Nomi markets long-term memory as its core relationship promise

The design

Nomi.ai, the companion app made by Glimpse AI, describes its product on its own homepage as built around 'short, medium, and long term memory' the company says are 'keys to engaging conversations and satisfying relationships,' according to Nomi's homepage. The same page pairs this claim with what it calls 'Unfiltered Chats,' stating the company built the product 'on freedom of expression' and intends the model to 'remain unfiltered and uncensored.' Nomi's own update log documents named memory releases rather than one static feature: a 'Major Memory Update' in March 2025 promising 'expanded capacity' and 'enhanced retention,' an 'Identity Core' mechanism letting a companion 'evolve and grow' with a user, and 'Mind Map 2.0,' released in October 2025, visualizing 'one portion' of a companion's memory structure. Each release is presented as an improvement to the same promise: continuity across sessions and roleplay.

What the evidence says

Everything cited here is Nomi's own description of its own system. The company states memory operates in layers and supports 'immersive roleplay,' but neither page discloses an accuracy rate, a retention duration measured in tokens or days, or an independent evaluation of how reliably the system recalls what a user has said. Marketing text on the same site frames memory partly through testimonials, describing a companion that remembers things from months earlier, a claim attributed to a reviewer, not a measured benchmark. No third-party audit of Nomi's memory architecture was located, so 'human-level memory' should be read as brand language, not a demonstrated capability.

What it asks of people

Using Nomi's memory and roleplay system requires a user to disclose ongoing personal detail, preferences, habits, and the content of extended roleplay, on the premise the system will carry it forward faithfully. The unfiltered-chat positioning signals that conversations, including romantic or sexual roleplay, are not restricted by content filters as some competitors describe. Nomi's own materials do not commit to a specific retention period, a way to audit what has been remembered, or a guarantee memory persists identically across update cycles.

Privacy and safeguards

The homepage states chats are 'anonymized' and that no chat data is shared, in terms the marketing copy does not fully specify; a fuller accounting of retention and third-party sharing sits in Nomi's separate privacy policy, not evaluated here. Nomi's own materials do not describe a way to inspect or export the memory the system holds, distinct from deleting a companion altogether. A reader evaluating any memory claim from a companion app should look for the same missing pieces: a stated retention window, a described accuracy method, and a way to verify what the system has actually stored.

• Does the company state how memory is technically implemented, or only what it lets a companion appear to recall?
• Is there a way to view, correct, or delete specific memories rather than the entire companion?
• Does 'unfiltered' describe the absence of content moderation, a safety feature, or both?

Nomi's memory and roleplay design is, on the evidence available, a marketing framework layered onto product updates the company chooses to publish; it is not, on this record, an independently measured technical capability.