README.md

---
title: Beyond Logic Labs
emoji: 🎮
colorFrom: blue
colorTo: purple
sdk: static
pinned: false
---

# Beyond Logic Labs

We build AI systems for games that respect player intelligence.

## Loreguard - Knowledge-Bounded NPCs

**Loreguard** is an NPC engine that ensures game characters only say what they actually know. No hallucinations, no breaking immersion.

### The Problem

```
Player: "Where is the secret base?"

Traditional LLM: "The secret base is in the northern mountains..."
                 (hallucinated - NPC shouldn't know this)

Loreguard NPC:   "I don't know anything about a secret base."
                 (grounded - respects knowledge boundaries)
```

### How It Works

- **Fine-tuned LLMs** - LoRA-trained on NPC dialogue with citations
- **Multi-pass verification** - RoBERTa NLI checks claims against knowledge
- **RAG retrieval** - Semantic search over character's actual knowledge
- **Local inference** - Runs on player's hardware via llama.cpp

## Models

| Model | Description |
|-------|-------------|
| [loreguard-vanilla-gguf](https://huggingface.co/beyond-logic-labs/loreguard-vanilla-gguf) | General NPC model with Unsloth Dynamic quantizations |

## Quick Start

```bash
pip install loreguard-cli
loreguard --token YOUR_TOKEN
```

## Links

- [Website](https://beyondlogiclabs.com/loreguard)
- [GitHub](https://github.com/beyond-logic-labs)

---

*Building games where NPCs feel real.*