Why Eleanor Has To Remember

Most AI companion demos can feel alive for one conversation. The harder test is what happens after the window closes. If a user returns tomorrow and the system has no accountable memory of what mattered, the relationship has been reset even if the language still sounds warm.

That is why Eleanor's launch checks treat memory continuity as infrastructure, not decoration. Account memory has to be written with provenance, read back in a new session without relying on the prior transcript, and kept inside user-controlled boundaries. A good answer is not enough if the system cannot show where the remembered context came from.

The launch rule is plain: Eleanor must not ship as a Pi-equivalent companion that starts over every time. Synthetic-account smoke tests check the full path before public use, including opt-out behavior, anonymous-session boundaries, and transition-ledger evidence. If cross-session recall fails, launch holds.

This also shapes the product promise. Eleanor is not claiming perfect memory or unlimited recall. She is claiming a more honest standard: remember when authorized, say when memory was not available, and leave evidence that the system did what it said.