AI Settings

Overview

AITuberKit works with multiple AI services to enable character conversation capabilities. This page explains the basic content of AI settings and the supported AI services.

Environment Variables:

# AI service selection
# openai, anthropic, google, azure, xai, groq, cohere,
# mistralai, perplexity, fireworks, deepseek, openrouter, orcarouter, apiroute,
# lmstudio, ollama, dify, custom-api
NEXT_PUBLIC_SELECT_AI_SERVICE=openai

# Selected AI model name
NEXT_PUBLIC_SELECT_AI_MODEL=gpt-4o-2024-11-20

# Whether to use custom model
NEXT_PUBLIC_CUSTOM_MODEL="false"

# Number of past messages to retain
NEXT_PUBLIC_MAX_PAST_MESSAGES=10

# Temperature parameter to adjust conversation randomness (0.0-2.0)
NEXT_PUBLIC_TEMPERATURE=0.7

# Maximum number of tokens
NEXT_PUBLIC_MAX_TOKENS=4096

Supported AI Services

AITuberKit supports the following AI services:

• OpenAI GPT
• Anthropic Claude
• Google Gemini
• Azure OpenAI
• xAI
• Groq
• Cohere
• Mistral AI
• Perplexity
• Fireworks
• DeepSeek
• OpenRouter
• OrcaRouter
• API Route
• LM Studio
• Ollama
• Dify
• Custom API

Using each service requires the corresponding API key.

TIP

API keys may incur usage fees. Please check the pricing structure of each service before using.

Conversation Settings

In AITuberKit, the following settings are available for conversations with AI:

• Number of Past Messages to Retain: Sets the number of messages to retain as conversation history. More messages provide better context understanding but increase API usage costs.
• Temperature Setting: Adjusts the randomness of responses. Higher values result in more diverse responses, while lower values result in more deterministic responses.
• Maximum Number of Tokens: Sets the maximum number of tokens for responses. This value varies depending on the AI model being used.

Reasoning Mode

Some AI models support Reasoning Mode, which enables the AI to generate responses through a deeper thinking process.

# Enable reasoning mode (true/false)
NEXT_PUBLIC_REASONING_MODE=false

# Reasoning level (none/minimal/low/medium/high/xhigh; supported values vary by provider/model)
NEXT_PUBLIC_REASONING_EFFORT=medium

# Reasoning token budget (for Anthropic, Cohere, Google)
NEXT_PUBLIC_REASONING_TOKEN_BUDGET=8192

# Always show thinking process (true/false)
NEXT_PUBLIC_SHOW_THINKING_TEXT=false

Reasoning Level

The available reasoning levels vary depending on the model.

• none: Explicitly disable reasoning
• minimal: Minimal reasoning (supported models only)
• low: Light reasoning (fast)
• medium: Standard reasoning
• high: Deep reasoning (high accuracy but slower)
• xhigh: Extra-deep reasoning (supported models only)

For Ollama, reasoning-capable models and custom models can use none / low / medium / high.

Thinking Text Display

Enabling "Always show thinking text" allows you to display the AI's thinking process in the conversation log.