How shared work stays reviewable
A local session joins Studio and MCP, with one writer and a shared record of changes.
On this page
Two access paths, one writer
Studio and MCP reach the same engine session. The engine owns the model and serialises writes so two callers do not independently save competing versions. Workspace locks also protect that model’s state. The visible History attributes changes to You, Semanticus or Your assistant.
Checks and current previews
Every DAX write is checked. Record verified edits controls an additional record, not whether the write is checked. Reviewed actions bind approval to the previewed state where supported. If work or the model moved, read it again. A successful syntax or metadata check does not prove every numeric context.
History and saved states
History records edits and their values. Undo reverses a recorded action or run at its supported boundary. Saved versions and published versions retain states for different destinations. A rollback still needs a reviewed remote write. Unknown or changed evidence remains distinct from Pass.
Local work and advice
Free runs no AI inference. Your own assistant uses your account and its own data policy. Pro can request typed advice from Jev after current permission for this model. Requests include declared metadata and typed words, not table rows, returned values, formula text or credentials. Local word matching can place some requests without sending them; it does not remove the Pro entitlement on the ask path.
Keep an isolated state folder
Technical tools can use the engine’s state-dir option or SEMANTICUS_HOME to keep sign-ins and per-user state separate. Workspace locks still prevent competing writers. Close older engine sessions before working on the same model with a newer isolated engine. Do not copy another person’s credentials to bypass a sign-in.