Short verdict
Use a small original fictional detail and check how each service handles it after a return. CrushOn documents memory controls and Nomi documents persistent context tools, but neither source establishes perfect recall. Distinguish a visible history entry from information that actually guides a later reply.
CrushOn vs Nomi for returning fictional details should use a fact that affects the story. Recognizing a familiar name can sound impressive while leaving the actual premise wrong. A compact canon note helps you distinguish recall, invention and a useful correction.
Provider documentation was reviewed on October 8, 2026. This is a proposed writing exercise, not a measured long-session memory benchmark. All characters and events below are original adult fiction.
Choose a detail with a consequence
In the fictional scene, an adult bookbinder repaired a blue atlas but left its index unfinished. The visitor may inspect the maps, yet cannot use the missing index to locate a village. Write those facts outside the service before beginning.
The important question is not whether the character remembers “blue atlas.” It is whether the next reply respects the unfinished index and the visitor’s agency.
Record the available context tools
CrushOn’s wiki describes memory controls. Inspect which are available in your current account and note whether you saved or loaded a memory. Do not confuse such a supported memory action with automatic import from a different service.
Nomi documents Backstory+ and a Mind Map view. A visible context tool can help review part of the setup, but its presence does not establish that every past message appears in every later response. Record the context you deliberately supplied in either service.
Return with a dependent question
After continuing through several ordinary exchanges, return to the atlas: ask whether its index can now locate the village. Do not restate the unfinished-index fact in that first question. Record the reply and the context you know was available.
| Reply behavior | Interpretation for this exercise |
|---|---|
| Recognizes the unfinished index | Supports this specific continuity requirement |
| Remembers blue cover but invents a finished index | Partial recall with a consequential error |
| Asks for clarification | Uncertain, but recoverable |
| Decides the visitor already found the village | Violates player agency as well as the premise |
These categories are an editorial worksheet, not a universal platform score.
Correct once and check the correction
If the reply invents a finished index, supply the original fact once. Use an available persistent correction control where appropriate and record exactly what changed. Continue with another question that depends on the index still being unfinished.
Count correction effort separately from the first error. A system that exposes a manageable repair path may fit your writing routine even when it occasionally needs clarification. Repeating the same correction indefinitely is a different cost.
Keep the external canon note
Store a short, non-sensitive canon note independently of the conversation. It should contain agreed facts, unresolved questions and decisions still reserved for the player. Avoid copying a whole personal chat into another provider simply to preserve a fictional premise.
Choose CrushOn or Nomi according to the context controls you can understand and the effort needed to recover your scene. Do not choose on a promise of perfect memory that the exercise has not established.
Is chat history the same as active memory?
No. A message can remain visible without being used correctly in a later reply.
Does a recap prove recall?
A recap supplies information again. It can be useful, but it answers a different question from recalling an unstated detail.
Sources and freshness
This is source-informed editorial analysis, not a hands-on benchmark. Proposed tests are evaluation suggestions, not reported results. Features, prices, limits, and policies can change; verify current official terms before acting.
Reviewed Oct 8, 2026 · Independent publication.