Klikněte na „Odebírat“ pro stažení položky
RiMind

This mod is a personal modification based on the respected juicy's RimTalk, which brought a new singularity to RimWorld. It was created by referencing some of RimTalk's core API calling and bubbler rendering logic. If you want to check out the original mod, search for RimTalk!

Permission to use the codebase was obtained from the original author Juicy.
I am optimizing the code for better performance, but this won't reduce the total ticks consumed by the "Bubble" mod itself.
I use Dubs Performance Analyzer for benchmarking. If the mod consumes more than 0.2ms of ticks, I consider it a memory leak.

Mod Description

Type-Based Memory Storage: Separately manages memories of what a pawn said, heard, witnessed, and experienced in combat or other events.
Smart Memory Filtering: Selects relevant memories for conversation topics based on freshness, importance, and relationships.
Long-Term Memory: Pawns remember past conversations and events, leading to more consistent dialogue.
Memory Viewer: A UI that allows you to view and manage each pawn's memories in real-time.

Multi-Model Support: Automatically switches between various AI models like Gemini-2.5-flash, Gemini-2.0-flash, and Gemma-3.
Quota Optimization: When one model's quota is reached, it automatically switches to another to maximize API usage.
Single API Key: Access multiple AI models with just one Google API key.
Intelligent Fallback: Automatically switches to an available model when the current one's quota is exceeded.

Event-Based Topics: Map events, combat, injuries, and skill improvements automatically generate new conversation topics.
Emotion Integration: Mood changes provide natural opportunities for dialogue.
Job Change Detection: New work assignments can also be a starting point for conversations.
Line-of-Sight Witnessing: Colonists use what they directly observe as conversation topics, with the ability to combine multiple objects to generate dialogue.

Automatic Urge to Speak: Pawns' desire to speak is adjusted based on their health, recent conversations, observations, and combat status.
Relationship-Based Responses: Dialogue patterns are adjusted according to relationships between pawns.
Varied Utterances: Pawns speak differently depending on the situation, whether it's combat, hunting, or daily work.
Consciousness Adaptation: Dialogue reflects the pawn's consciousness level, influenced by injuries or drugs.
Persona Editor: Customize each pawn's speaking style and personality individually.
AI Persona Generation: The mod can automatically generate a suitable personality based on a pawn's stats, background, and traits.

Instantaneous Reactions: Pawns can respond to another pawn's speech in real-time.
Delayed Responses: A natural system where pawns might respond later if they are busy.
Response Expectation: Pawns will accumulate disappointment if their questions are left unanswered.
Conversation Momentum: Implements a back-and-forth "tiki-taka" dynamic in conversations.

You can now remove the mod mid-playthrough and load your save file. (9/19 Patch)

1. Get a Google Account: You'll need a Google account.
2. Get a Free API Key:[aistudio.google.com]

• Click 'Get an API key' to generate a key (top right corner).
  
• Copy the generated key.

3. Paste the API Key: In the mod settings, paste the copied key into the API key field.

This mod is filled with experimental concepts and may exhibit unexpected behaviors depending on how the AI models operate. While these unpredictable interactions might bring greater enjoyment, they could also cause inconvenience.

This is a personal hobby project, so I can't guarantee continuous feedback and improvements.

I haven't been able to check compatibility with a wide range of mods yet.
There has been one report of a conflict with the Monolyn race mod.
Some API keys occasionally experience a 503 error. This is a known issue mentioned by Google on September 3rd.
A weighting system has been implemented for random and unpredictable utterances. Pawns can have various reactions in the same situation.

Please do not use arbitrary special characters in your prompts. This often causes the AI to return an error.
While I have no objection to providing various LLM environments, ensuring the core functionality works properly must always be the priority. I'll be focusing more on basic tasks like bug fixes, code optimization through refactoring (it's a real spaghetti mess right now), and improving conversation quality.


Q: Can I use it without any limits?

A: The goal of this mod is not to provide unlimited API usage. Once you've exhausted your request limit, each API will have a cooldown period. However, if you have multiple API keys, you can get usage that is effectively close to unlimited.

2025-09-23

I've added support for the SiliconFlow API. During this process, I made a small change to the Mode Settings menu. Now, you should select "Custom" from the first dropdown, and then choose a service like OpenRouter, Mistral, or SiliconFlow from the second dropdown.

Also, I've adjusted some values like Temperature and TopK to address the 429 errors that have been reoccurring with Gemini. This issue is not fully resolved yet.

2025-09-20

You can now use OpenRouter normally. Please enter your API key in the first field and the model name in the second field.
I've fixed the issue where models other than Gemini couldn't generate personas.

2025-09-19

I have fixed the issue where the log menu would break when adding the mod mid-playthrough and then removing it after saving.