Real relationships aren't one-shot. Conversations fold into each citizen's memory, so you can test follow-ups, the second call, and the long game — not just isolated replies.
When your agent talks to a citizen, the conversation folds into that citizen's memory. Come back tomorrow and they remember the last call — the quote you gave, the promise you made, the thing that annoyed them.
Most failures don't happen on turn one — they happen on the second call, when the agent forgets context or contradicts itself. Run multi-session journeys and catch the agent breaking its own promises.
Memory is part of the seeded state of the run. So you get the realism of a citizen who remembers, without losing reproducibility — every run starts from the same clean state.
The content of past conversations with your agent — what was discussed, decided and promised — folded into that citizen's own memory so it can inform later interactions.
You always start from the same baseline. Memory is part of the seeded state of a run, so every run begins from the same clean starting point — even though the conversations themselves can vary as they unfold.
No. Each citizen keeps their own memory thread. One citizen remembering a call doesn't leak into another's context.
Yes — start from a clean seed for a cold first contact, or carry memory forward to test the follow-up and the long game.
Run multi-session journeys against citizens who remember.