Learn the vocabulary of giving an AI agent a persistent memory across separate conversations.
0 / 5 completed
1 / 5
At standup, a dev mentions giving an AI agent a persistent store of facts from past conversations, so it can recall a user's stated preference in a later, entirely separate session. What is this capability called?
Long-term agent memory gives an AI agent a persistent store of facts from past conversations, letting it recall a user's stated preference in a later, entirely separate session rather than starting from a blank slate every time. Relying only on the current conversation's context window means anything discussed is forgotten as soon as that session ends. This persistent memory is what lets an agent feel like it's building a genuine, ongoing relationship with a user across many separate interactions.
2 / 5
During a design review, the team wants the agent to periodically summarize and compress an aging conversation history into a smaller, still-useful memory instead of letting the raw history grow indefinitely. Which capability supports this?
Memory summarization and compression periodically condenses an aging conversation history into a smaller, still-useful summary, rather than retaining every raw turn indefinitely, which would eventually overflow a practical memory or context budget. Retaining every raw turn indefinitely also makes it harder for the agent to surface the genuinely relevant fact buried among a large volume of old, less relevant detail. This summarization keeps long-term memory both scalable and still useful over time.
3 / 5
In a code review, a dev notices the agent's memory store scopes each stored fact to a specific user, preventing one user's private conversation history from being recalled during a different user's session. What does this represent?
Per-user memory isolation scopes each stored fact to the specific user it came from, preventing one user's private conversation history from being recalled or leaked during an entirely different user's session. Sharing one combined memory store across every user creates a serious privacy risk where a private detail from one person's conversation could surface for someone else. This isolation is a fundamental requirement for any agent memory system handling more than one user's data.
4 / 5
An incident report shows an agent recalled and referenced a fact from a user's conversation months earlier that the user had since explicitly asked to be forgotten, because no deletion mechanism existed for stored memory. What practice would prevent this?
Providing an explicit memory deletion mechanism that a user can trigger lets them permanently remove a stored fact, honoring a request to be forgotten rather than having that fact persist and resurface indefinitely. Storing every fact permanently with no deletion path ignores a legitimate user expectation and, in many jurisdictions, a real legal requirement around data deletion. This deletion capability is an essential piece of building a trustworthy, long-term agent memory system.
5 / 5
During a PR review, a teammate asks why the team builds persistent, long-term agent memory instead of relying solely on each conversation's own context window. What is the reasoning?
A single conversation's context window is forgotten entirely once that session ends, so an agent relying only on it starts fresh every time a user returns. Persistent memory lets the agent recall a relevant fact across separate sessions, creating a sense of continuity a context window alone can't provide. The tradeoff is the added responsibility of managing that memory safely, including per-user isolation and honoring an explicit deletion request.