Start a conversation

Every chat begins from a Moment — a scene attached to a character (see Moments and scenarios). Pick one from Explore or a character page, and the chat opens with the character's opening line. From there you write back, and the story unfolds together.

This guide covers a chat end to end: how replies are shaped, the tools around the input box, models and memory, and how to edit or rewind the story. Use the list on this page to jump to any feature.

How the character replies

When you send a message, Rubii builds the reply from several layered sources — not just your last line:

Character Setting — the character's personality, voice, and rules, written by its creator. Private, and always in effect (see Create your first character).
The Moment — the opening line and the scene you stepped into (see Moments and scenarios).
Your Persona — who you are in the story (see Personas, below).
Memory — a running summary of earlier rounds plus the recent transcript (see How memory works, below).
Your latest message — your words, and any instructions you give in the moment.

The character stays in character: it follows its personality and the scene's logic, reacts to what you do, and obeys reasonable directions you give along the way.

Writing your messages

You write speech and narration together, naturally. Put speech in double quotes and wrap actions in single asterisks to make them italic:

Example — "You're late again." She closes her book and finally looks up at you.

The asterisks are just ordinary italics, not a special command — the character simply reads everything you write. Replies come back as speech bubbles plus narration blocks, so dialogue and description stay visually separated.

To send, press the round send button (or Shift+Enter). A plain Enter just adds a new line.

Steering the story — good vs. bad

You are the co-author. The clearer your intent, the better the reply.

Good — put lasting facts in the right place. Who the character is goes in its Character Setting; who you are goes in your Persona; the scene goes in the Moment. Then, in chat, just play.
Good — give in-the-moment direction. Plain requests work: "Keep replies short," "Describe the room first," "Slow the pace down." Later instructions override earlier ones.
Avoid — writing the character's mind for them. Don't dictate their feelings or decisions ("He suddenly confesses he loves you"). Set up the situation and let them react.
Avoid — relying on a detail from 50 messages ago. Older turns get summarized (see How memory works), so fine details can blur. Restate anything that must stay exact.

Input suggestions

Not sure what to say? Tap the sparkle button at the left of the input box to get three suggested replies, written from your point of view.

Input bar and AI suggestions — The sparkle button at the left of the input returns three suggested replies — tap a card to send, or the pencil to edit it first.

Each card is one possible next line or action; the three are deliberately different (for example bold / cautious / playful).
Tap a card to send it right away, or tap the pencil to drop it into the input box and edit first.
Suggestions are free and automatically match the conversation's language.
They are cached for about an hour, so the same scene returns the same three until the story moves on. A card showing only “......” is a placeholder for a missed suggestion — just tap the button again.

Editing, regenerating, and rewinding

Every message has controls; the latest reply has a few extra. Use them to shape the story instead of starting over.

Message controls and the More menu — The latest reply shows a version pager, Regenerate and Continue; the ··· menu holds Copy, Share, Edit, Rewind (Unsend on your own messages) and Delete.

On the latest AI reply (inline icons):

Regenerate — rolls a different version of the same reply. Earlier versions are kept; use the ‹ 1 / 3 › arrows to compare and choose one.
Continue — extends the same reply with more text instead of replacing it.

In the ··· (More) menu on any message:

Edit — rewrite a message's text. This only changes the saved text; the character does not reply to the edit. (In Chat 2.0 you can edit only the most recent round.)
Rewind / Unsend — the same button, named by whose message it is:
- Rewind (on an AI message) deletes everything after that message and keeps it — roll the story back to that beat, then continue.
- Unsend (on your own message) deletes that message and everything after it, and puts your text back in the box to rewrite.
Delete — removes the whole round (your message and its reply).
Copy / Share — copy the text, or select messages to share.

Rewind, Unsend, and Delete can't be undone, so each asks you to confirm.

Good: Use Regenerate when you just want a different take on the same prompt — it preserves the alternatives. Use Edit for typos that shouldn't disturb the story.
Avoid: Rewinding far back to fix one small thing — everything after that point is gone for good.

New Chat and history

Each conversation with a character is an independent session.

New Chat starts a clean slate with the same character and Moment. The character remembers nothing from your other chats — memory is never shared between conversations.
History re-opens an earlier conversation exactly where you left it. One character can hold many parallel storylines at once.

New Chat and History — New Chat starts a clean slate; History reopens any past session — memory is never shared between conversations.

Good: Start a New Chat to try a different storyline, or to reset after a thread goes sideways. Open History to continue an existing story with full recall.
Avoid: Expecting a New Chat to remember your relationship or plot from a previous chat — it won't. Re-open the old session from History instead.

Personas

A Persona is your identity in the chat: a Nickname the character calls you, plus a short description of who you are. It shapes how the character treats you; it does not change the character.

Personas are saved per Moment, with two built-in cards (Moment default and Global persona) plus any you create. The active one shows an Active badge; tap Use to switch. Full details in Using personas.

Good Nickname: a short form of address — Alex, Captain, Sis.
Good Persona: concise context the character needs — "A 25-year-old detective; the character's longtime rival who secretly cares for them."
Avoid: writing the character's lines into your Persona, or padding it with unrelated lore.

Choosing a model

Tap the model button in the chat toolbar to pick which AI powers the replies. Models are grouped into series (Rubii, Gemini, Claude, GPT, and more).

The model picker — Models grouped into series, with All and Free tabs, a per-reply price in Rubies, an Exclusive (Rubii Plus) badge, and a congestion status tag.

Rubii sits at the top and auto-routes to a strong roleplay model for you.
The Free tab lists models that cost no Rubies. A status tag — Smooth / Moderate / Congested — shows current load.
An Exclusive badge marks Rubii Plus models. Each model shows its price (Rubies per reply) or Free.
Switching models only affects future replies; it never rewrites past messages.
Good: For long stories, prefer a higher-memory series, and check the status tag before you commit.
Avoid: Assuming any model can hold your entire history — every model has a capacity limit (see Memory Enhancement, below).

How memory works: the Memo

A character doesn't keep a perfect transcript forever. As a chat grows, Rubii writes a Memo — an AI summary of older rounds, grouped into stages — and feeds it back so the character "remembers" what happened. Recent rounds are still passed in full; only the older ones get summarized.

The Brief at the top of the thread is the rolling summary; the Memo panel (in the input menu) lists every stage.
For each stage you can Expand, Jump to scene, Edit, or Regenerate.
A short chat may show No memo yet — summaries only begin after enough has happened.

The Memo panel — Each stage is an AI summary of older rounds — Expand, Edit, Regenerate or Jump to scene. The Brief at the top of the thread is the live rolling summary.

Good: Edit a Memo to fix drift — a wrong name, a forgotten promise — in plain third-person, matching the existing text. This steers the rest of the session.
Avoid: Writing instructions or first-person lines into a Memo. It is a record of what happened, not a command, and the character can get confused.
Note: editing a Memo only affects future replies, not ones you already received. (Editing requires Rubii Plus; regenerating requires Pro or Premium.)

Memory Enhancement and tokens

By default, each reply works within a fixed budget of about 24,000 tokens of context. Memory Enhancement raises that budget so the character can draw on more of the setup and history at once. It is a Rubii Plus feature and must be turned on inside Chat 2.0 (see the Enabling Chat 2.0 section below).

Memory Enhancement and the budget slider — Turn on Memory Enhancement, then raise the extra token budget — 10K to ∞, with around 40K a practical sweet spot.

What is a token? A token is a small chunk of text — very roughly a short word, or part of one. The character's setting, your persona, the Memo, and recent messages all consume tokens. Each reply has a token budget; more context means more tokens.

Is longer always better? No. Two limits matter:

Model capacity. Every model can only hold so much (often around 128K tokens). Push past it and the oldest context is dropped anyway — and crowding a model near its limit tends to lower reply quality ("dumbing down").
Cost. Extra budget is billed per token actually used above the 24K base (rounded to 1K, at the model's per-1K price). You pay only for what is used, never for the ceiling you set.

A practical sweet spot is around 40K. Most models handle ~40K comfortably; only pricier, higher-capacity models hold much more. Set the extra budget to bring the total near a model's real capacity — not blindly to ∞.

Good: Raise the budget for long, slow-burn stories, or when you see a memory ("amnesia") warning. Heavier Character / Moment / Persona text eats into the base 24K, so rich setups need more budget just to avoid forgetting early.
Avoid: Maxing the budget to ∞ on an expensive model "just in case" — past the model's capacity it's cut anyway, quality can drop, and you spend more Rubies. Pair a high-capacity model and a higher budget only when you genuinely need deep recall.

Enabling Chat 2.0

Memory Enhancement (and the newest chat experience) live in Chat 2.0, an experimental mode you turn on once in your profile:

Open your Profile (the Mine page).
In the left sidebar, find the Experimental Features card at the bottom.
In the Chat Settings row, turn on the Enable Chat 2.0 (Experimental Feature) switch.

Enable Chat 2.0 — The toggle lives in the Experimental Features card at the bottom of your Profile (Mine) sidebar.

Once it's on, the Memory Enhancement card appears in the model panel during chats. You can turn Chat 2.0 off again the same way.

A note on content and cost

Chats follow the Content Rules, and a sensitive-content filter controls what appears across Rubii. Each AI reply spends Rubies (see Rubies and Rubii Plus).