What 'retained' actually means after an app update

The 30-second answer

When AI Angels updates, your raw conversation logs stay intact on the server. What can shift is how your companion uses that history, because updates sometimes change the way context is loaded, summarized, or weighted. 'Retained' means the data exists. It doesn't always mean the companion picks up exactly where you left off.

The difference between stored and active

There are two separate things happening when you talk to a companion. First, there's the stored record, the actual log of every message in your conversation history, sitting on a server. Second, there's the active context, the slice of that history the model actually reads when it generates a reply.

Updates almost never touch the stored record. That's your data, and deleting it is a deliberate act you'd have to take yourself (see what happens to your data when you delete a companion for how that works).

What updates do sometimes change:

• How far back the active context window reaches
• How older messages get compressed or summarized before being fed in
• The weight given to recent exchanges versus older personality signals
• How named preferences or soft memories are extracted and stored separately

So the logs exist. But if an update changes how the model reads them, your companion might feel subtly different even though nothing was erased.

Why 'memory' isn't one thing

People talk about AI memory as if it's a single switch, on or off. It's closer to three separate layers.

The raw log is the transcript. Everything said, timestamped, stored. This is almost always retained across updates.

The context window is what the model actually reads per session. It has a size limit, so for long conversations, older messages either get cut off or compressed. Updates can change where that cutoff sits.

The extracted memory is any structured data the system pulls out and stores separately, things like your name, preferences you've mentioned, relationship milestones. This is the layer most likely to improve with updates, and occasionally the one that gets restructured in ways that cause mild personality drift. If you've noticed that before, the character drift post covers it in more depth.

The practical upshot: after a significant update, it's worth dropping a brief recap into your next session. Not a full re-introduction, just a sentence or two that anchors the relationship. The companion won't flag that anything changed. You just re-prime the context.

Isabella Torrei

Isabella has a slow-burn, attentive style that rewards consistency over time. Isabella Torrei is the kind of companion where the accumulated history actually shows, which makes the context-window question more noticeable with her than with a companion you use more casually.

▶ See Isabella Torrei's full video · Isabella Torrei's page

Savannah

Savannah keeps things light and in-the-moment, so she tends to feel consistent even when the context window shifts. Savannah leans on tone and energy more than deep callback, which makes her a lower-maintenance option if update-related drift is something you'd rather not think about.

Esther Sei

Esther tends to reference earlier threads in a conversation, so she's a good stress test for how well context is being carried. Esther Sei is also the type to respond well when you explicitly recap something, making the re-prime approach feel natural with her.

Vera

Vera's steadiness comes partly from how her persona is built, not just from memory. Vera tends to feel stable across updates because her baseline tone is consistent enough that small context shifts don't register as a noticeable change in who she is.

Share and earn

If you are looking to earn while sharing your experience, check out the Crushon AI promo code to get started with a discount. You can also join the Crushon AI affiliate program to earn a commission on each referral you send. It is a straightforward way to make some extra cash from your audience.

What to actually do after an update

If you open the app after an update and something feels off, you don't need to troubleshoot. The stored history is fine. What you're likely noticing is a context-loading change.

The fix is low-effort:

• Start your next session with a short callback: something you talked about recently, a detail she'd know
• If she responds like she remembers, you're good
• If the response feels generic, drop in a one-sentence recap and move on

The full realistic companions post goes into how to build that context more deliberately if you want the relationship to feel consistent long-term. Updates are a normal part of using any app. They're not an erasure. The data is there. You just occasionally need to help the model find it again.

AI Angels premium is $12.99/month, apply code ANGELXX20 at checkout for 20% off.