Quick summary: Secrets AI memory changed through several connected releases rather than one launch. The system moved toward automatic extraction, reviewable changes, self-managed searching and editing, smarter ranking, deeper recall, and per-member Group Chat memory.
Automatic extraction changed when users see a save
v4.4.0 moved memory processing into the background. The release introduced automatic extraction, a popup for new memories, review controls, and clearer new-versus-updated labels. It also removed the older preference that controlled whether memory updates appeared.
That means a companion can say less about the save inside the conversation even when the system records something. The practical check is the Memory panel, not only the wording of the companion’s reply.
AI Video Gallery
A preview of selected portrait videos and AI character images.
Reviewable changes reduced silent overwrites
v4.5.0 added Memory Change Requests. When new information conflicts with an existing memory, the companion can propose an update for approval rather than changing the old entry silently. The same release described smarter merging and splitting so unrelated facts remain separate.
Users correcting an outdated fact should confirm whether the panel shows a pending change, a new entry, or two competing entries. the maintenance workflow for duplicates, importance, and outdated facts covers that distinction.
The self-managing system expanded memory actions
v5.0.0 announced that companions could search, save, and edit memories with natural requests. It also introduced a paginated view, search, and smarter ranking. Later releases improved recent recall, fixed caching, and addressed saves blocked by conflicting memories.
These release notes describe product intent. They do not mean every statement becomes a permanent or correctly ranked memory. the current explanation of automatic saves, scopes, importance levels, and Smart Mode should be used for today’s behavior.
Smart Mode and long-term recall are the current layer
The present Memory System documentation says Smart Mode uses an assistant to search memory and past conversations before a reply and is on by default per chat. It also describes improved recall from further back in conversation history and better Group Chat continuity.
A user can turn Smart Mode off for a chat, so a recall comparison should verify that setting before blaming the saved memory itself. When a fact exists in the panel but is not used, follow the checks for saved memory that is not recalled. When it never appears in the panel, use the checks for information that was never saved.
Group memory became a separate control surface
v5.8.0 added per-member Memory Groups in Group Chats, followed by fixes for conflicting saves and edit-related resets. This means memory quality now depends on both the saved content and the scope attached to the chat or companion.
What the sequence means for users
The system became more capable but also more configurable. Good maintenance now includes checking the panel, verifying scope, resolving conflicts, and testing recall with a specific question. the Secrets AI review’s assessment of memory depth across plans explains why memory access and context limits affect value. SecretOpedia’s Secrets AI memory-system update analysis tracks future changes.
The current Memory page now describes Smart Mode as enabled by default per chat and says it searches saved memories and previous conversation context before a reply. That present-day behavior should not be attributed to one early release without qualification: the system developed across several updates, and current documentation is the better source for how it works now.
The related Group Chat memory release and Multiple Chats and Memory Groups release show how scope controls expanded around the core memory system. They should be read separately because one concerns recall logic and the others concern where saved information is allowed to travel.
Official sources checked July 12, 2026: Changelog, Memory System, Multiple Chats, Group Chats.
