Definition
Selective Memory is an architectural pattern in AI agents that filters, summarizes, or prunes interaction history to retain only high-utility context. It balances the trade-off between maintaining long-term coherence and staying within the context window's token limits to minimize latency and noise.
In AI, this refers to algorithmic context management rather than the human psychological phenomenon of repressed recall.
"A highlighter pen used on a long transcript, where the agent only 'reads' the yellow-marked lines and ignores the rest."
- Context Window(Prerequisite constraint)
- Summarization(Primary compression mechanism)
- Sliding Window Memory(Simplified alternative strategy)
- Knowledge Retrieval(Complementary component for long-term storage)
Conceptual Overview
Selective Memory is an architectural pattern in AI agents that filters, summarizes, or prunes interaction history to retain only high-utility context. It balances the trade-off between maintaining long-term coherence and staying within the context window's token limits to minimize latency and noise.
Disambiguation
In AI, this refers to algorithmic context management rather than the human psychological phenomenon of repressed recall.
Visual Analog
A highlighter pen used on a long transcript, where the agent only 'reads' the yellow-marked lines and ignores the rest.