Skip to main content
Index memory is a memory space that holds facts about entities and the relationships between them, for the Agent to look up when it needs them. An entity is anything you can name: a customer, an order, a product. Use it for content where short facts pile up per entity. It comes in two index forms, Vector and Graph.

​
When to use it

  • Facts accumulate per entity when there’s a growing set of things to know about each entity, like customer profiles, product specs, or order history
  • Keeping only the facts from long records when you want to pull just what matters out of call transcripts, emails, or chat logs
  • Answers that require following relationships use the Graph form when a question takes several hops, like “who else is at the same organization as this customer?”
For content read start to finish, like procedures or guidelines, use Wiki memory instead.

​
Index forms

  1. Vector facts are embedded, indexed as vectors, and searched by semantic similarity. This is the default when a memory space is created.
  2. Graph facts are stored as (subject, relationship, object), linking entities to each other through relationships.
Despite the similar name, this is not the causal graph. The causal graph is a separate feature used for answer verification and cause-effect analysis; the Graph form is memory holding the entities and relationships accumulated through runs. Set it under Index form in the Settings tab, choosing Vector only or Graph.

​
Facts tab

The tab that appears when the index form is Vector only. Accumulated facts are shown as a list.

​
Fact list

  • Find facts with the search box at the top.
  • Each row shows the summary, a content preview, a kind badge, and a recurrence count (×N).
  • Facts with a higher recurrence count come first.

​
Kind badges

Each fact carries a badge for what kind of item it is — fact, preference, profile, entity, episodic, event. Hover a row and an × button appears on the right. Click it to exclude that fact from memory. The Agent no longer searches or references an excluded fact. See Forgetting for details.

​
Graph tab

The tab that appears when the index form is Graph. Accumulated entities and relationships are shown as a graph.

​
Exploring the graph

Click or drag nodes and edges to explore.

​
Entity merging

Items that refer to the same thing under different names should collapse into a single node. Nora merges items automatically when their IDs match or an alias connects them. To merge by hand, select several nodes and click Merge.

​
Log tab

Review when items were created or updated, in chronological order. Select a log entry to inspect the change.

​
How memories accumulate

Facts are created through three paths.
  • Automatic saves When an Agent run ends, entities, facts, and relationships are extracted from the conversation and documents per the policy prompt and saved. Most facts are created this way.
  • The Agent saves explicitly You can specify what to save in the Agent prompt.
  • Written via the CLI Run a script with the CLI. Useful for loading existing material in bulk or migrating memory from another system. For CLI details, see nora memory.