Secrets AI Time Travel: rewind and message editing guide

Editorial note: This article uses public documentation and editorial analysis. We have not completed hands-on testing of the products discussed. Our methodology Disclosure

By Daniel Kraus

Secrets AI Time Travel rewinds a conversation by removing what happened after a selected point. Treat it as a destructive edit of the current timeline. If you want to keep both versions of a story, preserve the existing chat and start a separate conversation for the alternative.

This guide provides an original decision worksheet and a before-and-after checking method. We checked the linked official documentation on October 7, 2026; we have not performed a rewind in a paid account. For plans, media tools and the overall product tradeoffs, read the Secrets AI review.

Choose the smallest change that solves the problem

First write down what is wrong. “The last reply is too long” calls for a different intervention from “the last twenty messages depend on a decision I want to replace.” A specific problem lets you avoid deleting useful material while trying to improve a minor detail.

Your goal	Starting approach	Check before acting
Change the tone of upcoming replies	A clear out-of-character direction	Can the existing scene remain intact?
Replace an earlier decision	Review a suitable rewind point	Which later messages would you lose?
Keep two alternative stories	A separate, clearly named chat	Which continuity facts should it share?
Begin again entirely	Review the reset option carefully	Have you preserved anything you want to keep?

This table is a decision aid, not a set of additional Secrets AI features. A new chat is not claimed to be a one-click copy of the old one. Where you want the same starting situation, prepare a short setup note yourself.

Understand the documented loss boundary

The official Time Travel guide says the clock control in the chat header enters rewind mode. Selecting an earlier message opens a confirmation. Later messages are removed permanently after confirmation; the guide says the action cannot be undone. The trash control resets the whole conversation instead, including relationship progress. Time Travel is documented for Premium and higher tiers, not the free tier.

Before confirming, identify the selected message by its actual text rather than its position on a small screen. Read the confirmation against your intention. If the selected point is ambiguous, cancel that action and inspect the history again. There is no reason to commit a destructive change merely to see what it does.

Make a checkpoint note you can understand later

Keep a short private checkpoint before experimenting. Record only the material you want to retain; a complete archive is not required for every scene. Do not treat this manual note as an account export or an automatic restore mechanism.

Checkpoint field	Illustrative entry
Chat and version	Observatory mystery — original route
Last decision to keep	The adult characters agree to inspect the damaged telescope.
First decision to replace	The narrator reveals who caused the damage too early.
Later material worth saving	A useful clue description and the planned interview scene
New direction	Keep the cause uncertain and let the reader choose the next question.
Success check	The next reply presents a clue without naming the culprit.

The observatory scenario is an original fictional example. Notice that the success check describes an observable answer. “Make the story better” would be difficult to assess, whereas withholding the culprit and leaving a choice open can be checked in one reply.

If you preserve a passage, label whether it is established history, a discarded idea or a possible future scene. Mixing those categories in the next prompt can reintroduce precisely the contradiction you were trying to remove.

Treat editing an older message with the same care

The Chat Features documentation says that editing a sent message generates a new reply and removes everything after that message. It recommends starting a new chat first when you want to keep the old branch. Do not assume that changing one word in an old message preserves every subsequent turn.

Use a practical question: is the typo causing a continuing misunderstanding, or is it merely unattractive in the transcript? For an inconsequential typo, a destructive rewrite may offer little value. For a wrong place name that affects the entire story, a focused correction or a carefully planned restart may be worth considering.

Our character profile guide can help you separate stable character facts from temporary scene details before you write the replacement. Avoid rewriting both categories at once unless you intend to change the whole premise.

Keep an alternate route without merging histories

The Multiple Chats documentation describes independent message histories with the same companion and states that histories cannot be merged. Legacy Memory Groups can share continuity; Lorebooks are now listed on Premium and Ultimate (checked October 7, 2026). The same book can serve multiple chats, while a separate book starts with separate content. Sharing memory does not merge the underlying transcripts. Before a new branch, use our Secrets AI continuity worksheet to state which facts it should share and which it should not. A separate book does not erase an existing chat history; start a new chat too if that is your goal.

For our observatory example, retain “Original route” and name the new conversation “Clue-first route.” Give the new version a brief opening containing only the facts it needs. Then record whether any memory is intentionally shared. This makes it easier to tell a copied premise from an unwanted detail crossing into the new story.

A useful branch note has three lines: where the characters are, what they know so far, and what must remain undecided. Do not include the old culprit reveal in the new setup while asking the model to forget it. Start from the intended information boundary.

Try a forward direction when the history is still useful

Secrets documents out-of-character instructions using an OOC marker within delimiters. These are meant to guide the scene, clarify details or give feedback. They are not a guaranteed way to erase account data or reverse an earlier deletion.

Here is an original, non-explicit example for a story that is moving too quickly:

(OOC: Keep the observatory scene where it is. Describe one clue the characters can inspect, then stop before deciding what they do. Leave the cause of the damage unresolved.)

Evaluate the next response against those three requests. If it satisfies them, continue from there. If it only partly follows them, simplify the direction before changing several settings at once. Our scene pacing guide explains how to give each turn a clear purpose.

Check the result before extending the story

After a deliberate change, inspect the visible conversation at the new endpoint. Then run a small continuity check with ordinary questions about the scene. Keep that check separate from your long-term memory assumptions.

Confirm the chat name and the last visible event.
Check whether the intended revision replaced the unwanted decision.
Ask about one established fact, such as the current location.
Check that an undecided event is still undecided.
Record any contradiction before adding several more turns.

For example, a reply that correctly names the observatory but immediately identifies the culprit passes the location check and fails the unresolved-event check. Write those outcomes separately. This is more useful than labelling the entire memory system either perfect or broken after a single response.

If unwanted facts persist, inspect the memory interface available to your account as a separate step. Our AI memory guide explains the difference between visible chat context and saved information. For privacy-related removal, consult the privacy checklist; story editing is not proof of service-wide data deletion.

Sources and limits of this workflow

The linked Time Travel, Chat Features, Multiple Chats and OOC pages were checked October 7, 2026. Recheck the interface and confirmation text before acting. The checkpoint table, observatory scenario and evaluation procedure are editorial examples, not results of a paid product test.

Compare the controls you need in our AI roleplay buying guide. The provider visit button below is an affiliate link, explained in our disclosure; it does not change the need to verify the plan and controls yourself.