NiVa

Anima 3 – what changed

Three things changed in Anima 3: she sees what you send her; she does not make things up when she could not make something out; and she now has a memory.

Published September 10, 2026Updated October 29, 20265 min read
Contents
  1. What this model is
  2. What she can do
  3. 1. How she reads your messages
  4. 2. She stopped making things up
  5. 3. Memory
  6. 4. Conversation
  7. 5. Limitations

Full model

Anima 3

Full memory. Understands your photos and voice messages, searches the web.

Response time

≈12 s

Memory

full

Your photos and voice messages

understands

Limit usage

the most
Anima
Levia
Lux

What this model is

Anima 3 is Nika's full model: she remembers facts about you, the diary, the chronology and promises; she sees photos, listens to voice messages, searches the web and can message first. She sends her first message in about 12 seconds and uses more limit per reply than the other models. Available on the Pro, Max 5x and Max 20x plans, on the Free plan and without a subscription for credits. Turn it on in your profile, in the bot @NiVaAibot in the “⚙️ Настройки Ники” (Nika settings) section, or with the /anima command in Nika's DMs.


What she can do

  • Same character and way of speaking
  • Remembers facts about you
  • Keeps a diaryfull
  • Remembers details of old conversations
  • Sees photos and stickers
  • Understands voice messages
  • Searches the web
  • Sends pictures
  • Writes first
  • Reacts to messages
  • Writes in parts, with pauses

1. How she reads your messages

Telegram formatting – bold, italics, spoilers, quotes, code blocks, links with a label, premium emoji – used to be dropped, and the model got plain text:

  • “I did *not* ask” and “I did not ask” were the same phrase to her.
  • A spoiler read as ordinary words.
  • A link came through as its label. In “Tap [here]” she saw the word “here” and did not see the address.
  • Quotes and code dissolved into the rest of the text.

Now formatting comes through in full.

Premium emoji

A custom emoji used to reach her as the ordinary emoji it is drawn on, and she did not see a reaction with a custom emoji at all. Now she understands custom emoji both in text and in reactions.

Stickers

She used to not recognize static stickers. Now she sees what is on a sticker.


2. She stopped making things up

Before, if an image could not be processed, she started guessing and confidently described a photo she had not seen. Now she gets a note that the content was not perceived, along with the reason, and she will say “I couldn't make it out”.

A broken reply no longer goes out

If the model outputs a meaningless string of characters instead of a reply, that reply is detected and not sent.


3. Memory

Before, she had a diary: entries about conversations and search by similarity. Now a memory works on top of the diary.

Facts

Facts about people are extracted from conversations: what you do for work, your dog's name, what you were sick with, what you like. She recalls them herself when they fit. A fact without a subject is not saved: “he yelled at me” without a name will not get into memory.

Age of a memory

Every memory has an age: “yesterday”, “last week”, “in March”. The older it is, the rougher the precision: yesterday is remembered to the day, March to the month. Before, time was not taken into account, and she could say “just now” about something from March.

Chronology

There is now a timeline: events can be gone through in order, from old to new. She uses it to answer questions like “how did August go”.

Promises and intentions

  • a promise (“I promised to text”) – an open promise comes up both by time and by context;
  • a decision (“I decided so”) – her own position, which she then sticks to.

Outdated facts are canceled

A new fact cancels the old one. The old one stays as “I used to think that…” but no longer comes up.

Working memory

Before, what she recalled was gone by the next reply, and she had to search again. Now it is kept between replies, refreshed when you come back to it, and fades when nobody mentions it for a long time.

Different sides

When she pulls eight memories, they cover different sides of the topic and do not repeat each other. Before, all eight could be about the same storyline.

Deja vu

If an old episode that has not been recalled in a long time resembles the conversation, but not enough to come up, she gets a feeling that something like this has happened before, without the content. If she wants, she will search. More often she will just say “some kind of deja vu”.

Spontaneous recall

Once in a while, an old episode that has not been recalled in a long time comes up for no reason. She decides whether to write “by the way, I just remembered…” or keep quiet.

What happens at night

Three jobs run at night:

  1. Missed facts. Facts that were not extracted from a conversation right away are extracted again.
  2. Updating cards. For anyone with new facts, the summary “who this is and where we are now” is rewritten.
  3. A dream. Episodes from the past day and a couple of random old ones are put together into a short dream in the first person.

Dreams do not take part in ordinary recall. In the morning, a fresh dream goes into her first reply: she decides whether to tell you about it or keep quiet.


4. Conversation

  • Search through your chat. “When did I send you that recipe?” – she will find it and quote it.
  • Reactions. She can react to a message when that fits better than a reply.
  • Swipe reply. She replies to a specific message without repeating your words in quotes.
  • GIFs that fit. GIFs are picked to match the conversation.
  • Typing rhythm. She types with pauses.

5. Limitations

  • She does not message first at night. She replies to messages at any time.
  • Once in a while a reply does not arrive. Then write again.