Agent Memory
Mechanisms in AI agents for storing and retrieving information. The concept encompasses three distinct problems: maintaining state within a single session, retaining information across sessions as long-term memory, and managing the context window.
Core Facts
Timeline (last 90 days)
For unverified Agent-generated memories, a default TTL of 7 days should be set for automatic expiration, and this can be shortened further in high-risk scenarios.
Agent memory is becoming a key battleground in AI infrastructure competition around 2026.
Traditional RAG retrieves from external knowledge bases, whereas agent memory retrieval retrieves from the user's own interaction history.
An Agent's memory mechanism is typically divided into two categories: short-term memory and long-term memory, with long-term memory requiring external persistent storage such as vector databases, relational databases, and graph databases.
When multiple agents share the same long-term context, erroneous memories and information that should not be shared are amplified together, forming a positive feedback loop of memory contamination.
All Facts (5)
An Agent's memory mechanism is typically divided into two categories: short-term memory and long-term memory, with long-term memory requiring external persistent storage such as vector databases, relational databases, and graph databases.
80%UnverifiedWhen multiple agents share the same long-term context, erroneous memories and information that should not be shared are amplified together, forming a positive feedback loop of memory contamination.
60%UnverifiedFor unverified Agent-generated memories, a default TTL of 7 days should be set for automatic expiration, and this can be shortened further in high-risk scenarios.
50%UnverifiedAgent memory is becoming a key battleground in AI infrastructure competition around 2026.
50%UnverifiedTraditional RAG retrieves from external knowledge bases, whereas agent memory retrieval retrieves from the user's own interaction history.
50%