In the product this is called RAG memory management, and some screens label it “semantic search store.” Its job is to chunk and vectorize your diaries so companions can find specific passages by meaning.
If this is your first setup, start with Chat in ten minutes to confirm the switch is on, an embedding model is assigned, and diaries are in the index. This page covers parameters, migration, and troubleshooting.
| Memory rings | RAG | |
|---|---|---|
| Form | Summary documents — weekly journals, monthly reports, and so on | Vector chunks plus a retrieval index |
| Purpose | Compress your life’s timeline | Pinpoint one specific passage |
| Who uses it | Generating and copying shared memories | Companions calling the semantic search tool |
The two run side by side without conflict: summaries handle the big picture, RAG handles the details.
You can also add memory snippets by hand and then look through them with semantic or full-text search.
The screen also shows things like the number of vector chunks, the dimensions, cache usage, and the diary vector count in the current workspace.
When search comes up empty, the threshold is usually set too high.
Different embedding models generally produce incompatible vector dimensions. After switching models, always run Embedding model migration as the client prompts you to; the flow gives you backup, continue, restore, and cancel options.
Go in this order:
For the difference between companion types, see AI companions & chat.
