Compare documented controls and repeatable tasks separately. A large context specification and a claim of long-term memory do not establish perfect recall.
What to do
-
List the controls
Check editable facts, pinned memories, context limits and user-persona features. Mark undocumented details as unknown.
-
Use the same conversation task
Introduce a few fictional facts, continue a normal exchange and ask a natural follow-up. Record model, tier and date.
-
Report the limits
Describe exactly what happened in the sample. Do not turn a short test into a claim that a service always remembers.
A starting point you can copy
Test facts: fictional name, hobby and weekend plan. Check them after the same conversation length. Record contradictions and unsupported assumptions.
This is an instruction to try. It is not a guaranteed platform control.
Your next step
Services with a relevant documented feature
Compare these if your current service lacks the control you need. Check the stated limitation before switching.
SpicyChat
Editable Memory Manager on paid tiers; context differs by tier.