Persistent context
Persistent context
Persistent context is the information an AI system retains about a particular person across sessions so that later conversations can build on earlier ones. It is what separates a system that knows you from one that is briefed on you, and it is the substrate every relational system depends on.
Also called durable personal context, AI long-term memory, AI memory
Persistent context is often described as a memory feature, which undersells what it is. Deciding what a system retains about a person is deciding what that person is, as far as the system is concerned. A record that keeps every stated preference forever produces a system that treats a person as fixed. A record that keeps only the last few exchanges produces one that cannot hold a thread.
The hard part is not storage. It is selection, revision, and proportion: what is worth keeping, what should be allowed to fade, what should be revised when the person changes, and what should never have been written down in the first place. A system with perfect recall of an offhand remark from two years ago is not more useful than one that forgot it. It is stranger, and often worse.
Persistent context also has a legibility problem. People cannot correct what they cannot see. A memory system that operates invisibly asks a person to live with a version of themselves they have no access to and no way to amend.
Why it matters
Persistent context is the point where a helpful system becomes a permanent record. Get it wrong in one direction and the system is amnesiac and useless; get it wrong in the other and it has quietly built a dossier that the person never agreed to and cannot inspect.
Where TAICU stands
TAICU’s position is that memory should support continuity while remaining legible, correctable, and proportionate. Those three properties are treated as design requirements rather than as features to add later, because a memory system that lacks them cannot be retrofitted into having them.
Common questions
- What should an AI remember about you?
- Enough to maintain continuity and not more: the things that make later conversations coherent. The useful test is proportion — whether what is retained is warranted by what it is used for — and whether the person could see and correct it if they asked.
- Is persistent context the same as a long context window?
- No. A context window is how much text a model can consider at once. Persistent context is what deliberately survives between sessions. A very long context window with nothing carried across conversations still produces a system that forgets you.