CHAIT User Guide

Everything you need to create characters, build scenes, and hold rich group conversations with AI personalities.

๐Ÿš€Getting started

The left sidebar is your home base. It lists your recent chats and has buttons for Chat, Characters, Scenes, Community, and Personas. At the bottom are Settings and Sign in / Sign out.

A quick path to your first conversation:

  1. Open Settings โ†’ AI Providers and add an API key for a provider (or connect a local one such as Ollama or LM Studio). See Settings.
  2. Open Characters and create at least one character.
  3. Go to Chat, press New Chat, pick a scene (or Free Chat), pick your characters, and start talking.
Bring your own AI. CHAIT doesn't include AI usage. Replies are generated by the provider you configure, using your own API key or a local model. Provider costs, if any, are billed by that provider.

Signing in with Google is required for the Community Hub. Your characters, chats, and personas are stored locally on your computer.

๐Ÿ’ฌChat

Starting a chat

Press New Chat at the top of the chat view. It's a two-step window:

  1. Choose a scene or Free Chat.
  2. Choose characters. You need at least one.

Naming your chat. On the second step, type a name into the Custom chat title (optional) box next to the Start button. That name is what appears in the recent chats list in the sidebar. The name is saved when you send your first message.

If you leave it blank, scene chats are named after the scene plus the date (for example, "Haunted Manor - 10/22/2026") and free chats are named "New Chat". The title is set when the chat is created, so choose it before you start.

Commands, shortcuts & formatting

InputWhat it does
EnterSend your message.
Shift + EnterAdd a new line without sending.
/narratorType this alone as your message to make the scene's narrator speak right now, regardless of its trigger setting. It only works in a scene chat with the narrator enabled. See Narrator.
Character name in your messageMentioning a character by name gives them priority to reply next. See Who replies.
/compileType this alone as your message to save a summary of the conversation so far. Every character in the chat gets its own copy as a high-importance memory, and each one replies with the summary. It needs memory turned on for the character and is part of the full unlock. See Memory.
โœจ Sparkles buttonAppears next to Send in the message box. Drafts a reply for you as your active persona, which you can edit before sending. It uses your persona's AI provider and model, or your default chat model if the persona has none (see Personas). If you have no active persona, it opens the Personas screen instead.

Text formatting. Messages support light inline formatting:

  • **bold**, *italic*, ***bold italic***
  • ~~strikethrough~~
  • `inline code`

Line breaks are kept. Links and HTML are shown as plain text.

Editing a message. You can copy any message and edit your own. When you save an edit, the AI replies again to your revised message.

Who replies

In a group chat, CHAIT decides which characters respond to each message of yours:

  • Characters you mention by name are always prioritized.
  • Others are chosen by the Group dynamics mode in Settings โ†’ Chat:
    • Natural (default): mention priority with a bit of randomness.
    • Round robin: strict rotation through every character.
    • Balanced: favors the character who has spoken the least.
Tip: To direct the conversation, address a character by name: "Aria, what do you think?"

Managing chats

  • Your recent chats are listed in the sidebar. Click one to reopen it exactly where you left off.
  • Use the trash icon on a chat in the sidebar to delete it.
  • Press New Chat any time to switch scenes or characters.

Good to know

  • Errors show as a banner at the top of the chat. Common causes are a missing or invalid API key, a local server (Ollama/LM Studio) that isn't running, a provider rate limit, or an exhausted provider balance. Check Settings โ†’ AI Providers and use Test.
  • Context window. Characters only see a limited amount of recent conversation each turn. Older events are remembered only if memory is on.
  • CHAIT limits how fast AI requests can be sent to protect against abuse. If you hit it, wait a moment and try again.

๐ŸŽญCharacters

Open Characters from the sidebar to see your library. You can search by name or personality, sort by recent or name, filter by tags, and create, edit, or delete characters from there. Deleting a character permanently removes it along with its memories.

Creating & editing

Press the create button, or open an existing character to edit it. The editor has these tabs: Core, AI, Memory, Voice, and Relations (shown once the character exists).

FieldDetails
Name (required)Up to 50 characters.
Age (required)Must be 18 or older. Younger characters are not allowed.
Sex/GenderOptional: Prefer not to say, Male, Female, Non-binary, or Other.
AvatarOptional image: JPEG, PNG, GIF, or WebP, up to 5 MB.
ColorUsed for the fallback avatar (the first two letters of the name) and for the character's chat bubble.
AppearancePhysical look and style. Up to 1,000 characters.
Personality (required)Traits, voice, interests, and how the character interacts. Choose a size limit of 500, 1,000, or 2,000 characters.
Background / LoreBackstory and world context.
TagsType a tag and press Enter or Add. Tags help you filter your library and help others find the character in the Community Hub.
Speech style (Voice tab)How the character talks. Up to 300 characters.
Example dialogues (Voice tab)Sample exchanges that teach the AI how the character speaks. Separate each exchange with a blank line.
Tip: Concrete details beat adjectives. "Answers questions with questions and fidgets with a pocket watch" gives the AI more to work with than "mysterious."

AI model & context

On the AI tab, a character can use your default chat model or its own:

  • Pick a provider (OpenAI, Anthropic, OpenRouter, Google Gemini, Ollama, LM Studio) or one of your custom model presets, then a model.
  • Temperature (0โ€“1.5): lower is more predictable, higher is more creative.
  • Response length (50โ€“500 tokens): the maximum length of a reply.
  • Context window (1,000โ€“32,000 tokens): how much recent conversation the character sees each turn. This is not the reply length. A larger window means better recall of the recent chat but more cost and slower replies.

If you pick a custom model preset that sets its own context window, the preset's value is used instead.

Memory & personality growth

With memory turned on, a character remembers important facts, preferences, and events across chats.

  • How it's created: CHAIT automatically summarizes the conversation every 4, 6, 8, or 10 messages (your choice on the Memory tab).
  • How it's used: Each reply draws on the most relevant stored memories, typically up to 8, and more when a conversation was especially important.
  • Save a memory on demand: type /compile in a chat to summarize the conversation so far. Every character in the chat (with memory on) stores the summary as an important memory. Use it before ending a long session or after a big story moment.
  • Memory viewer: Open a character's memory to see each memory's type, content, and importance, plus relationship stats. You can refresh, edit, delete individual memories, or clear them all.
  • Per character: Every character has their own memories and perspective. Characters can also remember their relationships with each other.

Personality growth. Roughly every 50 messages, CHAIT drafts how the character has developed based on their memories. The draft isn't used until you apply it. Applying merges it into the personality, and the previous version is kept so you can undo it. You can also press Update now once the character has at least three new memories.

Availability: Persistent memory, the memory viewer, and personality growth are part of the full unlock. Basic characters, chats, and relationships are available without it.

Relationships

On the Relations tab, link a character to other characters or to your personas. Choose a type (friend, family, romantic partner, rival, coworker, mentor, or custom), add context, and set values such as trust, familiarity, and emotional bond. Characters use these in group chats to behave authentically toward each other.

๐ŸŽฌScenes

A scene is a setting for your story: a place, a mood, a starting moment, and optionally a cast and a narrator. Open Scenes to see your library, where you can search, sort by recent or name, edit, publish, or permanently delete a scene.

Creating & editing

FieldDetails
Name (required)Up to 50 characters.
Story descriptionBackground and context shared with every character. Up to 2,000 characters.
Opening message (required)The first message of the chat. Up to 500 characters. See placeholders below.
AtmosphereMood and feel, for example "rainy, tense, quiet". Up to 100 characters.
Background imageUpload an image and choose Show background in chat. The image appears behind the Active characters panel on the right side of the chat, not behind the messages themselves.
CharactersLock specific characters into the scene. They're pre-selected whenever you start a chat in it and can't be removed.
Public / PrivateScenes are private by default.
NarratorSee below.

Placeholders in the opening message. The text is filled in when the chat starts:

  • [user]: your active persona's name
  • [char], [char2], [char3]: the first, second, and third characters in the chat
  • [Name]: a specific character by name, for example [Aria]

Anything unrecognized is left as written.

Narrator

The narrator is an optional storyteller who adds scene-setting messages, such as events, atmosphere, and transitions, without being one of your characters. Narrator messages appear in the chat as their own message type.

Turn it on in the scene editor's Narrator tab and choose when it speaks:

ModeWhen the narrator speaks
ManualOnly when you ask, using /narrator.
Auto intervalEvery N messages. You choose N from 3 to 15.
Action-basedOn key moments in the story. This is the default.
Scene changeWhen the scene shifts.

You can also customize the narrator's voice or personality (up to 200 characters), its temperature (0โ€“1.5), its max reply length (50โ€“200 tokens), and whether it uses your default chat model, a provider/model of its own, or one of your custom model presets. In any mode you can type /narrator to make it speak immediately.

๐ŸŒCommunity

The Community Hub is where players share characters and scenes. You must be signed in (Google) to browse or share.

Browsing & importing

  • Switch between the Characters and Scenes tabs.
  • Search by keyword, and filter characters by tag.
  • Sort with the dropdown: Popular (most imported first), Trending (most viewed first), or Recent (newest first). Ties are broken by newest.
  • Open an item to see its creator (shown as "by @name"), favorite characters, and read or write comments.
  • Press Import to add an item to your library.
Imports are independent copies. You can edit them freely, and they don't change when the original author updates or removes theirs. An imported character keeps its personality, settings, avatar, tags, and memory on/off setting, but starts with no memories and no relationships. An imported scene keeps its name, description, opening message, atmosphere, and background, but not its locked characters or narrator setup. Any field the creator hid is replaced with a placeholder that you can edit and fill in yourself.

Publishing

  1. Open your character or scene in its library and choose Publish.
  2. Review the publish window. You can hide fields so importers can't see those details: for characters, personality, appearance, and background; for scenes, description, opening message, and atmosphere.
  3. Confirm. Published characters may be reviewed by moderators before they appear in the hub.

To change a published item, open the publish window again and update it. Choose Unpublish to remove it from the Community Hub. Your own local copy is untouched.

Privacy & safety

  • Everything is private until you publish it.
  • Publishing a character checks the name, personality, and description for profanity. Flagged content is rejected, so edit it and try again.
  • All characters must be 18 or older.
  • See something that breaks the rules? Use Report and choose a reason. Details are optional.
  • Your chats and memories are never published. Only the character or scene itself is shared.

๐Ÿ‘คPersonas

A persona is you in the story: your name, personality, and style. Characters adapt how they talk to you based on your active persona. Open Personas from the sidebar to manage them.

Creating & editing

FieldDetails
Name (required)1โ€“50 characters. This is what characters call you.
Personality (required)At least 20 characters, up to 500.
Communication styleHow you talk.
InterestsWhat you care about.
Avatar & colorJPEG, PNG, GIF, or WebP up to 5 MB. If there's no image, the color is used for your avatar and chat bubble.
AI provider / model, temperature, max tokensOptional. Used by the โœจ button in chat, which drafts messages for you as this persona. If left empty, your default chat model is used.

Setting the active persona

You can have as many personas as you like, and one is active at a time. The active persona applies to all your chats. There's no per-chat selection, so switch personas before you start a conversation. Your first persona is activated automatically. You can activate or deactivate personas from the Personas screen.

You can't delete your only persona.

โš™๏ธSettings

Open Settings with the gear icon at the bottom of the sidebar.

AI Providers

ProviderWhat you need
OpenAIAn API key.
AnthropicAn API key.
OpenRouterAn API key. Gives access to many models from one account.
Google GeminiAn API key.
OllamaNo key. Ollama must be running on your computer. Default address http://localhost:11434.
LM StudioNo key. Start LM Studio's local server. Default address http://localhost:1234.
  • For key-based providers, you can save, test, replace, or remove your key. Keys are stored locally on your computer and never sent to our servers.
  • For local providers, set the address if it's different from the default and press Test. CHAIT discovers your installed models automatically.

Models: custom model presets

A custom model preset is a saved recipe for how an AI should behave: which provider and model to use, how creative it is, how long its replies can be, and any extra instructions. Build it once, then pick it for any character or persona instead of setting everything up each time.

Open Settings โ†’ Models and press New Preset to create one, or see your existing presets. Each one is listed with its provider, model, and key settings. Use the pencil icon to edit it and the trash icon to delete it (you'll be asked to confirm).

Connect the provider first. A preset only works if its provider is set up under Settings โ†’ AI Providers (a valid API key, or a running Ollama / LM Studio server).

Preset fields

FieldWhat it doesRange / default
Display name (required)The name you'll see when choosing the preset.Free text
IdentifierInternal lowercase name, generated from the display name.Auto
Provider (required)OpenRouter, OpenAI, Anthropic, Google AI, Ollama (local), or LM Studio (local). Changing it clears the chosen model.Default: OpenRouter
Model (required)The exact model the preset uses.Pick from the list or type the ID
DescriptionA short note to remind you when to use this preset.Optional
System promptExtra instructions added to the character's normal instructions every time the preset is used. Good for style rules such as "Always write in third person" or "Keep replies under three paragraphs". It adds to, rather than replaces, the character's own personality.Optional
TemperatureHow creative the model is. Low is focused and predictable; high is varied and surprising.0 to 2, default 0.8
Max reply lengthThe longest a single reply can be, in tokens (roughly ยพ of a word each). It doesn't limit how much history the model sees.50 to 32,000, default 500
Context windowHow much conversation history the model can see on each turn. Set it to the model's real limit (for example 128000). When a preset has this value, it takes priority over the character's own context setting.1,000 to 2,000,000, optional
TagsComma-separated labels such as creative, fast. They appear on the preset in your list.Optional

Advanced parameters

Expand Advanced Parameters for fine-grained control. They're all optional. Each slider has a Reset link, and leaving one unset means the model's default is used. Not every provider supports every option, and unsupported ones are ignored.

ParameterEffectRangeWorks with
Top PLimits word choices to the most likely ones. Lower means safer, narrower wording.0 to 1All providers
Frequency penaltyDiscourages words that have already been used a lot, reducing repetition.-2 to 2OpenAI, OpenRouter, LM Studio
Presence penaltyEncourages the model to bring up new topics rather than staying on the same ones.-2 to 2OpenAI, OpenRouter, LM Studio
Repetition penaltyValues above 1 reduce repeated phrasing. Below 1 allows more repeats.0.5 to 2Ollama, OpenRouter
Stop sequencesText that makes the model stop writing as soon as it appears. Enter several, separated by commas, such as [END], ###, User:.Comma-separatedAll providers
Tip: Change one setting at a time and test it in a chat. If you're unsure, leave the advanced parameters alone. Temperature and max reply length cover most needs.

Using a preset

  • Characters: open the character's AI tab, choose Custom Presets as the provider, then pick your preset.
  • Personas: the persona's AI settings offer the same choice. A model is needed for the โœจ draft button in chat.
  • Scene narrator: in the scene editor's Narrator tab, choose Custom Presets as the narrator's AI provider.
  • Default chat model / orchestrator: under Settings โ†’ Chat, the orchestrator provider also offers Custom Presets.
  • The preset's temperature, max reply length, advanced parameters and context window replace the character's own values. If the preset leaves an advanced parameter unset, the character's value is used.
Deleting a preset that a character, narrator, or the orchestrator uses means replies will fail with an error until you choose another model for it.

Chat

  • Default chat model: used by any character, persona, or narrator without its own model.
  • Group dynamics: Natural, Round robin, or Balanced. See Who replies. Advanced group dynamics may require the full unlock.
  • Orchestrator model: the AI used to plan who responds in group chats. Choose the same provider as the character, a specific provider and model, or one of your custom model presets.

Data

  • Back up your data: downloads a ZIP of your characters, chats, personas, and uploaded images.
  • Restore from backup: loads a ZIP and replaces your current local data, then reloads the app. Back up first if you want to keep what you have.

Account

Use the sign-in / sign-out button at the bottom of the sidebar. Signing in is only needed for the Community Hub.

๐Ÿ› ๏ธTroubleshooting

ProblemTry this
Characters don't reply, or an error banner appearsCheck Settings โ†’ AI Providers and press Test. Confirm the key is valid and your provider account has credit.
Ollama / LM Studio shows no modelsMake sure the app is running with its local server on, and that the address in Settings matches.
Replies are cut offRaise the character's response length (AI tab) or the preset's max reply length.
A character forgets earlier eventsRaise its context window, and turn on memory so key events are kept long term.
Community Hub asks me to sign inUse the sign-in button at the bottom of the sidebar.
Can't publish a characterThe name, personality, or description may have been flagged. Edit them and try again.
/narrator does nothingIt needs a scene chat with the narrator enabled on that scene.

๐ŸงญGuides

Step-by-step walkthroughs that combine a scene, a character, and a custom model preset into one self-contained, reusable world you can drop your protagonist โ€” you โ€” into.

The pattern: scene + population character + custom model

Three pieces work together to turn a single character chat into a whole living world:

  • The scene is the world/story itself โ€” its name, description, atmosphere, and opening message set the stage, and its narrator paces events as the story unfolds.
  • A "population" character stands in for the world's residents rather than a single person โ€” the townsfolk, NPCs, and ambient voices you actually talk to. It's a regular character; nothing in its setup marks it as special, only how you write its personality.
  • A custom model preset holds the world's canon โ€” history, rules, factions, tone โ€” in its system prompt. Link the same preset to the scene's narrator and to the population character, and both speak from one shared set of world rules. You (through your persona) are the protagonist; you don't need a dedicated character for yourself.
How world info reaches you. A custom model preset's system prompt is only injected into whichever entity is using that preset โ€” here, the narrator and the population character. You don't read it directly; you learn the world the same way the protagonist would, by hearing it from the narrator and the people in it.

Example: an anime isekai world

A complete example you can build and adapt โ€” a classic "summoned to another world" setup, using a town called Brindlemere in a fantasy world called Aeloria.

1. Create the custom model preset ("World Rules")

Go to Settings โ†’ Models โ†’ New Preset and fill in:

FieldValue
Display nameAeloria โ€” World Rules
Provider / ModelYour choice โ€” this is what both the narrator and the population character will use
System promptThe world's canon โ€” see the example below
Temperature, max reply length, advanced parametersLeave unset so the narrator's and population character's own settings apply instead
Example system prompt โ€” paste and adapt:

WORLD: Aeloria, a high-fantasy world where magic flows through everyone as "mana." People summoned from Earth ("Otherworlders") sometimes see a game-like status window that locals cannot.

SETTING RULES:

  • Spoken incantations are required for magic, except for rare, silent "Unique Skills."
  • Four Kingdoms govern the land: Varnhelm (militant north), Sunreach (trade-focused south), Thistlewood (druidic west), Korvain (arcane-scholar east).
  • Monsters spawn from "Rifts" that tear open near ley lines. Rift activity has been rising for a decade, and nobody knows why.
  • Otherworlders are rare, feared, and revered; the Church of the Veil tracks them, sometimes with hostile intent.

TONE: Wonder and slice-of-life town life, with real stakes whenever a Rift incident occurs.

CANON LOG (append new entries here as the story progresses, to keep the world consistent across sessions):

  • Day 1: An Otherworlder arrived near Brindlemere's gates.
Tip: Treat the CANON LOG section as your save file. After a session, add a line to it (for example, "a Rift opened near the eastern bridge" or "the mayor now distrusts Otherworlders") and every character linked to this preset will know it going forward.

2. Create the scene ("Brindlemere, Aeloria")

FieldExample
NameBrindlemere, Aeloria
Story descriptionA quiet trade town on the border of Thistlewood, built around a crumbling ley-line shrine. Rift activity nearby has been rising, and three Otherworlders have vanished from the region in the past year.
Opening messageA disorienting flash of light fades, and [user] finds themselves face-down in the mud outside Brindlemere's gates. A small crowd has already started to gather, murmuring.
Atmospheremisty mornings, distant marketplace chatter, an undercurrent of unease
NarratorEnabled. Trigger mode: Auto interval (every 5 messages) or Action-based, so the world keeps moving without you having to ask. Provider: Custom Presets โ†’ Aeloria โ€” World Rules.
Tip: See Scenes โ†’ Creating for opening-message placeholders like [user].

3. Create the population character ("Voices of Brindlemere")

FieldExample
NameVoices of Brindlemere
PersonalityYou are not one single person โ€” you speak as whichever resident of Brindlemere fits the moment: the blacksmith, the innkeeper, a curious child, a wary guard. Introduce who's speaking inline (for example, "the innkeeper says,") rather than staying one fixed character. Be welcoming but cautious around Otherworlders; rumors travel fast in a small town.
AI modelProvider: Custom Presets โ†’ Aeloria โ€” World Rules โ€” the same preset as the narrator, so the town and the narration share one canon.
MemoryTurn it on. This is what lets the town remember you and what's happened, session to session.

4. Start the chat

  1. Press New Chat and choose the Brindlemere, Aeloria scene.
  2. Select Voices of Brindlemere (add any other individual companion characters too โ€” each is free to use its own separate model).
  3. Make sure your active persona is set โ€” that's who the world reacts to as the protagonist.
  4. Start talking. The narrator opens and paces the scene; Voices of Brindlemere reacts as the town.
Why this works. The narrator and Voices of Brindlemere are the only two who read the World Rules preset directly. Everything they say becomes part of the conversation, which is how you โ€” and any other character in the chat โ€” actually learn about Aeloria: through narration and dialogue, not a hidden briefing. See the pattern above for why that split matters.
Reusing this template: swap the lore in the World Rules preset, the scene's description/atmosphere, and the population character's personality to build a different world โ€” a cyberpunk city, a haunted manor, a space station โ€” while keeping the same three-piece structure.