Skip to help content
Remembud HelpOpen journal in a new tab

Next-release help preview. These guides are being checked before publication. Availability depends on the release and account you use.

Candidate guidance

Can I trust a reflection?

A reflection is AI-generated and may be mistaken. It should use your facts without inventing a story, diagnosis or personal relationship. Treat it as something to consider, rather than medical advice or the named person’s own words. You can keep an entry without AI and report a concerning response.

Related questions

How is a reflection produced?

An AI model generates a character response from the input permitted for that request. Its response is an interpretation and may be wrong. It is not independent evidence of events in your life.

Does Remembud use the same model as Character.AI?

The researched competitor’s model description does not identify Remembud’s actual provider route or model. Those need release-specific evidence. A familiar character name does not mean a shared model or affiliation.

Can a reflection invent a source or fact?

Yes. AI may produce an inaccurate claim or nonexistent source. Check an actual cited entry or reference. A previous AI interpretation is not evidence that something happened.

Is a real person typing as the character?

The reflection is AI-generated, even if its voice sounds personal. It is not a live message from the named character, an author or a support worker. Managed encryption also must not be mistaken for a guarantee of operator-unreadable data.

What if a response drifts, repeats or is cut off?

AI can fail even when writing and saving worked. Keep the saved entry and inspect its actual result or error. A retry is a deliberate request, not an automatic quality guarantee. Report a concern without private text by default; support access needs a separate exact-record permission. No source wording can override platform permissions or prove a model is useful.

Can changing a response style or trying an alternative overwrite my history?

A new response request must preserve the original writing and prior results. A generated alternative is a separate result, not permission to erase later writing or silently change memory. Model-quality claims need the actual current recipe and human judgment, not a captured improvement percentage.

How can I tell whether a character change improved a reflection?

Use fictional cases with the same prompt and context, change one thing, and read the complete outputs for voice, grounding, useful advice and repetition. Check the saved and freshly loaded result too. Likes, repeated requests and a successful API call do not prove quality. A rating is feedback, not permission to train on your journal or a guarantee that the next response improves.

Was this guide useful?

Prepare feedback to copy or download, or review a metadata-only support request in your journal when contact is available. Nothing is sent from this guide.