AI Chat and Images

Ririko can talk with your members, translate messages and make pictures. It remembers each member's chat on its own, and it only does what the member is allowed to do.

1. Turn It On

1.1. For the Host

Ririko needs at least one AI provider.

1. Get a key from a provider. Gemini is the default.

   • Gemini: Google AI Studio
   • OpenAI: OpenAI, or any OpenAI-compatible service (set OPENAI_BASE_URL)
   • Ollama: a server you run yourself (set OLLAMA_BASE_URL)

2. Add the key to .env:

   DEFAULT_AI_PROVIDER=gemini
   GEMINI_API_KEY=your_key_here

3. Restart the bot.

4. Run ririko doctor to check the connection. See development.md.

Other variables:

Variable	Default	What it does
DEFAULT_AI_PROVIDER	gemini	The provider tried first: gemini, openai or ollama.
DEFAULT_AI_MODEL	Empty	A model for that provider. Empty uses the provider's default.
GEMINI_API_KEY	Empty	Gemini key.
OPENAI_API_KEY, OPENAI_BASE_URL	Empty	OpenAI key, and an optional address for compatible services.
OLLAMA_BASE_URL	http://localhost:11434	Your Ollama server.

If the first provider fails or runs out of quota, Ririko tries the next provider that you have set up. See adapters.md.

1.2. For the Server Admin

Members can chat with Ririko with /chat or /ai in any channel. For a better experience, make an AI channel.

1. Open the dashboard AI Chatbot page. See dashboard.md.
2. Pick the AI channel.
3. Pick a speaking style.
4. Optional: write a persona.
5. Save.

Or run /aichannel channel:#ririko-ai. With no channel, /aichannel uses the channel where you run it. To remove the AI channel, clear it on the dashboard or type !aichannel reset. Ririko answers every message in the AI channel without a prefix.

2. How Chat Works

• In the AI channel, Ririko answers every message.
• In any other channel, Ririko answers when you mention it (@Ririko how are you?) or run /chat or /ai.
• Messages that start with the command prefix go to the command, not to the AI.
• Ririko remembers the last 20 messages of each member in each channel. Members do not see each other's memory, even in the same channel.
• Ririko sees only your public Discord name, display name and the server name. It never sees internal IDs, keys or account details.
• A reply longer than 2000 characters is sent in more than one message.

2.1. Consent Before Any AI Feature

Ririko sends nothing to an AI provider until the member agrees. This holds for every bot, public or self-hosted. There is no setting to turn it off.

The rule covers every feature that sends a member's words to a provider: chat (by mention, in the AI channel, /chat and /ai), Translate to English, and /imagine with its Regenerate button.

A member who has not agreed gets this question instead of an answer:

Before you use AI features

To answer you, Ririko sends what you write (chat, text to translate, image prompts) to: Google Gemini, OpenAI. Ollama runs on the host's own machine. These providers keep what they receive under their own policies. The bot owner cannot change or recall it.

Ririko stores your AI chats for 90 days, then deletes them.

ai clear deletes the chat in this channel. ai clear all deletes all your chats. ai consent withdraw takes back your consent and deletes your AI data.

Terms of Service · Privacy Policy

Press I agree to use AI features.

[I agree] [No thanks]

• The providers named are the ones this bot is set up with, so the list differs from bot to bot. Ollama and ComfyUI run on the host's own machine. The offline image maker sends nothing anywhere and is not listed.
• The Terms of Service and Privacy Policy links come from TERMS_URL and PRIVACY_URL (see deployment.md).
• Slash commands and Translate to English show the question to you alone. A chat message or a prefix command gets a reply to that message, because Discord cannot hide those. In a busy AI channel, Ririko asks each member once every ten minutes and ignores their other messages until they answer.
• Only the member the question was shown to can press its buttons. I agree also records that they accepted the terms. They then send the request again: Ririko does not remember what they asked before.
• No thanks records nothing. AI stays off for that member, and Ririko can ask again.
• Ririko stores nothing for a member who has not agreed, and does not store a reply when the member withdraws while Ririko is still writing it.
• If Ririko cannot check the consent (for example the database is down), it treats the member as not having agreed.
• When the AI providers or what they receive change in a way that matters, the consent version is raised and every member is asked again.

To take the consent back, type ai consent withdraw (see below).

3. Commands

Command	Aliases	Options	Who can use it
/chat	c	prompt (required)	Everyone
/ai	ask	action, consent, scope, prompt, model, channel, style	Everyone. See below.
/aiclear	clearchat, reset-ai	None	Everyone
/aichannel	setaichannel	channel	Manage Server or Manage Channels
/aipersona	aistyle	style, prompt	Manage Server to change. Everyone to view
/aimodel	setmodel	model	Manage Server to change. Everyone to view

/ai does different things depending on action:

Action	What it does	Who can use it
chat (default)	Chats. Use prompt for your message.	Everyone
model	Shows or sets the model. Use model.	Manage Server to set
channel	Sets the AI channel with channel. With no channel, it shows the current AI channel. To remove the channel, type !ai channel reset.	Manage Server or Manage Channels
persona	Sets the style and persona. Use style and prompt.	Manage Server
clear	Clears your memory in this channel. Use scope to choose what to clear.	Everyone
consent	Shows or withdraws your AI consent. Use consent to choose.	Everyone

/ai action:clear has two scopes:

Scope	Prefix form	What it deletes
here (default)	!ai clear	Your conversation in this server and channel.
all	!ai clear all	Every AI conversation and message of yours, in every server and channel. Ririko answers with the number of conversations it deleted. With the slash command only you see the answer.

• Clearing does not change your AI preferences (nickname, timezone and language).
• Deleting a conversation removes it from Ririko's database only. Messages that Ririko already sent to an AI provider cannot be recalled: the provider keeps them under its own terms.

/ai action:consent has two choices. Only you see the slash answer.

Choice	Prefix form	What it does
status	!ai consent status	Says whether you agreed to AI features and on which day.
withdraw	!ai consent withdraw	Takes back your AI consent, deletes all your AI conversations in every server and channel, and deletes your AI preferences (nickname, timezone and language). Ririko answers with what it deleted. AI stays off for you until you agree again. Messages already sent to an AI provider cannot be recalled.

Without a choice, Ririko tells you to use status or withdraw. These commands send nothing to a provider, so they work without consent.

• Speaking styles: FRIENDLY_ANIME (default), TSUNDERE, KUUDERE, DANDERE, GENKI and FORMAL.
• The model is a choice for the whole server. Run /aimodel with no option to see what you can choose. A choice is provider, such as openai, or provider/model, such as gemini/gemini-2.5-pro. Type default to reset it.
• Models you can choose:
  • Gemini: gemini-2.5-flash, gemini-2.5-pro, gemini-3.5-flash-lite, gemini-3.7-flash
  • OpenAI: gpt-4o-mini, gpt-4o, o3-mini
  • Ollama: llama3.3, llama3.2, mistral, qwen2.5, deepseek-r1
• The server chooses the provider that Ririko tries first. If it fails, the next provider answers with its own default model.

4. Server Settings

Open the dashboard AI Chatbot page.

Setting	What it does
AI channel	The channel where Ririko answers every message.
Speaking style	The tone of Ririko.
Persona	Extra instructions for Ririko's personality, up to 1500 characters.
Tools	The actions Ririko may take while chatting. Remove every tool to turn tools off.
Provider and model	The provider and model to try first. The page lists only providers the host has set up.

You can also use the command line: ririko guild:config <server> ai.<key> <value>.

5. Tools

A tool lets Ririko do something for a member while it chats. Ririko checks the permissions of the member before it uses a tool.

Tool	What it does
Current time	Tells the time in your time zone. Ririko uses your own zone, then the server zone, then UTC.
Coin flip	Flips a coin.
Anime search	Looks up an anime.
Check a balance	Shows your credits.
Play music	Adds a song to the music queue.
Create a reminder	Sets a reminder for you.

Ririko cannot run programs, read files, make web requests, change the database or use admin functions. It cannot ban, kick or change roles. If you ask it to do something you may not do, it refuses.

6. Translate to English

Right-click a message, choose Apps, then Translate to English. Only you see the result. See commands.md. Translation needs your AI consent (section 2.1).

7. Images

/imagine makes a picture from a text description. It needs the member's AI consent (section 2.1), and so does its Regenerate button.

7.1. Turn It On

The host sets at least one image provider in .env:

Variable	What it does
GEMINI_API_KEY	Uses Google Gemini Imagen.
REPLICATE_API_TOKEN	Uses Replicate (Flux models).
COMFYUI_URL	Uses your own ComfyUI or Stable Diffusion server. COMFYUI_BASE_URL and SD_WEBUI_URL also work.
IMAGE_DEFAULT_PROVIDER	The provider to use when nobody picks one: gemini, replicate, comfyui or mock. Default: gemini.
IMAGE_DAILY_QUOTA	The number of images each member may make in 24 hours, across all servers. Default: 30. 0 means no limit.

The mock provider makes a simple test picture. Ririko uses it when the first provider fails and the member did not pick one.

Never put image keys in Discord. The /setup-stablediffusion-api command only explains this.

7.2. Commands

/imagine (aliases img, ai-image, generate-image). The cooldown is 5 seconds.

Option	Required	What it does
prompt	Yes	What to draw.
preset	No	A style: anime (default), photoreal, pixel-art, fantasy, cyberpunk or none.
aspect_ratio	No	1:1 (default), 16:9, 9:16, 4:3 or 3:4.
negative_prompt	No	What to leave out.
provider	No	auto (default), gemini, comfyui, replicate or mock.

Prefix form: !imagine a castle at night --ar 16:9 --preset fantasy --negative people.

/stablediffusion-model (aliases sdmodel, sd-model, imagemodel, image-model) sets or shows the defaults of the server. It needs Manage Server.

Option	What it does
preset	The default style.
provider	The default provider: gemini, replicate, comfyui or auto.

7.3. Server Settings

Open the dashboard Image Generation page.

Setting	What it does
Default provider	The provider for /imagine when a member picks none. Only providers the host set up are listed.
Images per member (24 hours)	A limit for your server. It can be lower than the host's quota, not higher.
Default style	The style used when a member picks none.

8. Limits

• Ririko needs a working provider and quota. If the provider fails, Ririko tells the member.
• Ririko remembers the last 20 messages of each member in a channel.
• Image requests wait in a queue. A request that waits long can take time.
• Input for translation is limited to 12,000 characters.