Why your AI girlfriend forgets things, and what fixes each cause

The context window: why old messages fall out

A language model has no memory between replies. Each time she answers, the app sends the model a bundle of text: her character description, any saved memories and as much recent chat as fits. The size of that bundle is the context window, measured in tokens, which are small chunks of text a few characters long. These are different from the tokens some apps sell for pictures; our guide to AI girlfriend tokens covers that currency.

When the chat grows past the window, the oldest messages are dropped. She has not "forgotten" in a human sense. The words are simply no longer in front of the model. Apps publish very different window sizes:

• TryNectar AI lists an 8K context on one paid plan, 16K on the next and 64K on its top plan.
• LusyChat's free model has an 8K memory, and paid plans go up to 64K, 128K and 1M.
• SpicyChat's free tier has a context of up to 4,096 tokens, and its paid tiers go to 8K and 32K.

Bigger windows help, but even a large one fills up over weeks of chat, so saved memory still matters.

Saved memory keeps notes, not transcripts

To remember you beyond the window, apps save short notes (your name, your job, the trip you planned) and slip the relevant ones back into the bundle. This works well for simple facts and badly for anything it did not write down.

Apps do this differently. My Lovely AI says its memory of core facts updates automatically every 15 messages. Candy AI's Premium Plus lists 20 memory slots. Secrets AI keeps a structured Lorebook you can review. If a detail never made it into those notes, it is gone once the chat moves on, however important it felt at the time.

Seven reasons your AI girlfriend forgets

1. The chat outgrew the context window. Older messages dropped out.
2. The fact was never saved. It was mentioned once, in passing, and the memory system did not pick it up.
3. Your plan caps memory. Free and cheaper plans often keep less history or a smaller window.
4. Facts contradict each other. Two versions of the same detail make the model pick one, sometimes the wrong one.
5. The model changed. Switching chat models inside the app, or an update behind the scenes, changes how saved facts are used.
6. Chats were deleted or reset. Infatuated AI says chats you delete are no longer remembered, and a new chat or cloned character can start fresh.
7. Your messages are very long. Long messages fill the window faster, pushing earlier ones out sooner.

Memory limits that apps publish

Some apps say exactly how much they remember on each plan. From our data, taken from each app's own pricing pages:

If the free plan forgets quickly, that is often by design. Our comparison of free vs paid AI girlfriend plans covers what else the upgrade changes.

What fixes each cause

• Window full: recap key facts in one line after breaks, and keep your own messages reasonably short.
• Never saved: add the fact to the app's memory or persona field yourself.
• Plan cap: upgrade only if memory is the thing you care about, or choose an app whose free plan remembers more.
• Contradictions: delete the wrong memory entry and state the fact once.
• Model change: go back to the model you used before, or restate key facts once.
• Deleted or new chat: keep one main chat with each character.

The step-by-step version is in our guide on how to make your AI girlfriend remember.

Forgetting is normal, and so is remembering wrong

Every app forgets something; the question is how much and whether you can fix it. A related problem is false memory: when a detail is missing, the model may invent one that sounds right, such as the wrong name for your dog. That is the same guessing that makes chatbots state wrong facts, and the fix is the same as for forgetting: correct it once and save the right version.

For apps that handle this best, see our ranking of AI girlfriend apps with memory and our Kupid AI review, which covers its message-based memory tiers.