Bulk-provision Google accounts into 9router for Antigravity AI model access (ag/claude-*, ag/gemini-*).

Automates the same Google OAuth flow used by Antigravity CLI, Gemini CLI, and Cloud Code extensions — but headlessly handles login, consent screens, and token injection at scale.

Playwright (headed Chromium)
  → Google OAuth login (email + password + consent)
  → Localhost callback captures auth code
  → Exchange for access_token + refresh_token
  → Discover/provision Antigravity project (loadCodeAssist API)
  → Insert/update into 9router's providerConnections SQLite DB

After injection, the accounts are immediately available as ag/ model connections in 9router.

git clone https://github.com/rullxd/ag-login-inject.git
cd ag-login-inject
pip install -r requirements.txt
playwright install chromium

# Interactive menu — prompts for everything
python main.py

# Batch: inject 20 accounts from a file
python main.py -b emails.txt

# Parallel: 3 browsers at once (~12s/account vs ~30s sequential)
python parallel.py -b emails.txt -w 3

Passwords are resolved in this order (first wins):

1. --password / -p flag
2. AG_PASSWORD environment variable
3. Interactive prompt (visible while typing)

The password is never stored on disk. For batch mode with many accounts sharing the same Workspace password:

# Option 1: flag (visible in shell history — use for testing only)
python main.py -b emails.txt -p "password"

# Option 2: env var (recommended for scripts)
AG_PASSWORD="password" python main.py -b emails.txt

# Option 3: prompted (most secure)
python main.py -b emails.txt
# → Password (shared for all accounts): mypassword

python main.py

python main.py -b emails.txt           # skip existing accounts
python main.py -b emails.txt --no-skip # re-auth all

python parallel.py -b emails.txt -w 3

• -w = concurrent workers (browsers). 3 recommended, 5 max on most machines.
• Each worker uses a unique callback port (8121, 8122, 8123, ...).
• Approximately 3x faster than sequential.

user1@example.com
user2@example.com
# commented-out@example.com  (skipped)

See emails.example.txt for a template.

The script launches a headed Chromium browser (Google blocks headless) and handles:

1. Email input → clicks #identifierNext
2. Password input → clicks #passwordNext
3. Intermediary screens — Workspace welcome page, Terms of Service, security prompts. Scrolls down and clicks through common button labels (English + Indonesian).
4. OAuth consent — checks "Select all" scopes, clicks "Continue" / "Allow"
5. Callback — captures the authorization code from localhost:8121/callback

Google blocks real headless Chromium. On Linux servers without a display, use --headless to auto-start a virtual framebuffer (Xvfb):

# Install Xvfb
sudo apt install xvfb
pip install xvfbwrapper

# Run headless
python main.py -b emails.txt --headless
python parallel.py -b emails.txt -w 3 --headless

The browser still runs "headed" internally, but renders to a virtual screen — invisible to the user, but appears real to Google.

constants.py       OAuth credentials & API endpoints
google_auth.py     Token exchange, user info, project discovery
login.py           Playwright Google login automation
router_db.py       9router SQLite read/write operations
main.py            Interactive CLI + sequential batch
parallel.py        Parallel batch injector

Q: Why does it open a browser window? Google blocks automated login from headless browsers. The script needs a visible Chromium instance. On Linux servers, use --headless to run inside a virtual display (Xvfb).

Q: Are the OAuth credentials safe to have in source code? Yes. These are the same public OAuth credentials used by Antigravity CLI, opencode-antigravity-auth, and Google Cloud Code extensions. They are not secret.

Q: What if an account shows "unavailable" status? Use menu option 4 ("Test all connections") to trigger a token refresh via the 9router API.

Q: Can I use this with 2FA-enabled accounts? Not automatically. The script handles password-only login. For 2FA accounts, you'll need to manually complete the challenge in the browser window.

Q: Port 8121 is in use — what do I do? The parallel mode auto-assigns unique ports (8121 + worker index). For single mode, you can modify DEFAULT_CALLBACK_PORT in constants.py.

MIT — see LICENSE.