · undo
Read the shape of what exists: tables, variables, dimensions, formulas. The map your agent orients by.
Which workspaces your key can touch. Auth is per human, one sign-in; the agent works within it.
Tool names and operations are drawn from the engine's source. Schemas, error shapes, and the full tool set arrive with the public docs. Meanwhile: /llms.txt carries this in plain text for your agent.