Replika’s memory controls show a useful boundary between correction and erasure

Two different controls

Replika’s memory help page says that some memories are visible in a Memory tab, that users can add memories manually, and that saved memories can be deleted. It describes this layer as working alongside a deeper system informed by conversation patterns. The companion can therefore be corrected at the fact level without necessarily resetting everything it has learned. The same page cautions that deleting memories can affect recall.

A separate deletion article draws a sharper boundary: it says there is no way to delete an entire message history without deleting the account. Its suggested alternative for specific details is the Memory section, where users can view, add or update, and delete remembered details. Account deletion is described as permanent. This is a practical control distinction: editing a remembered fact is not the same operation as erasing the conversation record behind a relationship product.

A mechanics lens

Evidence: the help pages describe Memory-tab editing and the account-deletion path. Inference: a companion’s memory UI is also a governance UI, because it gives users a way to correct an inaccurate self-description without abandoning a conversation. It is not a complete explanation of model training, backups, or all data handling. The cited pages should not be read as a privacy-policy substitute, and this note offers no legal interpretation of deletion rights.

Evidence limit: Help-center guidance may change and does not establish all backend retention or privacy practices.