Game Model Development Studio — Emotional AI Engine for NPCs

NPCs that think.
Games that remember.

Gamemodes applies behavioral psychology to NPC behavior — an AI-native engine that generates psychologically authentic, dynamic dialogue across 6 projects and 79K+ lines of code. Runs on consumer hardware with 1-bit quantized models.

Explore Projects ↓ Get in Touch
6
Active Projects
79K+
Lines of Python
22K+
Training Records
~1GB
VRAM at Runtime
Live
NPC Demo
The Technology

Game NPCs deserve better than canned lines.

For decades, NPCs have been limited to pre-written dialogue trees. Gamemodes applies multi-layered behavioral psychology to NPC behavior. The engine generates contextually aware, psychologically grounded dialogue in real time, at a cost and compute footprint that works on any gaming PC.

🧠

Psychology Engine

NPCs are driven by multi-layered personality profiles — protective, reactive, and vulnerable behavioral states that govern behavior based on emotional state and relationship context. Every response reflects the character's internal behavioral system.

Runs on Consumer Hardware

Our engine uses 1-bit quantized models (~1GB VRAM) to run entirely on local hardware — no cloud dependency, no latency, no per-call API costs.

🔧

Game-Agnostic Engine

The core engine is designed for general applicability. We've proven it across three distinct game worlds — the architecture adapts to any game genre or setting.

/* Core Data Flow */ Game EngineStates + Profile + ContextPrompt BuilderLLMNPC Dialogue /* What the engine tracks per NPC: */ Behavioral State System (protective, reactive, vulnerable) 7 Physiological States (stress, hunger, thirst, ...) 3 Psychological Modes (protective, reactive, vulnerable) 20 Personality Archetypes (base traits, speech patterns) 7 Relationship Levels (hostile to bonded) 5 Awareness Tiers (emotional regulation capacity) 8 Game Genres (noir, fantasy, sci-fi, ...)
🔀

Provider Pattern Architecture

Model-agnostic design. Swap between any OpenAI-compatible LLM at runtime — from 1-bit models to frontier models — without changing a line of game code.

🛡️

Graceful Degradation

If the LLM is unavailable, the engine falls back to curated dialogue organized by psychological mode. The game never breaks.

📊

Training Data Pipeline

Integrated pipeline for generating, validating, and filtering fine-tuning data — producing thousands of high-quality Alpaca-format examples.

Projects

Six projects. One engine.

Gamemodes operates across six interconnected projects — each proving a different dimension of the psychology engine's capability.

🌃 Testbed Game — Phase 12

Shadow City

Shadow City is a noir RPG designed to showcase the Gamemodes NPC engine. Set in a rain-soaked dystopian metropolis where six factions wage silent war, it demonstrates the full capability of the psychology engine — quest systems, faction progression, mortality, and dynamic NPC-NPC interactions — all powered by psychologically grounded dialogue generation.

  • Phase 12 — 6 factions with full game loop and victory conditions
  • Cell phone, comedy, entertainment, tutorial, and asylum engines
  • Full quest tree system with pool-based job assignment
  • Medical, mortality, and addiction simulation systems
  • Tick system in development — duplicate event bugs being resolved
faction "Police" {
  theme: "Internal corruption vs. duty",
  victory_quest: "Operation: Takedown",
  mortality_rate: 0.004,
  ranks: ["Recruit", ..., "Leader"]
}

// NPC response driven by psychological state
npc.generate_dialogue({
  stress: 72,
  psychological_mode: "reactive",
  affinity: -1
}) // → contextually generated response
⚔️ Skyrim Integration — Active Development

Skyrim NPC Dialogue Engine

A Skyrim Special Edition mod that replaces canned voice lines with real-time, psychologically grounded dialogue. Players type messages to NPCs via a console command; the engine generates responses through a C++ SKSE plugin communicating with a local Python dialogue server with dual-provider architecture.

  • SKSE C++ plugin: http_client, npc_resolver, papyrus_functions, subtitle_display
  • Full LLM server with dual provider, caching, and priority routing
  • Lydia-1B: First fine-tuned NPC model (Llama 3.2 1B + QLoRA)
  • MiMo v2 Pro data generation pipeline validated (100% pass rate)
  • NPCs across all major Skyrim factions with rich personality profiles
// SKSE Plugin → Python Server → LLM
Player: "talk lydia 'How are you?'"

// 1. C++ plugin resolves NPC
npc = NPCResolver.ResolveByName("lydia")

// 2. HTTP POST to localhost:8080
HttpClient.SendDialogue({
  npc_id: "lydia",
  stress: 35,
  psychological_mode: "protective"
})

// 3. Response displayed as subtitle
"I am sworn to carry your burdens,
 but tonight... I wonder who carries mine."
☢️ Fallout 4 Integration — Only Trained Model

Fallout 4 NPC Dialogue Engine

The only project with a fully trained model in the portfolio. A Fallout 4 mod using F4SE plugin to capture player input, route through a Python dialogue server, and generate wasteland-appropriate dialogue. Nine NPCs across five factions with full psychological profiles and an affinity system.

  • F4SE C++ plugin with fine-tuned 4B personality model (checkpoint-250)
  • 9 NPCs across Brotherhood, Institute, Railroad, Raiders, Settlers
  • 22,250+ Alpaca-format training entries
  • Full affinity system tracking relationship progression
  • Faction-specific psychological templates
// Personality model: fine-tuned 4B checkpoint-250
model: "fallout4-personality-4b"

// Factions with NPC profiles
brotherhood { vibe: "disciplined, hierarchical" }
institute { vibe: "detached, clinical" }
railroad { vibe: "devoted, paranoid" }
raiders { vibe: "chaotic, impulsive" }
settlers { vibe: "resilient, weary" }

// Training data: 22,250+ entries
npc_dialogue_profiles: 2,065
faction_voice: 1,501
affinity_modulated: 1,500
moral_dilemma: 800
... and 5 more categories
⚡ Core Engine — Pipeline Ready

Gamemodes01 — The Training Pipeline

The core fine-tuning and engine project. Trains lightweight LLMs to generate psychologically authentic NPC dialogue, with a complete data pipeline from generation through deployment. Pipeline is built and ready — data generation is the current focus.

  • Training pipeline built — data generation in progress
  • Data schemas, seed data, fine-tuning scripts, annotation guidelines ready
  • LLM judge quality filtering across 5 scoring dimensions
  • Alpaca format conversion with augmentation and multi-template support
  • 8 game genres supported: noir, fantasy, sci-fi, post-apocalyptic, and more
// Training pipeline
generatevalidatemergefilterfine-tune

// Quality scoring (5 dimensions)
behavioral_authenticity: 0-1
state_visibility: 0-1
archetype_voice: 0-1
genre_consistency: 0-1
affinity_accuracy: 0-1

// Auto-accept if avg ≥ 4.5/5
// Flag for review if 3.0-4.5
// Reject if avg < 3.0

Talk to Our Characters

Chat with NPC personalities powered by the Gamemodes engine. Each character has unique traits, backgrounds, and dynamic relationship progression.

Roadmap

Building the future of NPC intelligence

Our phased plan to move from research prototype to production deployment and enterprise API.

Phase 1 — Foundation Q2 2026 — Next 3 Months

Foundation

  • Complete training data collection for Gamemodes01
  • Train Gamemodes01 model based on Meta-Llama-3.2-1B
  • Implement testing framework for live deployment
  • Launch NPC demo service
Phase 2 — Live Deployment Q3 2026 — 3-6 Months

Live Deployment

  • Deploy Gamemodes01 for live NPC service
  • Build enterprise deployment plan
  • Begin Gamemodes02 research (Bonsai models + Bankai training protocol)
  • Scale character library across game universes
Phase 3 — Frontier Q4 2026+ — 6+ Months

Frontier

  • Gamemodes02 deployment with next-gen training
  • Frontier AI research tracking and integration
  • Expand to new game engines and platforms
  • B2B behavioral engine API launch
Coming Soon

$GMODE Token

$GMODE is the access token for the Gamemodes Behavioral Engine API. A software utility asset powering NPC intelligence at scale.

Token-2022 on Solana API access payment for behavioral engine inference Tiered access: Free / Standard / Professional / Enterprise
📝 Development Log

Latest Updates

A chronological view of milestones, launches, and stabilization work across the Gamemodes portfolio.

2026-04-09

Wendy Speaks + Investor Page + Company Spokesperson

  • MiMo TTS voice synthesis live — Wendy now speaks with 🔊 buttons on every response
  • Wendy serves as company spokesperson with tiered project knowledge
  • New investor relations page with tokenomics, roadmap, and transparency dashboard
  • Xiaomi MiMo-V2-TTS integration with auto-play toggle
2026-04-08

Website Restructure + Roadmap Launch

  • Navigation consolidated from 11 tabs to 7
  • 3-phase roadmap added (Q2–Q4 2026)
  • $GMODE token "Coming Soon" card live — Solana Token-2022
  • Mobile responsive hamburger menu added
2026-04-06

Portfolio Evaluation Complete

  • Full portfolio assessment across 6 projects
  • Grade: B+ (7.6/10), Estimated value: $50K-$150K
  • Behavioral framework validated as core differentiator
  • Priority: Generate training data, fix Shadow City tick bugs
2026-04-05

Patent Strategy & Shadow City Stabilization

  • Patent application guide for emotionally-aware dialogue system
  • Shadow City Phase R6 stability work
  • Balance testing framework for NPC behavior validation
  • 11 faction playthroughs completed across 5 archetypes
2026-04-04

Gamemodes Site Launched & Wendy Live

  • GitHub Pages deployment at gamemodes.xyz
  • Wendy chatbot live at chat.gamemodes.xyz
  • Live stats integration between sites
  • Wendy Live Demo Integration plan finalized
2026-04-03

Wendy v2 Development & Shadow City Expansion

  • Complete Wendy rewrite with modular architecture
  • Affinity system, LLM client, database, session management
  • Shadow City Phase 7+: Cell phone, comedy, entertainment engines
  • NPC extras system with 11 trait categories
2026-04-09

Lydia-1B Model Investigation & Training Data Pipeline

  • Completed comprehensive Lydia-1B model investigation
  • Root cause identified: body language failure due to narrow gesture vocabulary
  • Validated MiMo v2 Pro API for training data generation (100% pass rate)
  • Updated quality standards with body language diversity requirements
  • Prepared 800-1500 example generation pipeline for model retraining
2026-04-02

Behavioral Engine v2 & Skyrim Server Overhaul

  • Behavioral Engine v2 with 3-layer personality depth
  • Affinity-driven emotional depth (protective, reactive, vulnerable)
  • Skyrim LLM server with dual provider, caching, priority routing
  • Dynamic prompt tier system
2026-04-01

Skyrim SKSE Plugin & Multi-Project Kickoff

  • Skyrim SKSE plugin foundation (http_client, npc_resolver, papyrus)
  • Emotionally-aware dialogue system with NPC resolver and subtitle display
  • Fallout 4 F4SE plugin scaffolding
  • Multi-project architecture established
View Full Development Log →

Let's build the future of NPCs.

Whether you want to contribute, invest, or just learn more — we'd love to hear from you.