Skip to content

Context scopes

A context scope is a governed grouping of entities, metrics, business rules, and findings for one team, domain, or workstream — Revenue, Build, Retention. It is simultaneously an organisational unit, a boundary on what the AI can reason over, and a unit of access control.

The instinct is to give the agent the whole schema. It does not work well.

Whole-schema context is noisy — most of it is irrelevant to any given question, and more context means more chances to pick a plausible-looking wrong table. It is expensive, since every irrelevant column is in every request. And it is a governance hole: a person asking about revenue should not be able to read the HR schema because it happened to be in scope.

A scope fixes all three. The agent works from one domain’s context, so answers are tighter, cheaper, and confined to what that team is responsible for.

  • Entities and metrics assigned to it.
  • Business rules and investigation findings — the living business knowledge for that domain. See grounding in practice.
  • Owners — up to three people who are responsible for it.
  • Featured metrics — the ones surfaced on the scope’s board.

Only a label is required. Add a description, a colour, and up to three owners. See scopes for the walkthrough.

Every person has a primary scope, and it is what makes the app feel personal:

  • Home shows that scope’s summary rather than an empty page.
  • Chat opens in that scope’s context.
  • Scope board defaults to it.

Without a granted scope, someone logging in for the first time sees an empty product with a note to ask an admin. Admins grant scopes under Admin → Users; people can change which of their granted scopes is primary in Settings.

You can bookmark scopes to pin them to the top of your scope list. It is a personal convenience — it does not change access or defaults.

A scope holding entities or metrics cannot be deleted. Reassign its content first. This is deliberate: deleting a scope should never silently orphan the catalogue work attached to it.

They look a bit like folders and behave more like permissions. A folder hides things; a scope governs them — it decides what the agent treats as in-context, what a board shows, and who can see the area at all.