Objects

Memory Data

Memory objects are structured records stored in the platform’s shared memory. Each object belongs to a schema, holds real business content, and is indexed so agents can retrieve it later through memory-enabled tools.

What objects contain

Every memory object has:

Field	Description
Type	The schema this object conforms to (for example FAQEntry or ProductSpec)
Key	An optional human-readable identifier unique within the type
Content	The structured payload, validated against the schema
Embeddings	Search vectors generated automatically from the stored content
Created / Updated	Timestamps tracking when the object was first stored and last modified

Creating objects manually

1. Navigate to Admin > Memory > Objects and select Create Object.

2. Choose a published schema (type) that defines the structure for this object.

3. Fill in the object’s content fields. The form is generated from the schema configuration.

4. Optionally provide a key — a human-readable identifier for the object. If omitted, one is generated automatically.

5. Save the object. It is stored immediately and becomes available to memory-enabled workflows.

Tip

Choose meaningful, human-readable keys for your objects. Keys like vpn-reset-guide or printer-laserjet-500 are easier to reference and debug than auto-generated identifiers.

Importing from documents

For bulk knowledge ingestion, you can upload a document and let the platform extract candidate objects from it automatically.

1. Upload a document. Select the file and choose the target schema the extracted objects should match.

2. Wait for extraction. The platform analyzes the file and prepares structured candidate objects based on the chosen schema.

3. Review the extracted objects. You can inspect the proposed values, edit them, and remove entries you do not want to import.

4. Confirm the import. Accepted objects are validated against the schema before they are created.

5. Finish. Imported objects are stored and indexed automatically.

Note

Extraction quality depends on the source document and on how clearly the target schema is described.

Updating object content

To edit an existing object:

1. Open the object from the Explorer or the object list view.

2. Modify the content fields in the form editor.

3. Save the changes. The platform validates the updated content against the schema before storing.

Changes to object content automatically update the object’s search index so later retrieval uses the latest version.

Validating objects

Objects are validated against their schema at creation, update, and import time. Validation checks include:

• Required fields — all required properties must be present
• Type constraints — field values must match their declared types (string, number, boolean, etc.)
• Format rules — fields with format constraints (e.g., email, URL) are validated accordingly

When validation fails, the UI highlights the invalid properties so you can correct the data before saving or importing.

Embeddings generation

Every object is indexed automatically for semantic search when it is created or updated. In practice the platform:

1. Reads the object’s content
2. Applies the schema’s optional embedding schema if one is defined
3. Generates the search representation
4. Stores the result for later retrieval

Note

If a schema defines an embedding schema, only the fields included there are used for indexing. This helps focus retrieval on the parts of the object that matter most.

Document view

The Document View is only available for objects that originated from an uploaded document. It lets you compare extracted objects with their source material.

This is useful when you want to:

• Check whether extracted fields match the original document
• Review where a piece of knowledge came from
• Audit imported content later when quality questions come up

Deleting objects

To delete an object, open its detail page and select Delete. Deleting an object:

• Removes the object and its content from memory
• Removes all edges (graph connections) involving the object
• Removes the object’s embeddings from the semantic search index
• Removes the object’s entry from the Explorer tree (if it has one)

Caution

Deletion is permanent. Agents will no longer be able to find or reference the deleted object. Ensure the object is no longer needed before deleting.

Next steps

Schemas Define the data types that structure your memory objects.

Knowledge Graph Connect objects through typed relationships for richer agent context.

Explorer Browse and organize objects in a visual folder hierarchy.