Getting started

NJORD Engine is bring-your-own-keys (BYOK). You supply the AI, so costs go straight to the provider with no markup from us, and your keys never leave your machine. You can also run everything locally with no cloud key at all.

1. Get the app

Download the latest Windows build from the releases page (early testing). Linux and macOS are coming.

2. Choose how the AI runs

You have two ways to power the agents that build your game, and you can mix them.

Option A: Cloud (bring your own key)

An Anthropic (Claude) key is the default for the build agents. Paste it into Settings ▸ AI providers. Each agent (orchestrator, code, story) can use its own provider, so you can mix and match, and cloud models are currently the strongest for heavy reasoning.

Option B: Local / OpenAI-compatible

Prefer to keep everything on your machine? Point NJORD at any OpenAI-compatible endpoint, including Ollama and LM Studio (or OpenAI itself):

1. Install Ollama or LM Studio and pull a capable coding model, then start its local server (Ollama on http://localhost:11434, LM Studio on http://localhost:1234).
2. In Settings ▸ AI providers, pick a capability, set the provider to Local / OpenAI-compatible, and enter that endpoint URL.

With a local endpoint there's no key, no per-token cost, and nothing leaves your computer.

3. Optional: a 3D key

To generate 3D models from a description, add a Meshy or Tripo key. Or skip it and bring your own glTF models.

4. Describe and build

Prompt your game, refine it, and export a native build. When you're ready to go deeper, the Docs cover every part of the engine.

On keys and privacy: cloud keys are stored in your OS keyring, never in project files and never sent to us. With local models, nothing leaves your machine at all. You can also work with no AI and edit everything by hand.