36 AI RPG Dialogue

Inspiration

Modern RPGs still rely on hand-written dialogue trees that are slow to author and identical for every player. We wanted NPCs that feel fast, personal, and alive—without breaking console budgets or production pipelines.

What it does

36 AI RPG Dialogue creates NPC conversations that respond instantly and adapt to who the player is. A fast reflex agent reacts immediately, while a deeper AI response runs in parallel. The same NPC changes tone based on the player’s character sheet.

How we built it

We split NPC cognition into parallel agents: a low-latency reflex layer and a deeper dialogue layer. Player character sheets drive behavior instead of dialogue trees. The system is provider-agnostic and uses modular agents with voice input and output.

Challenges we ran into

Managing latency without relying on gated “thinking” models was tricky. Designing the system so it felt responsive while still delivering meaningful dialogue required careful orchestration and timing.

Accomplishments that we’re proud of

NPCs feel instant without being scripted

Dialogue adapts socially without branching trees

The same architecture supports small models and future fine-tuning

Clean separation between agents, UI, and AI providers

What we learned

Perceived speed matters more than raw model speed. Structuring context and timing produces better results than relying on larger or “smarter” models alone.

What’s next for 36 AI RPG Dialogue

We plan to generate large-scale character-dialogue datasets to train SLMs or LoRA adapters, reduce costs further, and enable fully offline or console-friendly AI NPCs.

Built With

• cline
• openai