AI memory and configuration governance: keeping context trustworthy

Reusable context can help an AI system interpret an organisation's language and recurring business circumstances. That context needs ownership and review because an outdated or incorrect instruction can influence later answers.

A remembered explanation is not a transaction, approval or accounting entry. It should support interpretation without replacing the underlying records that establish business facts.

AI Memory in DNA

DNA's AI Memory interface separates creating, editing, approving, rejecting, activating, deactivating, deleting and restoring entries. The reviewed entry form carries content and related context, while the page uses distinct API operations for lifecycle changes and exports.

Active status and review status are separate attributes. An entry being visible in a list is not by itself evidence that it is approved, applicable to the current question or used in every AI answer. Those outcomes depend on the relevant service behaviour.

Make context specific

A useful entry explains a stable definition or a bounded circumstance. “Northstar's installation review refers to the pre-handover inspection” is more precise than “always assume installation is complete.” The latter can contradict real task and inspection records.

Time-sensitive context should identify its intended dates or conditions where the configuration supports them. Review it when the underlying business changes, rather than allowing a temporary exception to become an unexplained permanent assumption.

Govern configuration as a business asset

Across DNA, a sound change has a purpose, an owner, a defined scope and verification of the saved outcome. Permissions, imports, report definitions, templates and integrations each require evidence appropriate to their effect.

Keep a record of what changed and why. Verify representative workflows and unresolved exceptions before relying on the change widely. The aim is consistent business behaviour supported by evidence, not simply a collection of settings that saved without an error.