TLM Self-Talk

A mod that adds self-talk / welcome greetings / maid-to-maid chats to Touhou Little Maid maids.

Features

• Maid self-talk: The maid randomly talks to herself when the owner is online, or when the owner is offline but other players are nearby
• Welcome greetings: The maid greets her owner when they log in — no distance limit; any maid in loaded chunks can trigger
• Maid-to-maid chat: A maid will randomly strike up a conversation with maids near her
• No persona, no talking: Maids without a persona never trigger and none is auto-generated, avoiding the token cost of auto-generating a persona
• Structured persona: The persona page supports whole persona and structured persona modes; the structured mode has four sections (identity, personality, background, hobbies) and can selectively inject persona sections as the situation requires
• Bounded history: When self-talk history exceeds the configured limit, old entries are forgotten and only the latest one is kept, preventing context bloat
• Random in-game context: Randomly injects context such as location / nearby entities / equipment to make self-talk more varied
• Environment awareness: Maids notice nearby deaths, player injuries, their own injuries / burning / drowning, and their own shrine revivals and kills, and bring them into the next self-talk or inter-chat
• Custom Prompt: Customize the prompts used for self-talk / maid-to-maid chats for your own maids; supports both an all-maids setting and independent per-maid settings
• Tool calls: Off by default. When enabled, maids may switch states, fetch environment info, etc. on their own during self-talk / maid-to-maid chats. Incurs extra token cost
• Per-player switch: Each player can toggle self-talk and maid-to-maid chat for individual maids; the host controls the master switch
• Trigger cooldown: A configurable random interval range between triggers
• Quiet while sleeping: Allows maids to stay quiet while sleeping — self-talk and inter-chat won't trigger while they sleep
• Visible in chat: Self-talk and maid-to-maid chats are visible to nearby players
• Prebuilt content (experimental): see the prebuilt content guide

Requirements

Installation

1. Install the mod: download the jar for your version (both client and server must install it, using the same platform and version)
2. Enable the master switch: Maid AI settings → Global AI settings → "Maid Self-Talk" → Enable; Maid Inter-Chat needs to be enabled separately in its own sub-page

Usage

Configuration

Maid AI chat settings → Global AI settings → "Maid Self-Talk":

• State 1 / State 2: enable switch, min/max trigger interval (seconds, typed input with range validation), player radius, and self-talk keep count for owner online / offline
• Welcome: the greeting trigger window after the owner logs in
• Tool Calls: off by default; when on, maids may call tools during self-talk / inter-chat
• Environment Awareness: controls whether maids notice nearby deaths, player injuries, their own injuries / burning / drowning, and creatures they kill themselves

Maid AI chat settings → Global AI settings → "Maid Inter-Chat":

• Basics: master switch (off by default), min/max trigger interval (seconds), player range (blocks), and maid range (blocks)
• Context & chain: keep rounds (a Q/A counts as one round; only the latest message remains when exceeded), chain probability (chance to continue after each reply), max chain length (hard cap on messages per session, guarding against maxed-out chain probability)

Per-player settings

Open the maid's AI chat input screen (usually by pressing T) and click the 💬 button on the left to open the Maid Settings:

• Self-talk / Inter-chat: turn self-talk / inter-chat off individually
• Sleep Quiet: when on, self-talk / inter-chat no longer trigger while the maid is sleeping
• Custom Prompt: fill in separate prompts for all maids and for this maid
• Tool: global and per-maid switches; when enabled, this maid's self-talk / inter-chat can call tools
• Context Settings: choose per entry whether environment info enters the random pool, is always included, or is never included; history context can be full, compact, or retrieval mode
• Thinking Bubbles: controls only the waiting, tool-status and reasoning bubbles you see during self-talk / inter-chat; final replies, welcome messages and normal chat are unaffected
• Prebuilt content (experimental): off by default; when on, generated content is automatically saved as a prebuilt source and reused when conditions match; see the prebuilt content guide for custom prebuilt content

License

MIT