Skip to main content
Scoping is how you make sure one user’s memory doesn’t leak into another user’s. Every memory read and write carries a scope key. Reads only see items with matching scope; writes are tagged with the current scope.

​
Why scoping matters

Without scoping, an Agent that helps Alice would also see notes it wrote about Bob. This is:
  • A privacy problem.
  • A quality problem (Bob’s context contaminates Alice’s answer).
  • A compliance problem (GDPR, HIPAA).
Scoping fixes all three with one setting.

​
Scope shape

A scope key is a JSON-ish object. Common shapes:
You choose the shape. Nora doesn’t mandate keys — it just enforces equality.

​
Setting scope at runtime

Two ways:

​
From the Trigger

The Trigger passes the scope key into the run. All memory operations in that run use it automatically. For chat triggers, the scope defaults to { conversation_id: <session> }. Override in the Trigger block config. For webhook triggers, extract from the payload:

​
Per-tool override

A tool can call memory with an explicit scope, overriding the run default. Rare — usually a smell. Use only when the Agent legitimately needs to read cross-scope data (support agent reading their own knowledge across all customers, for example).

​
Scope hierarchy

Scopes can be nested. Writing at a broader scope makes an item visible at all narrower scopes. Example:
  • Write at { tenant: "acme" } — everyone in Acme sees it.
  • Write at { tenant: "acme", user: "alice" } — only Alice sees it.
  • Alice’s reads see both her own items and tenant-wide items.
Configure hierarchy under Space settings → Scope schema — declare which keys nest under which.

​
Cross-scope access (careful!)

Some Agents legitimately need cross-scope reads — support agents, admin dashboards. Grant per-space:
  • Cross-scope read — Agent can read any scope. Writes still respect current scope.
  • Cross-scope everything — read and write. Only for admin use cases.
Both require explicit opt-in on the space. Off by default.

​
Testing scope isolation

Space settings → Test isolation. Enter two scope keys and a query. Nora runs the query under each scope and shows that no items leak between them. Fails loudly if isolation is broken. Recommended before shipping any Flow with multi-user memory.

​
Debugging leaks

If an Agent seems to know something it shouldn’t:
  1. Check the Trace’s memory ops — every read and write is logged with its scope.
  2. Look for items written under a broader scope than intended.
  3. Look for accidental cross-scope tool calls.
The most common bug: forgetting to set scope on the Trigger, so everything lands in a single global scope. Test isolation catches this.