Try it yourself

Chat with Zero, a hacker NPC. Browse his knowledge files on the right to verify every claim is grounded.

#underground -- Zero 9B open-weights model

* Now talking in #underground

* Zero is online

Intent classification

RAG retrieval

Internal monologue

Speech synthesis

NLI verification

knowledge

What Loreguard can offer?

zero inference costs

Grounded in your world

Upload your game bible, lore documents, character sheets. NPCs only speak from what you've written.

                  lore/

                    world-history.md

                    factions.md

                    characters/

                      elder-mira.yaml

Verified against your lore

Every NPC statement is checked against your documents using NLI. Contradictions get filtered out before reaching the player.

Player GPU inference

The model runs on the player's machine. Min 6GB VRAM (RTX 2060+).

Trust-gated secrets

Gate NPC knowledge behind trust levels. Players build relationships to unlock information. Social engineering and interrogation built in.

                "trust":
                75,
                "secrets_unlocked": ["hideout"]
              

Works with any engine

Unity, Godot, Unreal. Anything that speaks HTTP or WebSocket. Cloud or bundle, same API.

                # Cloud: connects via WebSocket

                # Bundle: localhost:8090

                POST
                /v1/chat

Persistent memory

NPCs remember past conversations and build knowledge about each player over time. Relationships evolve across sessions.

Cloud or Bundle: you choose

Cloud: we orchestrate the pipeline, you focus on NPCs. Bundle: everything runs on the player's machine, no cloud at runtime. Both run inference on the player's GPU.

Multi-pass verification

RAG retrieval, grounded generation, NLI check, trust gate, safety filter. Five stages before any response reaches the player.

Inference runs on the player's GPU. No per-token API costs. Works offline.