Memory

Memory Semantic Structured

What is Memory?

Memory is the platform’s long‑term storage engine. It is designed to give AI agents access to your organization’s structured business data – past interactions, support tickets, invoices, or documentation.

Unlike a traditional file system or database that only stores text, nara Memory indexes the meaning and relationships of your data, so agents can find relevant information even when keywords don’t match exactly.

Core building blocks

The Memory system is organized into three main concepts.

Data Types

The blueprints for your data. A type defines the JSON Schema that every object must satisfy before it can be saved.

Memory Objects

The actual records. Each object is linked to a specific type version and validated before it is stored and embedded.

Knowledge Graph

The relationships. Links between objects (for example relates to, depends on) form a graph that gives agents richer context.

How data flows

1. Define Data Types
   Create JSON‑Schema‑backed blueprints that describe tickets, invoices, customers, incidents, and more.
2. Write Memory Objects
   Store validated records against those types. Each write is checked against the schema.
3. Connect via the Graph
   Link objects together to express relationships and dependencies between entities.
4. Query via Agents & Retrieval
   Use agents or the Retrieval API to run semantic searches over your stored objects.

Semantic search: finding meaning

When you save a Memory Object, the platform automatically analyzes the text and generates an embedding. This allows agents to perform “fuzzy” retrieval based on intent rather than exact keywords.

Search query	Traditional search	nara Semantic Memory
“The internet is down”	❌ No results (matches “internet” only)	✅ Finds: “Network outage on Floor 3”
“Who bought the gold plan?”	❌ No results	✅ Finds: “Invoice for Premium SLA Subscription”

Use Memory with Retrieval

Memory handles storage and indexing. For semantic queries in production we recommend using the high‑level Retrieval API or agents instead of calling embeddings directly.

Common use cases

• Agent long‑term memory – store summaries of past conversations so agents don’t repeat themselves.
• Knowledge base retrieval – upload FAQs or internal wikis that agents can cite when answering questions.
• Business entity tracking – keep a synchronized record of invoices, users, or incidents that tools can reference in complex workflows.