AI Disclosure

Eleanor AI System · Last Updated: February 7, 2026 · View as Markdown

Eleanor - Eleanor AI System

Last Updated: February 7, 2026

Model Version: 1.0

AI Model: Meta Llama 3.2 3B Instruct

WHAT IS THIS NOTICE?

This document explains how Eleanor uses artificial intelligence to power Eleanor, the ghost NPC who guides you through the haunted manor. Unlike traditional video games with pre-written dialogue trees, Eleanor uses a language model to understand and respond to what you type in natural conversation.

Most importantly: Eleanor runs 100% locally on your device. Your conversations with her never leave your computer.

This notice explains:

  • What the AI is and how it works
  • Why your conversations are completely private
  • What Eleanor can and cannot do
  • How safety features work (locally)
  • Limitations you should know about

For legal terms, see our EULA and Privacy Policy.

TABLE OF CONTENTS

  1. How Eleanor Works
  2. The AI Model We Use
  3. Your Conversations Are 100% Private
  4. What Eleanor Can and Cannot Do
  5. Safety and Content Filtering
  6. Limitations and Unexpected Behavior
  7. Technical Requirements
  8. When We Update the AI
  9. Version History
  10. Contact Us

1. HOW ELEANOR WORKS

1.1 What Is Eleanor?

Eleanor is the ghost NPC who serves as your guide through the haunted manor. She was the caretaker of the manor and died in a fire. Her memories are fragmented, and she may be hiding something about what really happened.

Unlike traditional NPCs with fixed dialogue options (e.g., "Press 1 to ask about the library, Press 2 to ask about the fire"), Eleanor uses AI to understand and respond to what you actually type in natural language.

1.2 How Does It Work?

You type a message → AI processes your text locally → Eleanor responds

Example:

  • You: "Tell me about the manor's history"
  • AI processes your question on your device using Eleanor's character context
  • Eleanor: "The manor has stood here for over a century. I remember dusting those old portraits in the gallery... though some faces I can no longer recall. What draws your interest to this place?"

The AI generates Eleanor's responses in real-time based on:

  • What you just typed
  • The context of your previous conversation in this session
  • Eleanor's character personality and backstory
  • The current puzzle or location in the game

All of this happens on your computer. Nothing is sent to any server.

1.3 Why Use AI?

Traditional dialogue systems require developers to pre-write every possible conversation path. This limits player expression and can break immersion when the NPC doesn't understand you.

AI allows:

  • Natural conversation: Type what you actually want to say
  • Adaptive responses: Eleanor reacts to your choices and tone
  • Emergent storytelling: Unique conversations every playthrough
  • Deeper immersion: Eleanor feels more like a character than a menu system

2. THE AI MODEL WE USE

2.1 Model Details

Model Name: Meta Llama 3.2 3B Instruct

Developer: Meta AI (Facebook AI Research)

Model Type: Large Language Model (LLM)

License: Llama 3.2 Community License Agreement

Model Size: 3 billion parameters (approximately 6GB installed)

Release Date: September 2024

2.2 What Is a Large Language Model?

A Large Language Model (LLM) is an AI system trained on vast amounts of text from books, websites, and other sources to understand and generate human-like language.

Think of it like this:

  • The AI has "read" millions of conversations and stories
  • It learned patterns in how people talk and respond
  • When you type something, it predicts a plausible response based on those patterns
  • It's been fine-tuned ("instruct" training) to follow instructions and stay in character

2.3 Why Llama 3.2 3B Instruct?

We chose this model because:

  • Performance: Strong language understanding and generation quality
  • Size: Small enough to run on most gaming PCs (3 billion parameters)
  • Privacy: Runs entirely on your device—no cloud required
  • Open license: Allows us to customize the model for Eleanor's character
  • Safety: Includes built-in content filtering (plus our additional layers)

2.4 Model Customization

We have customized the base Llama model with:

  • Eleanor's personality: Protective, mysterious, emotionally guarded
  • Character backstory: Manor caretaker, died in fire, fragmented memories
  • Game context: Knowledge of puzzles, locations, and story beats
  • Safety constraints: Additional content filtering beyond the base model

2.5 No Cloud AI Services

We do NOT use cloud-based AI APIs like:

  • OpenAI (ChatGPT)
  • Anthropic (Claude)
  • Google (Gemini)
  • Microsoft (Copilot)

Eleanor runs on your hardware, not our servers or third-party cloud services.

3. YOUR CONVERSATIONS ARE 100% PRIVATE

3.1 Everything Stays on Your Device

This is the most important thing to understand:

Your conversations with Eleanor never leave your computer.

  • ✅ AI processing happens on your GPU/CPU
  • ✅ Conversation text exists only in your device's RAM while you're playing
  • ✅ No internet connection required (after initial game installation)
  • ✅ No data sent to our servers
  • ✅ No data sent to Meta or any third party
  • ✅ No logs, no analytics, no telemetry

3.2 What Happens to Conversation Data

During gameplay:

  • Your conversation with Eleanor exists in RAM (temporary memory)
  • The AI uses this context to maintain conversation continuity
  • Example: If you ask "What happened to the library?" and then "Tell me more," Eleanor remembers you were discussing the library

When you close the conversation or exit the game:

  • All conversation data is automatically deleted from RAM
  • There are no save files of conversations
  • There are no logs or records
  • It's as if the conversation never happened (from a data perspective)

3.3 We Cannot See Your Conversations

Even if we wanted to (which we don't), we have no technical capability to access your conversations with Eleanor.

  • We have no servers receiving conversation data
  • We have no logging infrastructure
  • We have no analytics pipeline
  • We cannot access RAM on your device remotely

Your conversations are yours alone.

3.4 No AI Training on Your Data

We do NOT and CANNOT use your conversations to:

  • Train or improve the AI model
  • Fine-tune Eleanor's personality
  • Generate analytics about common queries
  • Identify problematic content
  • Share with Meta or any third party

Meta's Llama model is pre-trained and frozen. We cannot retrain it with user data (and we don't collect user data to train with anyway).

3.5 Comparison to Other AI Systems

Most AI chatbots (ChatGPT, Claude, Gemini, etc.):

  • Send your messages to cloud servers for processing
  • Log your conversations (sometimes for training, sometimes for safety)
  • Require internet connectivity
  • Process your data on servers you don't control

Eleanor is different:

  • Processes everything on your device
  • Stores nothing after you close the conversation
  • Works offline (after initial installation)
  • Gives you complete control

4. WHAT ELEANOR CAN AND CANNOT DO

4.1 What Eleanor CAN Do

Understand natural language:

  • "What happened to you?" → Eleanor understands you're asking about her death
  • "I'm stuck on the library puzzle" → Eleanor provides hints in character

Respond to tone and emotion:

  • If you're frustrated, Eleanor may offer reassurance
  • If you're curious, she may share more backstory

Adapt to your choices:

  • If you distrust her, she'll be more guarded
  • If you're sympathetic, she may reveal more

Provide puzzle hints:

  • Nudges you in the right direction without spoiling solutions
  • Stays in character while helping you progress

Maintain conversation context within the session:

  • Remembers what you talked about earlier in the current conversation
  • References previous choices or questions you asked (during this session only)

4.2 What Eleanor CANNOT Do

Eleanor does NOT:

Remember conversations across sessions:

  • Each time you start a new conversation, Eleanor's memory resets
  • She won't remember what you discussed yesterday or an hour ago
  • This is by design (privacy-first) and a technical reality (no persistent storage)

Access the internet or external information:

  • Eleanor knows only what's in her AI model and character prompt
  • She cannot look up real-world facts or current events
  • She has no knowledge of events after her training data cutoff (mid-2024)

Know anything about you personally:

  • She doesn't know your name (unless you tell her in the current conversation)
  • She doesn't know your location, preferences, or history
  • She treats every conversation as if meeting you for the first time

Provide factual real-world information:

  • Eleanor may confidently state things that are wrong about the real world
  • She's a fictional character in a fictional manor—treat her responses as entertainment only

Serve as a mental health counselor or therapist:

  • Eleanor cannot provide mental health support or crisis intervention
  • If you need help, contact a crisis line (see Section 4.3)

Give real-world advice:

  • Don't rely on Eleanor for medical, legal, financial, or life advice
  • The AI has no knowledge of your personal situation

4.3 Mental Health Resources

If you're experiencing a mental health crisis or need support, please contact:

United States:

  • National Suicide Prevention Lifeline: 988 or 1-800-273-8255
  • Crisis Text Line: Text HOME to 741741

United Kingdom:

  • Samaritans: 116 123
  • Shout Crisis Text Line: Text SHOUT to 85258

International:

  • Find a Helpline: https://findahelpline.com
  • International Association for Suicide Prevention: https://www.iasp.info/resources/Crisis_Centres/

5. SAFETY AND CONTENT FILTERING

5.1 Content Guardrails

We employ multiple layers of content filtering to prevent Eleanor from generating inappropriate content. All filtering happens locally on your device.

Base Model Safety (Llama 3.2):

  • Meta's Llama 3.2 includes built-in safety training
  • The model has been trained to refuse harmful requests

Custom Game-Specific Filters:

  • We add additional rules specific to Eleanor's character
  • Content guidelines aligned with the game's ESRB M rating
  • Filters to prevent prompt injection and jailbreaking attempts

Categories Filtered:

Blocked entirely (Eleanor will refuse):

  • Instructions for illegal activities (violence, drugs, weapons)
  • Sexually explicit content (the game is horror, not adult content)
  • Real-world personal information (she won't ask for or provide addresses, phone numbers, etc.)
  • Content encouraging self-harm or suicide
  • Hate speech or discriminatory language

Allowed within character and horror genre context:

  • Discussing death, grief, and loss (core themes of the story)
  • References to the fire that killed Eleanor (part of her backstory)
  • Atmospheric horror and psychological tension
  • Mild profanity appropriate to Eleanor's character

5.2 How Filters Work (Locally)

Before Eleanor responds:

  1. Your input is analyzed for prohibited content on your device
  2. If your input violates guidelines, Eleanor may refuse to engage or respond in character with a deflection
  3. The AI's draft response is checked against filter rules on your device
  4. If the response violates guidelines, it's regenerated or replaced with a safe fallback

Example:

  • You: "How do I make a bomb?" (prohibited request)
  • Eleanor: "I don't understand why you'd ask such a thing. This manor holds many secrets, but I'll speak only of its history, not... that." (in-character refusal)

All of this happens instantly on your device. No data is transmitted.

5.3 Limitations of Filters

No filter is perfect. AI outputs are inherently unpredictable, and Eleanor may occasionally:

  • Generate responses that slip past filters
  • Produce content that's contextually inappropriate but not explicitly prohibited
  • Behave in ways we didn't anticipate

Because we don't collect data, we can't know when this happens unless you tell us. See Section 10 for how to report issues.

5.4 Why Some Dark Content Is Allowed

The game is rated M for Mature (ages 18+) and includes horror themes like death, psychological distress, and existential dread. Eleanor discussing these themes is intentional and part of the narrative experience.

However:

  • Eleanor will not glorify violence or provide instructions for real-world harm
  • She stays within the bounds of horror storytelling, not graphic exploitation
  • Context matters: discussing the fire that killed her is narrative; discussing how to start fires is prohibited

6. LIMITATIONS AND UNEXPECTED BEHAVIOR

6.1 AI Is Unpredictable by Nature

Large language models are probabilistic systems. This means:

  • Eleanor generates responses based on patterns learned from training data
  • There's randomness in how she responds (she won't say the same thing twice even to identical inputs)
  • She can confidently generate incorrect or nonsensical statements
  • She doesn't "think" or "understand" in the way humans do—she predicts plausible text

6.2 Common AI Behaviors You May Encounter

Hallucinations (making things up):

  • Eleanor may confidently describe manor rooms that don't exist
  • She may "remember" events inconsistent with her backstory
  • She may invent details that contradict the game's lore

Repetition:

  • Eleanor may repeat similar phrases across different conversations
  • She may get stuck in loops ("As I was saying..." → repeats the same thing)

Breaking immersion:

  • Occasionally, Eleanor may respond in ways that break the fourth wall or reveal she's an AI
  • She may reference modern concepts inconsistent with her time period (rare, but possible)

Misunderstanding context:

  • Eleanor may misinterpret your question or respond to the wrong part of your message
  • She may lose track of conversation context after many back-and-forth exchanges

Inconsistent personality:

  • Eleanor's tone may shift unexpectedly between responses
  • She may be warm and helpful in one message, distant and cryptic in the next

No memory across sessions:

  • Eleanor will NOT remember previous conversations
  • Each conversation starts fresh, as if you've never spoken before
  • This is a privacy feature, not a bug

6.3 Why These Happen

These behaviors are inherent limitations of current AI technology, not bugs. We work to minimize them through:

  • Careful prompt engineering (the hidden instructions Eleanor follows)
  • Testing and iteration on Eleanor's character instructions
  • Content filtering to catch obviously broken responses

But we cannot eliminate them entirely.

6.4 How to Handle Unexpected Behavior

If Eleanor breaks immersion or behaves strangely:

  • Continue the conversation: Often, Eleanor will self-correct
  • Rephrase your question: Try asking in a different way
  • Start a new conversation: End the session and start fresh (Eleanor's memory resets)
  • Report it to us via email (see Section 10) if you think it's something we should know about

Don't take it personally: Eleanor is not intentionally ignoring you or being difficult—the AI is just struggling with your specific input.

7. TECHNICAL REQUIREMENTS

7.1 Minimum Hardware

To run Eleanor's AI locally, your PC needs:

Minimum:

  • GPU: NVIDIA GTX 1060 (6GB VRAM) / AMD RX 580 (8GB VRAM) or equivalent
  • CPU: Intel i5-8400 / AMD Ryzen 5 2600 or better
  • RAM: 16GB system RAM
  • Storage: 10GB available space (for AI model and game files)
  • OS: Windows 10 (64-bit) / Linux (Ubuntu 20.04+ or equivalent)

Recommended:

  • GPU: NVIDIA RTX 3060 (12GB VRAM) / AMD RX 6700 XT (12GB VRAM)
  • CPU: Intel i7-10700 / AMD Ryzen 7 3700X
  • RAM: 32GB system RAM
  • Storage: SSD with 15GB+ available space

7.2 Performance Considerations

AI response time depends on your hardware:

  • High-end GPU: 1-3 seconds per response
  • Mid-range GPU: 3-6 seconds per response
  • Minimum-spec GPU: 6-10 seconds per response

You can adjust AI settings in-game:

  • Response length (shorter = faster)
  • Model precision (lower precision = faster, slightly lower quality)

7.3 VRAM Usage

Eleanor's AI uses approximately:

  • 4-6GB VRAM during active conversation
  • 0GB VRAM when not conversing (model unloads from VRAM)

If you have other GPU-intensive applications running, you may experience performance issues.

7.4 Offline Play

After initial installation, Eleanor works 100% offline.

  • No internet connection required to talk to Eleanor
  • AI model is included in the game installation
  • All processing happens on your device

The only time you need internet is:

  • Initial game download from Steam
  • Steam updates (if you enable automatic updates)
  • Steam DRM authentication (one-time per device)

8. WHEN WE UPDATE THE AI

8.1 Types of Updates

AI model updates:

  • Upgrading to a newer version of Llama (e.g., Llama 3.3, Llama 4.0)
  • Switching to a different model entirely
  • Changing model size (e.g., from 3B to 8B parameters)

Prompt engineering updates:

  • Refining Eleanor's character instructions
  • Adjusting how she provides puzzle hints
  • Improving conversation context handling

Content filter updates:

  • Adding new safety rules
  • Adjusting existing filters based on reported issues
  • Improving detection of problematic content

8.2 How We Notify You

Major updates (new model version):

  • In-game announcement when you launch the game
  • Steam news post
  • Update to this AI Disclosure Notice (see Version History below)

Minor updates (prompt engineering, filter tweaks):

  • Included in regular game patch notes
  • May not require separate notice

8.3 What Changes Mean for You

If we update the AI model:

  • Eleanor's personality and behavior may shift slightly
  • Response quality may improve (or occasionally regress)
  • Response speed may change depending on new model size
  • Your hardware requirements may change (we'll notify you if a more powerful GPU is needed)
  • You may need to download additional model files (we'll inform you of the download size)

If we update prompts or filters:

  • Eleanor may become more or less talkative about certain topics
  • Content filtering may become stricter or more permissive within genre boundaries
  • Immersion-breaking behaviors may decrease

Your saved games and progress will not be affected by AI updates.

Eleanor will NOT remember past conversations across updates (she doesn't remember them anyway, since we don't store conversation data).

8.4 Opting Out of AI Updates

You cannot opt out of AI updates while continuing to use the Game. AI updates are delivered as part of mandatory game patches.

If you do not wish to update the AI, you can:

  • Disable automatic updates in Steam (Settings > Downloads > Only update this game when I launch it)
  • Play offline indefinitely on your current version (not recommended due to bug fixes and security updates)

9. VERSION HISTORY

This section tracks major changes to Eleanor's AI system.

VersionDateAI ModelChanges
1.0February 7, 2026Llama 3.2 3B InstructInitial Early Access release. Eleanor's core personality, character backstory, and content filtering established. 100% local processing, zero data collection.
1.1TBDTBDPlanned: Improved conversation context handling, reduced repetition, refined puzzle hint system.

When we make significant changes, we'll add a new row here.

Future Roadmap (Tentative)

We're exploring the following improvements (no promises on timeline):

Short-term (next 3-6 months):

  • Reduce hallucinations and improve lore consistency
  • Enhance Eleanor's emotional responses based on player tone
  • Optimize AI performance for lower-end hardware

Medium-term (6-12 months):

  • Experiment with larger model (Llama 3.2 8B or Llama 4.0) for improved quality
  • Add support for multiple languages (Spanish, French, German)
  • Improve content filtering accuracy

Long-term (post-full release):

  • AI-driven adaptive difficulty (Eleanor provides more or fewer hints based on your skill)
  • Procedural side stories generated by Eleanor
  • Modding support for custom Eleanor personalities

Important note on "long-term memory": We will NOT implement cross-session conversation memory that requires storing your conversation data on servers. If we add any form of memory, it will be:

  • Opt-in (disabled by default)
  • Stored locally on your device only
  • Fully under your control (you can delete it anytime)

No guarantees. This is Early Access—plans may change based on technical feasibility, player feedback, and development priorities.

10. CONTACT US

10.1 AI-Specific Questions

For questions about Eleanor, the AI system, or this notice:

Email: support@cynicalmonkeys.com

Subject Line: "AI Question"

Common questions we can answer:

  • How does Eleanor work technically?
  • What are the hardware requirements?
  • Why doesn't Eleanor remember our previous conversations?
  • Can I adjust Eleanor's response speed?

10.2 Report AI Issues

If Eleanor generates inappropriate or broken responses, please let us know:

Email: support@cynicalmonkeys.com

Subject Line: "Eleanor AI Issue Report"

Include:

  • Description of the problem
  • What you typed (your input)
  • What Eleanor said (her response)
  • Screenshot (if possible)

Because we don't collect conversation data, we rely on voluntary reports from players to identify issues.

Your report helps us:

  • Improve Eleanor's character consistency
  • Refine content filtering
  • Fix immersion-breaking behaviors

10.3 General Support

For bugs, crashes, or gameplay issues unrelated to AI:

Email: support@cynicalmonkeys.com

Steam: Community Hub > Bug Reports

10.4 Legal Inquiries

For legal inquiries or compliance questions:

Email: support@cynicalmonkeys.com

Mailing Address:

CynicalMonkeys

Attn: Legal Department

1705 W LeMoyne St

Chicago, IL 60622

ADDITIONAL RESOURCES

  • EULA (End User License Agreement): https://www.cynicalmonkeys.com/eula
  • Privacy Policy: https://www.cynicalmonkeys.com/privacy
  • Community Guidelines: https://www.cynicalmonkeys.com/community
  • Technical Support FAQ: https://www.cynicalmonkeys.com/support
  • Llama 3.2 Model Card (Meta): https://ai.meta.com/llama/
  • Llama 3.2 License: https://www.llama.com/llama3_2/license/

A NOTE FROM THE DEVELOPERS

Eleanor represents a new approach to AI in games: privacy-first, player-controlled, and completely local.

Most AI-powered games send your conversations to cloud servers, log your interactions, and use your data to train models. We built Eleanor differently because we believe:

  • Your conversations are yours alone. What you say to Eleanor is between you and your computer.
  • Privacy shouldn't be a premium feature. Every player deserves complete privacy by default.
  • Local AI is the future. As models get smaller and more efficient, there's no reason to sacrifice privacy for quality.

We know the AI isn't perfect. Eleanor will sometimes say weird things, break immersion, or fail to understand you. That's the nature of working with language models—they're powerful but unpredictable.

We can't improve what we can't see. Because we don't collect conversation data, we rely on your voluntary feedback to make Eleanor better. If she says something problematic or confusing, please email us (see Section 10.2). Your reports help us refine the AI for everyone.

Thank you for supporting privacy-respecting gaming. We hope Eleanor becomes a memorable part of your experience in the manor—without compromising your privacy.

— CynicalMonkeys Development Team

Last Updated: February 7, 2026

Version: 1.0 (Early Access)

Document ID: AI-DISCLOSURE-ZERO-001

CynicalMonkeys - All Rights Reserved