Files
wehub-resource-sync aa1a5d5763
CI / Release hygiene (push) Waiting to run
CI / Python targeted tests (push) Waiting to run
CI / Frontend build (push) Waiting to run
Pages / Build and deploy public pages (push) Waiting to run
chore: import upstream snapshot with attribution
2026-07-13 13:18:17 +08:00

3.8 KiB

GOD Quick Start

The shortest path from a clean machine to the setup wizard and then a live GOD control room.

Chinese version: QUICKSTART.zh-CN.md


1. Prerequisites

You will need:

  • macOS/Linux: Python 3.11+, Node.js & npm, uv, and screen
  • Windows: PowerShell 5.1+ and winget; the startup script installs missing Git, Node.js LTS/npm, and uv

macOS:

brew install python node uv screen

If git clone is not available on Windows yet, install Git first with winget install --id Git.Git -e --accept-package-agreements --accept-source-agreements or download the repository ZIP.

2. Clone

git clone https://github.com/XiaoLuoLYG/GOD.git
cd GOD

3. Start

macOS/Linux:

./scripts/god.sh start

Windows PowerShell:

.\scripts\god.cmd start

start is the normal one-command path. It is idempotent, so running it again reuses services that are already up.

On first run, GOD will:

  1. Create .env from .env.example.
  2. Install Python + Node dependencies.
  3. Start the setup backend/control room.
  4. Open the browser setup wizard at /setup.
  5. Wait for you to save model settings and choose an experiment.
  6. Bring up the full stack for that current experiment, create a live session, run the first step, and open the control room.

GOD setup wizard

In the wizard you can choose one of three paths:

  • Open GOD Town launches the built-in The Ville baseline.
  • Open PKU Trump Visit launches the built-in PKU campus experiment.
  • Create Custom Experiment lets you describe a scenario, generate an editable experiment draft, adjust agents/steps, and then click Save and Launch.

Three settings are required:

Variable Example
GOD_LLM_API_KEY sk-...
GOD_LLM_API_BASE https://api.openai.com/v1
GOD_LLM_MODEL gpt-5.4

Any OpenAI-compatible endpoint works. The API key is saved only in local .env; the browser only receives redacted status. Experiment selection is saved separately in .god/current_experiment.json, so .env does not control which map or experiment starts.

To switch between built-in experiments or create another experiment later, run:

./scripts/god.sh configure

4. Open the control room

When startup finishes, the script opens the control room for the current experiment and prints a URL like:

http://127.0.0.1:5174/pixel-replay/god_town/1

If the browser did not open automatically, open that URL. You should see the pixel town, the resident roster, step controls, and the live console.

5. Verify

./scripts/god.sh status

You should see every service marked up.

6. Restart or run fresh

Use restart when you want a clean process restart without wiping the current run:

./scripts/god.sh restart

Use new-run when the UI shows old replay data or you want a fresh live session:

./scripts/god.sh new-run

new-run prints the current experiment run directory, stops services, wipes only that current run state, and starts a clean live session.

7. Day-to-day commands

./scripts/god.sh start     # idempotent; reuses running services
./scripts/god.sh setup     # install/check dependencies only
./scripts/god.sh configure # switch built-in experiments or create a custom one
./scripts/god.sh restart   # stop everything cleanly, then start again
./scripts/god.sh new-run   # wipe the current experiment run and start fresh
./scripts/god.sh status    # print URLs, ports, and model status
./scripts/god.sh stop      # stop everything
./scripts/god.sh tail      # follow logs
./scripts/god.sh open      # open the frontend pages in the browser

On Windows, replace ./scripts/god.sh with .\scripts\god.cmd.