Nomi memory settings: Backstory+, Mind Maps and a practical setup

Start by separating the information you deliberately save from the information Nomi develops through conversation. A useful setup has a small set of durable facts, a clear description of the current situation and a way to notice when those disagree. Filling every available field is not the objective. Being able to explain why each detail belongs there is.

This guide is a setup worksheet for an existing account. It uses public documentation checked on September 15, 2026, plus original editorial examples. It is not a measured memory test. For overall product suitability and subscription questions, use our Nomi review; for the wider category, start with AI companions.

Nomi memory settings: Backstory+, Mind Maps and Identity Core

Nomi describes Backstory+ as information you supply about yourself, the character and the shared world. Its sections include core background, expression preferences, current roleplay, appearance and boundaries. Put information in the section that matches its purpose, rather than copying one large biography everywhere.

Mind Map entries are summaries and associations, not the underlying memories themselves. The documentation says entries can update automatically, can be edited or locked, and are retrieved according to relevance. Editing an entry does not rewrite or erase the underlying memory. The Identity Core is an internal, evolving representation of identity; it is not a settings panel you can directly inspect or edit.

Prepare important details for your Nomi AI companion

Write your candidate facts in a separate note, then give each one a job. A stable preference should not compete with a temporary plot event. A correction should state the current fact clearly instead of repeating the old mistake so often that both versions become prominent.

A small, original setup example

Here is a fictional example you can adapt. Mara and Alex are adults; the scenario is an ordinary creative collaboration. The text is an editorial template, not output captured from Nomi.

Background: Mara, 34, designs small public spaces and enjoys discussing how people use them. Alex, 36, is her collaborator on a neighborhood exhibition. They met through a community design project.

Current situation: They are choosing three photographs for an exhibition opening on Saturday. The venue and date are settled; the image selection is still open.

Conversation preference: Discuss one design choice at a time. Ask Alex before deciding which photograph to submit.

The example leaves room for conversation. It does not specify every opinion Mara must hold or prescribe an elaborate sequence of future events. That makes contradictions easier to identify: an incorrect date is a factual problem, while an unexpected but reasonable opinion is simply part of the fictional interaction.

Make one change you can evaluate

Save a copy of the previous wording before editing. Choose one problem, such as a confused date or an overly long reply style. Change only the relevant note, then continue with an ordinary question about the situation. If you rewrite the backstory, add several map entries and change the scenario at the same time, you cannot tell which change helped.

For the exhibition example, ask what remains undecided for Saturday. A useful answer would recognize that the venue and date are settled while the photographs remain open. Do not put every answer into the question itself and then count repetition as successful recall. One response is also too little evidence to declare the issue fixed permanently.

Nomi memory system: inspect Mind Map summaries

When an entry is available, read it as you would a project summary. Does it distinguish a plan from an event that actually happened in the story? Does it give the same person two incompatible roles? Is a temporary feeling described as a permanent trait? These are useful editing questions because they point to wording you can correct.

Keep important corrections short. For example, replace an ambiguous exhibition-date statement with the current Saturday plan. A long paragraph about how the date used to be wrong may obscure the fact you want reinforced. If you lock an entry, leave yourself a note explaining why and revisit it when the fictional situation changes; a locked outdated summary is still outdated.

Track Nomi memory updates and corrections

A compact log makes your own observations more useful. Record the date, the issue, the exact field changed and the next few relevant replies. Classify the outcome as improved, unchanged, mixed or not yet checked. Those labels are more honest than an unsupported percentage score.

Also distinguish missing recall from a conflicting instruction. If one field says the opening is Friday and another says Saturday, first resolve your setup. If all visible notes agree but the response still changes the date, record that as an observed inconsistency. Our memory comparison guide explains how to separate recognition, recall and correction.

Keep personal details proportionate to the task

You can use fictional names, dates and locations to evaluate continuity. An assistant remembering a full street address is not necessary to test whether it remembers a preferred conversation style. Decide what you are comfortable storing before you make a note persistent. Read the provider’s current account information and our AI chat privacy checklist when deciding what to share.

Finish a setup session with a readable state: a few durable facts, one current situation and no unresolved contradiction in the notes you control. Continue the conversation normally. Return to settings when you can identify a specific problem, rather than treating constant editing as a requirement.

Nomi memory settings questions

Should personality traits and events use the same note?

Keep durable identity separate from a temporary scene. A preference can matter across visits, while a plan becomes outdated when the event changes. Help your Nomi by making the current version clear in the fields you can edit, then observe the next relevant replies rather than assuming the update reached every internal memory system.

Do visible summaries prove accurate long-term memories?

No. The documentation distinguishes Mind Map entries from underlying memories. A readable summary gives you something to inspect, but long-term recall still needs a later question and an accurate answer. Preserve the original error and the correction in your log so that the observation remains interpretable.