Waifu

Waifu Plugin

A LangBot plugin that creates realistic chatbots by combining programmatic logic with LLMs, exploring an authentic chat experience.

Language / 语言: 简体中文 | English | Back to Main

The community group is the LangBot community group. For questions about the main program or plugins, please ask there. Check the documentation and Issues before asking.

Important Notice: The Waifu project will only receive essential maintenance updates going forward. Version 2.0 and subsequent features will be presented as an independent project: Husbando.

Features

• Multi-level Memory System: Short-term + long-term memory summarization
• Character Card System: Separate templates for private/group chats
• Story Mode: Narrator, affection system for immersive storytelling
• Humanized Mode: Typing delay, segmented replies for natural conversation
• Proactive Greeting: Automatic initiation when user is inactive
• Emotion Perception: Analyze user emotions and adjust reply tone
• Life Events: Fictional daily activities for characters
• Web Search: AI autonomously decides whether to search
• Jailbreak Mode: Multi-position prompt injection
• Chain of Thought: Enhanced humanization through reasoning

Installation

1. Configure the LangBot main program
2. Search for Waifu in the LangBot Plugin Market to install
3. Or enter the repository URL in the plugin page of the admin panel:

https://github.com/ElvisChenML/Waifu

Or see the detailed Plugin Installation Guide

Configuration

Configure the following parameters in the LangBot plugin management interface after installation:

Basic Settings

Config Item	Type	Default	Description
character	String	default	Character card name. default auto-selects template, off disables preset
llm_model	Model Selection	First available model	LLM model used by Waifu
summarization_mode	Boolean	true	Enable long-term memory summarization; if disabled, truncates when limit exceeded
short_term_memory_size	Integer	2000	Maximum character count for short-term conversation memory

Mode Settings

Config Item	Type	Default	Description
story_mode	Boolean	false	Story mode (narrator, affection system), only effective in private chats
thinking_mode	Boolean	false	Chain of thought, enhances humanization, may affect reply style
personate_mode	Boolean	true	Humanized mode: typing delay, segmented replies
jail_break_mode	Select	off	Jailbreak prompt injection position: off / before / after / end / all
emotion_mode	Boolean	true	Emotion perception: analyze user emotions, auto-adjust reply tone
event_mode	Boolean	true	Life events: fictional daily activities for characters, enrich conversation topics

Reply Settings

Config Item	Type	Default	Description
response_rate	Float	1.0	Reply probability after reaching minimum message count in group chat (0.0-1.0)
group_response_delay	Integer	0	Seconds to wait before generating reply in group chat (message merge window)
person_response_delay	Integer	5	Seconds to wait before generating reply in private chat

Proactive Greeting Settings

Config Item	Type	Default	Description
proactive_greeting_enabled	Boolean	false	Enable proactive greeting when user is inactive
proactive_target_type	Select	person	Proactive greeting target type: person (private chat) / group (group chat)
proactive_target_id	String	Empty	Target user/group ID for proactive greeting, leave empty to disable

Search Settings

Config Item	Type	Default	Description
search_mode	Boolean	false	Web search: AI autonomously decides whether to search and include results
search_api	Select	duckduckgo	Search engine: duckduckgo / bing / google
search_api_key	String	Empty	API key for Bing or Google (Google format: api_key:cx_id)

Commands

Command components are divided into two categories:

• Command: Instructions that need to be actively sent
• EventListener: Automatically triggered by message events, no explicit call required

Basic Commands

Command	Description
列出命令	List all supported commands and descriptions
全部记忆	Display all current long and short-term memory
删除记忆	Delete all long and short-term memory
加载配置	Reload all configuration files
撤回	Delete the last dialogue from short-term memory

Story Commands

Command	Description
修改数值[value]	Modify affection value, e.g., 修改数值100
态度	Display current attitude based on affection value
停止活动	Stop narrator timer
开场场景	Trigger narrator to output opening scene from character card
旁白	Manually trigger narrator to advance plot
继续	Manually trigger Bot to continue reply and advance plot
控制人物[character]|[dialogue]	Control character speech or trigger AI to generate character messages
推进剧情[character_name]	Auto-sequentially call: narrator → control character

Debug Commands

Command	Description
请设计[content]	Design a list
请选择[question]|[option1,option2]	Choose from given list
回答数字[question]	Return numeric answer
回答问题[system_prompt]|[user_prompt]	Q&A mode with custom system prompt

Character Cards

Directory Structure

data/plugins/Waifu/
├── cards/      # Character card files (.yaml)
├── config/     # Configuration files
└── data/       # Runtime data (conversation logs, memory, etc.)

Template files are stored under plugins/Waifu/templates/. Files under data/plugins/Waifu/ are generated from templates.

Character Card Format

# Required fields
user_name: Wang           # How to address the user
assistant_name: Susu      # Character name
language: English         # Conversation language
Profile:
  - Bio: You are... (character description)

# Optional fields
Skills:
  - You speak gently with humor, avoiding aggressive words, making people feel comfortable.
Background:
  - You chat with the user via QQ.
Rules:
  - When introducing yourself, only mention your name, don't include occupation or other info.
Prologue:
  - Every day, Wang sells watermelons at the school gate... (opening scene description)

# Story mode related (optional)
max_manner_change: 10
value_descriptions:
  - max: 100
    description:
      - Interaction behavior: Maintain appropriate distance, formal language.
  - max: 500
    description:
      - Interaction behavior: Enter romantic phase, words reveal tenderness and affection.

Using Custom Character Cards

1. Create a character card referencing the template plugins/Waifu/templates/default_person.yaml, e.g., example.yaml
2. Place the file in the data/plugins/Waifu/cards/ directory
3. In LangBot Plugin Management Panel → Waifu → Configuration, set the character field to example, save to take effect

Using different character cards for specific users/groups: Create or edit waifu_{session_ID}.yaml under data/plugins/Waifu/config/, add character: "example" to override the global configuration.

How It Works

1. Message Listening: Listen to private/group chat message events
2. Emotion Analysis: Analyze user emotional state (if enabled)
3. Memory Retrieval: Retrieve short-term memory and long-term memory summary
4. Life Events: Get character's current fictional activities (if enabled)
5. Web Search: AI decides whether search is needed (if enabled)
6. Chain of Thought: Generate internal reasoning process (if enabled)
7. Reply Generation: Generate character reply based on context
8. Humanized Output: Simulate typing delay and segmented replies (if enabled)
9. Memory Update: Update short-term memory, summarize to long-term memory when necessary

Version History

Click to expand/collapse

Waifu! 3.0

• Compatible with latest plugin system

Waifu 2.0

• Emotion perception: Analyze user emotional state, AI auto-adjusts reply tone
• Life events: Fictional daily activities for characters, enrich conversation topics
• Web search: AI autonomously decides whether to search and include in context
• Migrated to LangBot new plugin architecture, supports plugin management interface configuration

Waifu 1.9

• Support running with other plugins simultaneously, no longer blocks subsequent events
• No longer need to modify random response probability in group message response rules
• Jailbreak adds all mode, enables all jailbreak positions simultaneously

Waifu 1.6

• Added character: off option, disables character preset

Waifu 1.5

• Added group chat blacklist blacklist config item
• Added repeat repeat_trigger config item
• Added max narrator word count, max thinking word count configs

Waifu 1.4

• Added thinking_mode to disable chain of thought
• Added character card built-in attribute Prologue (opening scene) and [开场场景] command

Waifu 1.3

• Added integrated jailbreak, supports three positions: before/after/end of system prompt
• Added continued_rate for auto-continue speech
• Added display_value to control whether to display value each time
• Added 继续, 剧情推进 commands

Waifu 1.2

• Group chat supports @ commands
• Messages support images
• Added nickname recognition

Waifu 1.1

• Each session uses independent configuration file
• Added undo function
• Added message accumulation processing
• Added humanized mode (typing time, segmented replies)

Waifu 1.0

• Basic framework: Q&A module, character module, thinking module, memory database, narrator, emotion perception, status bar

Contributing

1. Clone LangBot
2. Clone Waifu
3. Create plugins/ directory in the LangBot root directory
4. Place Waifu in the LangBot/plugins/ directory

License

MIT

Acknowledgments

• LangBot — Provides Bot functionality and base methods
• LangGPT — Provides character preset prompt paradigm
• Tencent TexSmart — Provides text understanding API
• CyberWaifu, koishi-plugin-aikanojo, Spit_chatBot — Provided ideas and code references