Long-term memory stores information that should remain available after a single task ends. It can include durable facts, preferences, decisions, project history, and reviewed knowledge that later sessions may need.
Persistence creates responsibility. Long-term records need provenance, update and deletion rules, access controls, conflict handling, and protection against stale or untrusted information becoming a repeated source of error.
ELI5
Long-term memory is the shelf where an agent keeps information that may matter again later. Unlike a temporary scratchpad, it remains available after the current session ends.
For example, a team agent might retain an approved architectural decision and the reason behind it. A future session can retrieve that decision instead of reopening the same debate from nothing.
What should go into long-term memory?
Only durable, relevant, authorized information with clear provenance and a reason to remain useful across sessions.
Why can long-term memory become risky?
Incorrect, outdated, sensitive, or malicious records can influence many future tasks if retention and review are weak.
