How LoreGuard works

Knowledge-grounded AI with multi-layer verification.

Why LoreGuard?

The problem with AI NPCs isn't intelligence. It's trust.

Grounded responses

Every NPC response is verified against your lore documents. Unverified claims are caught and removed before players see them.

Debug before shipping

See citations for every line of dialogue. Know exactly which documents informed each response before players see it.

Works your way

Cloud API for online games with per-request billing. Offline bundle for single-player with zero ongoing costs.

The pipeline

A multi-stage system designed to keep NPCs honest.

IN
Player Message
0
Intent Classify
BART zero-shot
1
Retrieval
LlamaIndex
1.5
Compaction
if >10K chars
1.7
Entity Resolve
"he" → name
2
Monologue
chain of thought
2.1
Relevance Filter
top sources
2.4
Claim Retrieval
self-RAG loop
2.5
Fact Check
NLI verify
4
Speech
character voice
4.5
Final Check
citations
5
Actions
state update
OUT
Response
grounded dialogue
6
Memory Gen
async reflection
6.5
Synthesis
10 mem → summary
W
Warmup
KV cache
P
Character
personality
V
VFS
notes, mail, irc
GPU
LLM Worker
llama-server
NLI
RoBERTa
WebSocket
S
State
trust, flags
LI
LlamaIndex
vector search
Retrieval
Generation
Verification
Output
Context
Memory

Technical deep dives

Read the engineering details behind LoreGuard.

Deployment options

Choose the architecture that fits your game.

Cloud API

For online games and multiplayer. We handle the infrastructure.

REST API with WebSocket support
Per-request pricing
Automatic scaling
Steam/Epic/custom auth integration

Offline Bundle

For single-player and offline games. One binary ships with your game.

Works with any engine - Unity, Godot, Unreal, whatever
WebSocket and API support
Zero per-player costs
No internet required

Questions?

Reach out and we'll help you figure out the right setup.

Get in touch