Definition
The state management mechanism in AI Agents or RAG pipelines that captures and persists interaction history within a specific user session to provide contextual continuity. It involves a trade-off between keeping full history for maximum context and pruning or summarizing logs to minimize token latency and costs.
Distinct from 'Long-term Memory' or Vector DBs; Session Memory is localized to a single conversation thread and is often ephemeral.
"A rolling transcript on a teleprompter that the AI must read before every new response to stay in character."
Conceptual Overview
The state management mechanism in AI Agents or RAG pipelines that captures and persists interaction history within a specific user session to provide contextual continuity. It involves a trade-off between keeping full history for maximum context and pruning or summarizing logs to minimize token latency and costs.
Disambiguation
Distinct from 'Long-term Memory' or Vector DBs; Session Memory is localized to a single conversation thread and is often ephemeral.
Visual Analog
A rolling transcript on a teleprompter that the AI must read before every new response to stay in character.