Skip to main content

Model Agents and Usage

SkyrimNet uses specialized language models for different tasks that collectively simulate intelligent NPC behavior. These models coordinate to deliver a rich, reactive Skyrim experience, from dialogue to memory, action selection, and emotional response.

There are six of them. You can configure model providers and parameters in:
🔧 Advanced Configuration > OpenRouter

✨ Default Model​

The central model responsible for generating spoken lines and character dialogue. Use a good, smart model, suitable for roleplay.

🗨️ Text Generation​

  • Converts scene, memory, and personality context into natural dialogue.
  • Uses retrieved memories and character profiles to remain consistent and grounded.
  • Can speak to the player, to other NPCs, or monologue when alone.

🎮 Game Master​

Acts like a Dungeon Master, guiding the scene and ensuring the world feels alive.

🎬 Narrative Scene Control​

  • Creates ambient interactions, especially between NPCs.
  • Responds to changes in the world state or player behavior.
  • Manages NPC–NPC chatter, conflict, or alliances.

🧠 Memory Generation Model​

Handles memory creation, summarization, and storage.

📝 Memory Creation​

  • Summarizes recent event streams into first-person memories.
  • Assigns:
    • importance_score
    • emotion
    • tags
    • type (TRAUMA, EXPERIENCE, etc.)
  • Used later for recall, relationships, and mood shaping.

📍 View in-game under: UI > Memories

🔍 Memory Search Model​

Optimized for fast, accurate memory recall when context is needed.

🔎 Query Generation​

  • Builds search queries using:
    • Semantic vector similarity (MiniLM-L6-v2)
    • Keyword match filters (tags, actors, locations)
  • Feeds the most relevant memories into the Default Model for improved dialogue quality.

🧬 Character Profile Generation Model​

Creates and updates NPC identity data.

🧾 Profile Creation​

  • Auto-generates profiles for modded or uninitialized NPCs.
  • Includes:
    • Goals
    • Personality traits
    • Background
    • Relationship tendencies

🔄 Dynamic Updates​

  • Modifies existing profiles over time based on:
    • Important memories
    • Recent emotional shifts
  • Allows characters to evolve naturally.
  • 🛠 Configurable at: Advanced Configuration > DynamicBio

🎭 Action Evaluation Model​

Decides what an NPC does after speaking. The model just needs to be "decent", no need for an expensive one.

🎯 Action Selection​

  • Chooses appropriate gameplay actions tied to dialogue intent.
    • Examples: FollowPlayer, SlapTarget, PickUpItem
  • Prevents dialogue from feeling disconnected from in-game behavior.

✋ Gesture Generation​

  • Selects expressive gestures or animations that match speech tone.
    • e.g., nodding, shrugging, dramatic motions

⚔️ Combat Evaluation Model​

When the characters are currently engaged in combat, instructs the LLM to prioritize recent combat-related events and dialogue in its responses.

  • Make NPC speech reactive to active combat.
  • Emphasize recent combat-related events such as:
    • Enemy actions
    • Player moves
    • Damage taken or inflicted
    • Tactical changes (retreat, push, heal)
  • Maintain immersion by blending character personality with battle context.

🧪 Meta Evaluation Model​

Performs scene-level analysis to keep interaction flow realistic. Use a fast, exact model. Its tasks are small and frequent.

😐 Mood Evaluation​

  • Updates NPC emotional state based on:
    • Dialogue tone
    • Memories
    • Player actions
  • Affects:
    • Voice tone (XTTS)
    • Facial expressions
    • Decision-making

👥 Speaking Turn Selector​

  • Decides who should speak next:
    • Based on relevance, social rules, and proximity
  • Ensures realistic pacing (e.g., not everyone talks at once)

🔗 Coordination Between Models​

All models work in tandem:

  1. Memory Search finds what’s relevant
  2. Profile & Mood Evaluation sets emotional/character context
  3. Default Model generates dialogue using that context
  4. Action Model makes the NPC act it out
  5. Game Master spices up the world with ambient interaction