Choosing your AI setup

All text, image, and voice generation in Aikami flows through one abstraction — AiProviderGateway — so the game never cares which mode is active. That means you can mix and match freely, and change your mind later without touching your saves.

The three modes

Mode	What it means
Offline / local	Ollama (text), ComfyUI (image), Kokoro (voice), running as Docker services on your own hardware
BYOK	Your own key for Anthropic, OpenAI, Gemini, ElevenLabs, Stability AI, or any OpenAI-compatible endpoint
Service (coming soon)	Fully managed pay-as-you-go hosting — no GPU, no Docker, no setup

What’s actually required

A text engine is required to play. The Game Master is the core of the game; without a text model there’s nothing refereeing your actions.

Image and voice are optional. The LPC sprite system covers the visual baseline with zero AI dependency, so a text-only setup is a fully supported configuration, not a degraded one.

Picking a text provider

Local (Ollama / vLLM) — free to run, private, works fully offline. Quality and speed depend entirely on the model and hardware you choose. This is the demanding part of a local setup; plan for a GPU.
BYOK cloud — best quality per unit of effort, no hardware requirements, and you pay the provider directly at their rates. Aikami is never a proxy in that path — your key and your prompts go straight to the provider you picked.

A common arrangement is a cloud model for text (where GM quality matters most), local ComfyUI for images (where you’d otherwise pay per generation), and local Kokoro for voice.

Images and voice

Images — ComfyUI locally by default; DALL·E, Stability AI, NovelAI, and fal.ai via BYOK. See Image Generation for style profiles and contextual triggers.
Voice — Kokoro locally by default; ElevenLabs, OpenAI TTS, VOICEVOX, and Fish Speech via BYOK. You can assign voices per NPC.

Where to configure it

Everything lives under Settings in the client — providers, keys, endpoints, and per-NPC overrides. Keys are stored client-side.