Configuration¶

GOD separates machine settings from experiment selection.

`.env`¶

.env stores local model, port, and startup settings. It is created from .env.example and ignored by Git.

Required model settings:

Variable	Purpose	Default
`GOD_LLM_API_KEY`	OpenAI-compatible API key	empty
`GOD_LLM_API_BASE`	OpenAI-compatible API base URL	`https://api.openai.com/v1`
`GOD_LLM_MODEL`	Chat model used by GOD and the runtime	`gpt-5.4`
`GOD_EMBEDDING_MODEL`	Embedding model	`text-embedding-3-large`

`.god/current_experiment.json`¶

The setup wizard writes the active experiment here. start, open, and new-run read this file so .env does not accidentally decide the current map or scenario.

Explicit overrides¶

For one-off runs, these environment variables override current-experiment state:

GOD_EXPERIMENT=pku_trump_visit GOD_EXPERIMENT_RUN=1 ./scripts/god.sh new-run

Common ports¶

Variable	Default	Service
`GOD_BACKEND_PORT`	`8001`	FastAPI backend
`GOD_FRONTEND_PORT`	`5174`	GOD control room
`RUNTIME_AGENT_PORT`	`19092`	JiuwenClaw agent WebSocket
`RUNTIME_WEB_PORT`	`20000`	JiuwenClaw web service
`RUNTIME_GATEWAY_PORT`	`20001`	JiuwenClaw gateway
`RUNTIME_UI_PORT`	`6173`	Runtime UI

Browser opening¶

Disable automatic browser opening:

GOD_OPEN_BROWSER=0 ./scripts/god.sh start