Your Characters.
Their Stories.
Living NPCs that ship inside your game.
No cloud. No latency. No monthly bills.
Every NPC in your game
says the same 5 lines.
Pre-written. Static. Players stop listening after 2 minutes.
Dynamic. Context-aware. Every conversation is unique.
Dialogue examples from books & screenplays
Character archetypes built in
Smallest model - fits in any game
A new paradigm
For decades, they were Non-Playing Characters.
Scripted. Static. Forgettable.
VoxLore changes that. Your characters remember. They react. They feel. They gossip, argue, and form opinions. They are no longer NPCs.
They are Playing Characters.
Three steps to living characters
Build Your World
Describe your setting, factions, history, and lore. The model uses this as ground truth so every NPC stays consistent with your world.
Create Characters
Define personality via OCEAN traits, speech patterns, knowledge boundaries, and relationships. Each character feels distinct and memorable.
Ship It
Drop the model into Godot, Unity, or Unreal. Runs locally via llama.cpp or ONNX. No server, no API keys, no internet required.
One event. Every NPC reacts differently.
Push world events into the SDK and watch your characters respond naturally - each filtered through their own personality, fears, and ambitions. The world feels alive because it is.
"That bridge was the only route for supply wagons. I need to send scouts east to find an alternate path before the garrison runs out of provisions."
"Were there travelers on the bridge when it fell? Please, if anyone is hurt, bring them to the chapel. I have herbs enough for a dozen wounded."
"Collapsed, you say? That means the merchant caravans will detour through the forest road. Less guards, more confusion... could be profitable."
Personality-Filtered Reactions
Events are filtered through each NPC's OCEAN traits. A high-neuroticism character panics while a low-neuroticism one shrugs it off. Same event, completely different reactions.
Works in Every Conversation
Events flow into both player-NPC and NPC-to-NPC conversations. Your world's background chatter evolves alongside the main story without any extra scripting.
Everything NPCs need to feel real
Not just better dialogue. A complete system for characters that see the world, react to events, talk to each other, and animate with purpose.
Animation Constraints
Define your animation vocabulary per character. The model outputs machine-readable flags your engine maps to clips - not English descriptions you have to parse.
NPC-to-NPC Conversations
NPCs talk to each other autonomously. Define conversation pairs, set a timer, and watch your world come alive with background chatter that responds to events. Player requests always take priority.
Game State Awareness
NPCs reference inventory, quests, location, time, and reputation. They notice the broken sword you carry, remember favors you did, and warn you about dangers in the area.
World Events System
Push events into NPC context with a single API call. The model filters each event through personality traits - warriors worry about threats, healers worry about safety, rogues see opportunity.
Every field drives your game engine
Not just dialogue text. Every response includes emotion, facial expression, body language, and gesture data your engine can parse directly into animations.
Maps to audio tone parameters. Your TTS or bark system uses this to adjust pitch, speed, and timbre.
Direct mapping to facial rig blend shapes. Brow, eyes, and mouth values drive your character's expression in real-time.
Body maps to animation state machine states. Gesture triggers one-shot animations layered on top.
Ship-ready engineering
Four battle-tested techniques stolen from production LLM infrastructure and embedded directly into every SDK. No cloud required - your game handles everything locally.
Stream Cancellation
Player skips dialogue mid-sentence? The SDK cancels generation instantly and frees the GPU slot in under 50ms. No wasted compute, no stale responses blocking the next conversation.
Tool Calling for Game State
Stop cramming inventory, quests, and world state into the system prompt. NPCs request only what they need via structured tool calls. Cleaner prompts, smaller context, better responses.
Emergency Fallback
If the local model crashes, stalls, or runs out of memory, the SDK downgrades gracefully through tiered quality modes. At the lowest tier, pre-written character-banked lines keep the scene playable. Your game never breaks.
Custom Schema Extensions
Bolt on your own fields - relationship deltas, knowledge reveals, quest hints, combat intent - without forking the SDK. Extensions are enforced at the schema level so the model always produces valid, typed data.
One model. Three sizes.
Pay once, ship forever. No subscriptions, no per-player fees, no usage limits.
Full narrative model - unlimited archetypes
- Highest dialogue quality and coherence
- Full OCEAN personality expression
- Long-term conversation memory
- All 20+ archetypes included
- Recommended: 8 GB+ RAM
Perfect for indie games
- High dialogue quality
- Personality-aware responses
- Shorter context window
- 15 archetypes included
- Recommended: 4 GB+ RAM
Per-archetype models for mobile & Steam Deck
- Functional dialogue generation
- Basic personality traits
- Fastest inference speed
- Pick individual archetypes
- Recommended: 2 GB+ RAM
Enterprise - Need custom archetypes, fine-tuning on your own lore, or volume licensing? Let's talk.
Same JSON. Every engine.
One character definition works across all supported engines. Write once, ship everywhere.
Ready to build living characters?
Start defining your first character in minutes. No sign-up required during early access.