Frequently asked questions.

Aikami is an open-source AI RPG engine that fuses tabletop-style D&D mechanics with LLM-driven roleplay. Every NPC is procedurally generated — archetype, personality, a full six-stat ability score block, backstory, and expressions — and an AI Game Master uses those stats to referee what you do: skill checks, persuasion, combat, all of it.

It’s built game-first, not chat-first: you launch into a spatial 2D world rendered with PixiJS, not a chatbot dashboard. The AI narrates; deterministic rules decide.

Aikami itself is free and MIT licensed — there is no subscription and no paid tier required to play.

What can cost money is the AI you point it at. If you run local models (Ollama, ComfyUI, Kokoro) on your own hardware, the whole thing is free. If you bring your own cloud API key, you pay that provider directly at their rates — we never sit in the middle of that transaction.

A managed pay-as-you-go hosting tier is planned for people who’d rather not run their own AI, but it will never be required. Self-hosting will always be a supported option.

No. You describe what your character does in plain language — “I try to talk the guard into letting us through” — and the AI Game Master decides whether it’s feasible, which check to roll, and how the world responds. There are no skill menus to memorize and no combo system to learn.

The D&D-style mechanics underneath (ability scores, skills, HP, AC) are there so outcomes are decided by consistent rules instead of the AI’s mood. You can look at them if you want the detail, or ignore them entirely and just play.

Yes, if you’re running local AI. Campaigns, saves, and chat history live in a local Turso (libSQL) database, and the game boots and plays with no network connection. Run the full Docker stack — Ollama for text, ComfyUI for images, Kokoro for voice — and nothing needs to leave your machine after the initial pull.

If you’re using BYOK with a cloud provider instead, you obviously need a connection to reach that provider. Your saves are still local either way; cloud sync via Firebase is optional and never a boot dependency.

Yes. Every world in Aikami — maps, tilesets, sprites, props, NPCs, items, dialogue, quests, encounters, factions — is defined in a structured content pack, not hardcoded into the engine. Packs are plain files: copy one, change a line, and the world shifts.

The pack that ships with Aikami, Emberwatch: The Fading Ward, is a full village built from a single manifest — an inn, a merchant, a persuasion-or-combat encounter, a faction system, and a branching quest. Packs can be generated by AI or written by hand.

The Hub — where creators will browse, upload, remix, and publish packs the way people share mods today — is still being built.

Not yet. Aikami is currently a single-player experience — you and the AI Game Master, with an AI-driven party alongside you.

The same world state does stream across your own devices: play in the browser, pick up on the desktop app, same world, same NPCs. That’s multi-device, not multi-player. Shared sessions with other people are not in the current roadmap phases; if it’s something you want, open a feature request — direction is genuinely still up for discussion at this stage.

All text, image, and voice generation flows through a single abstraction (AiProviderGateway), so you can mix and match:

• Text — Anthropic (Claude), OpenAI, Google Gemini, or any OpenAI-compatible endpoint. Locally: Ollama or vLLM.
• Images — ComfyUI locally by default; DALL·E, Stability AI, NovelAI, and fal.ai via BYOK.
• Voice — Kokoro locally by default; ElevenLabs, OpenAI TTS, VOICEVOX, and Fish Speech via BYOK.

A text engine is always required — that’s the core of the game. Image and voice are optional; the LPC sprite system covers the visual baseline with no AI dependency at all.

Yes — that’s the intended setup for a lot of people. Because everything goes through one provider gateway, the three pipelines are configured independently in Settings.

A common arrangement is a cloud model for text (where quality matters most for the Game Master), a local ComfyUI instance for images (where you’d otherwise be paying per generation), and local Kokoro for voice. You can also assign TTS and image providers per NPC if you want different characters to sound different.

Only if you want to run the AI models locally.

• BYOK or the web client — no GPU needed at all. The game itself is a 2D PixiJS renderer; any machine that runs a modern browser smoothly will run Aikami smoothly.
• Fully local (Docker stack) — you’ll want a GPU. Text generation via Ollama is the demanding part; image generation via ComfyUI more so. The exact requirements depend entirely on which models you choose to run.

You can also split the difference: local text on your own hardware, cloud image generation via an API key, or vice versa.

No — Docker is the easiest path to a fully local setup, not a requirement. There are four ways to run Aikami:

• Web, BYOK — no install at all. Open the hosted client, drop in an API key, play.
• Desktop app — a native Tauri v2 build for Windows, macOS, and Linux. Works with a local Ollama/vLLM instance or your own cloud keys.
• Docker — docker compose up aikami for the whole local stack (client + Ollama + ComfyUI + Kokoro), or aikami-client for just the client.
• From source — Bun + Moon, if you’re contributing or want to modify it.

Full commands are on the tech page.

The desktop app ships for Windows (.exe installer), macOS (universal .dmg), and Linux (AppImage). It’s built with Tauri v2, so it’s Rust-powered and small.

The web client is a Progressive Web App and runs in any modern browser, desktop or mobile.

One Linux note: the AppImage needs glibc 2.35+ and libfuse2 to mount. Some newer distros (Ubuntu 22.04+, Fedora 36+, Debian 12) ship libfuse3 instead and need libfuse2 installed separately — sudo apt install libfuse2 or sudo dnf install fuse-libs.

On your machine. Campaigns, saves, and chat history are stored in a local Turso (libSQL) database — that’s the source of truth, not a server-side copy.

Firebase (auth and optional cloud sync/backup) is layered on top and is never a boot dependency. Your world plays and saves fine without ever signing in. If you do sign in, sync is there so you can continue on another device — but it’s opt-in, not the default storage location.

No. When you use BYOK, your key is stored client-side and requests go directly from your client to the provider you configured — Anthropic, OpenAI, Gemini, or whatever endpoint you pointed it at. We are not a proxy in that path and we never see the key or the traffic.

The same goes for your prompts and the AI’s responses: with a local setup nothing leaves your machine at all, and with BYOK the conversation is between you and your provider under their terms.

If you’d rather verify than take our word for it, the client is open source — the provider gateway is all right there.

No — Aikami is early and moving fast. Expect rough edges, missing pieces, and breaking changes between releases. We’d rather say that plainly than have you find out after an hour of setup.

The core loop works today: chat interface, character cards, dice rolling, an AI Game Master, persistent NPC memory, content packs, and the 2D world renderer. Deeper systems — cross-chat memory, AI-generated lorebooks, richer relationship graphs, branching story tools — are actively being built.

If you hit a bug or want something to exist, open an issue or drop into Discord. Direction is still genuinely up for discussion.

Roughly, in order:

• Now — core loop: chat interface, character cards, dice rolling, AI backend, chat persistence.
• Next — D&D depth: character stats, world info/lorebooks, user personas, group chats.
• Later — living world: character relationships, chat summarization, AI-generated lorebooks, persistent world state, cross-chat memory.
• Future — premium & scale: branching stories, knowledge-graph visualization, voice cloning, and managed pay-as-you-go hosting for people who don’t want to self-host.

Open feature requests live on the feature label in GitHub Issues.