Preparation, preview, approval
The process has three steps. First, Copilot understands the request and prepares the relevant record — an order, for example. Then it shows this preparation to the user as a preview: the user sees what is about to be written and can change it if needed. Only once the user gives explicit approval does the platform turn the record into an actual transaction. Nothing reaches Logo before that approval.
Why approval matters this much
Keeping this order — and not reversing it — is a deliberate choice. AI prepares, the person decides: Copilot's job isn't to make the decision, it's to prepare it. A record written automatically could carry an unnoticed wrong assumption straight into Logo; the preview-and-approval step puts that risk in front of the user before the record is created.
How the write actually happens
Once approval is given, the record is written through iyibir PaaS Agent, installed on your server, to Logo's own REST service. This is a separate path from reading: Copilot reads Logo data through a read-only database user, while writing goes through the Logo REST service. The two paths don't mix.
There is no deletion — Copilot does not delete a record. Every transaction you approve, and the approval itself, is kept on the platform, so the question of "who approved what, and when" can always be answered later.
A concrete example
When a user says "create an order for this customer," Copilot looks at the relevant customer, product, and stock information and prepares an order draft. That draft is shown on screen; the user can change the quantity, the product, or another field. If there's nothing to change, or once the correction is made, the user approves — and only then is the record written to Logo. If approval isn't given, the draft never reaches Logo.
The limits of this model
- Copilot doesn't write a record without user approval.
- Copilot doesn't delete any record in Logo.
- The decision always belongs to the user; Copilot doesn't replace the decision, it prepares it.
- The approval record and audit trail are kept on the platform.
- Reading and writing are separate paths: reading uses a read-only database user, writing goes through Logo's own REST service.
In the end, Copilot's write model aims to give speed and control at the same time: AI handles the preparation, the person makes the decision, and the record only reaches Logo after that decision.