Canonical definitionRelational AI
Persistent Relational Memory is the long-term memory system that makes continuity between conversations possible.
It is different from a context window, which is temporary, and from retrieval-augmented generation, which is document-centric. Relational memory is person-centric and time-aware.
It stores what matters to a specific person, how it evolves over time, and how it should be surfaced when relevant.
What it is not
- —Not a longer context window.
- —Not RAG over a person's transcripts.
- —Not a log — memory is curated, weighted and allowed to change.
In practice
Knowing someone's sister is called Marta is retrieval. Knowing that the relationship changed in March, and when to bring it up gently, is relational memory.
Frequently asked
- How is this different from RAG?
- RAG answers 'which document is relevant to this query?'. Relational memory answers 'what matters to this person, how has it changed, and should it surface now?'.
- Can a person see and delete their memory?
- Memory that cannot be inspected or removed is not trustworthy. Control over memory is part of the design.
Related terms
Cite this definition
Almaia. "Persistent Relational Memory." almaia.tech/relational-ai/persistent-relational-memory.