Skip to main content

​
What are clusters?

3ngram automatically groups semantically similar memories into clusters. Each cluster gets a human-readable label like “Authentication Architecture” or “Q2 Planning Decisions”, making it easy to browse your knowledge by theme.

​
How clusters form

Clustering runs automatically in the background:
  1. Your memories’ content is converted into numerical representations (embeddings)
  2. A clustering algorithm groups memories with similar content
  3. An AI generates a descriptive label for each cluster
  4. Clusters update automatically as you add more memories
You do not need to organize anything manually. The system finds patterns for you.

​
Viewing clusters

In the dashboard:
  • Navigate to the Clusters view
  • Each cluster shows its label, member count, scope breakdown, and last activity
  • Click a cluster to expand it and see individual memories

​
Cluster context in recall

When you recall memories, 3ngram automatically includes related memories from the same cluster. This helps surface relevant context you might not have searched for directly. 
3ngram: Recall what we decided about the database
The response includes direct matches plus cluster siblings tagged as “Cluster context”. You get a more complete picture without needing to know the exact search terms.

​
Minimum requirements

You need at least 10 non-archived memories for clustering to activate. Memories created by automated hooks (like git commit captures) are excluded from clustering by default.

​
Refreshing clusters

Clusters update periodically via a background job. You can also trigger a manual refresh from the dashboard’s Clusters view using the refresh button.

​
Pin and exclude

You can fine-tune clusters to better match your mental model:
  • Pin a memory to keep it in a specific cluster across re-clustering runs
  • Exclude a memory to remove it from a cluster’s display
These overrides persist even when clusters are recalculated, so your customizations are not lost.