Configuration¶

World Simulation Engine has two configuration layers:

1. Startup configuration is read from environment variables. These settings let the backend start, find Neo4j, write media files, and set service-wide limits.
2. Runtime service configuration is managed in the web interface. These settings decide which model providers, prompts, image services, and voice services a world or simulation uses.

The first layer belongs in .env, Docker Compose, or your shell. The second layer is stored in Neo4j and can be changed without editing environment variables.

Configuration sections¶

• Environment variables: backend startup settings and service-wide limits.
• Connections: reusable provider endpoints and API keys.
• LLM model configs: chat model profiles and per-provider generation settings.
• Embedding configs: retrieval and memory embedding profiles, including the expanded embedding provider support.
• Image generation configs: ComfyUI image profiles and image generation behavior.
• TTS configs: AllTalk backend configs and per-simulation voice generation.
• STT configs: whisper.cpp speech recognition settings.

In practice, start with environment variables, create provider connections, create model/media configs, then assign those configs to worlds or simulations.

Runtime hierarchy¶

Runtime settings are intentionally reusable:

• Connection: the network endpoint and optional API key for a provider.
• Model config: a reusable model profile for one service type, such as an LLM model with sampling settings or a ComfyUI image config with image dimensions.
• World assignment: default model, image, embedding, TTS, and prompt assignments for a world.
• Simulation assignment: overrides for a specific simulation. Use this when one run needs different models, image behavior, or prompts from the world default.

Open the web interface, then go to Configurations or Connections from the main navigation.