All SUTRA models are built with a hierarchical memory architecture consisting of short-term working memory, a rolling session context window, and, as an add-on, Long Term Memory.

Memory Index extends the model’s capabilities to provide even more engaging experiences for users. The integration of Memory Index allows for the preservation of conversational context over extended periods. This feature ensures that dialogues are coherent and consistent, providing a seamless conversational experience. Memory Index is crucial for maintaining the continuity of multi-session conversations, making interactions with applications more natural and user-friendly.

The following is an example of a user query and response with and without LTM enabled:

With Memory Index: 

User: "Can you recommend a good book for me?"
SUTRA: "Absolutely. Since you mentioned before
        that you enjoyed '1984' by George Orwell,
        you might also like 'Brave New World' by
        Aldous Huxley. It's another thought-provoking
        dystopian novel."

Without Memory Index

User: "Can you recommend a good book for me?"
SUTRA: "Sure, 'To Kill a Mockingbird' by Harper Lee is a
       classic and highly recommended by many."

SUTRA can answer well even without memory index. But with Memory Index add-on, SUTRA's responses get enhanced with nuances and context from memory, providing rich and engaging conversations. Note how it refered to conversation from past and provided response based on previous reference.

Search IndexFreshness and Factuality