Data types

Data types define the structure of memory objects. Each data type is a JSON-schema definition that describes which fields an object has, how values are validated, and which parts matter for identity and semantic search. Every object in memory conforms to exactly one data type.

The Data Types page

Open Data Types at /admin/memory-types (admin sidebar, under Knowledge). The list shows each type with its Type name, Usage groups, whether it is Managed, its current Version, when it was last Updated, and its Status. Unpublished types are hidden by default — enable Show unpublished to see them. A canvas view is also available.

Managed types are provided and maintained by nara; you create your own types alongside them.

Creating a data type

Click Create Data Type and fill in the definition:

1. Name it. Set a Display Name (shown in the UI), a technical Type name (used by agents and the API), and a Description that tells agents what the type represents.

2. Build the JSON Schema. Define the object’s fields in the visual builder or switch to code view. Available property types: String, Number, Integer, Boolean, Enum, Object, and Reference. A Reference points to another data type and links objects in the knowledge graph; choose its version mode — Latest to always follow the referenced type’s newest version, or Pinned to lock it to a specific version.

3. Define the Identity Schema. Select a subset of the fields that uniquely identifies an object. nara uses it to recognize when two records describe the same thing, for example during ingestion.

4. Configure the Embedding Schema. Select the fields whose content should be embedded for semantic search, and enable Automatic embeddings to have nara compute them whenever objects are created or updated. Only embedded fields influence Query Memory results.

5. Set usage categories. Choose where the type may be used: Knowledge (stored organizational knowledge), Parameter (input to workflow agents), or Return (output of workflow agents). A type can belong to several categories.

6. Set visibility. Visible types appear in normal object views; System hidden types are used internally and kept out of user-facing lists; Hidden legacy marks retired types.

Publishing and versioning

Use Publish to make a data type available; unpublished types only appear in the list with Show unpublished enabled, and Unpublish withdraws a type again. Data types are versioned — the list shows each type’s current version, and references from other types resolve according to their Latest or Pinned mode, so you control whether dependent types follow schema changes automatically.

Each type shows a usage count, so before unpublishing or reworking a type you can see how much depends on it.