What API key options does agy support and when to use each?

[kumaakh]

The agy (Antigravity) CLI supports several ways to supply credentials. What are they, when should I use each, and which one does apra-fleet use?

[kumaakh]

Summary: Agy supports four auth paths. ANTIGRAVITY_API_KEY is the primary one for headless/CI use. For desktop use, browser OAuth is the default and requires no env var.

---

Auth options at a glance

Method	Env var	When to use
Browser OAuth	None -- login is stored in OS keyring (Windows Credential Manager / macOS Keychain / Linux libsecret)	Default for desktop use. Run agy once, sign in via browser; login persists across sessions. No env var needed.
ANTIGRAVITY_API_KEY	ANTIGRAVITY_API_KEY	Primary choice for headless, CI, Docker, or scripted environments. Obtain from Google AI Studio. Set it in the environment before invoking agy or fleet commands.
GEMINI_API_KEY	GEMINI_API_KEY	Supported via config mapping for routing background models, subagents, or specific plugins to Gemini. Not a primary auth mechanism for the agy CLI itself. Do not use this as a substitute for ANTIGRAVITY_API_KEY.
GOOGLE_API_KEY	GOOGLE_API_KEY	Supported via config mapping for GCP / Vertex AI enterprise backends. Requires GOOGLE_GENAI_USE_VERTEXAI=true. Advanced use; not needed for typical agy setups.

---

Which one does apra-fleet use?

apra-fleet's agy provider sets authEnvVar = 'ANTIGRAVITY_API_KEY'. When you register a remote agy member and need non-interactive auth, set ANTIGRAVITY_API_KEY on that machine.

provision_llm_auth does not apply to agy. Agy auth is per-machine and set up out of band -- either via the one-time browser OAuth flow or by planting ANTIGRAVITY_API_KEY in the member's environment. The provision_llm_auth fleet tool only applies to Claude and Gemini CLI.

---

Quick setup guide

Desktop (interactive):

1. Install agy: npm install -g @google/antigravity-cli
2. Run agy and sign in via browser when prompted.
3. Done -- the token is stored in the OS keyring and reused automatically.

Headless / CI / Docker:

1. Obtain an API key from Google AI Studio.
2. Set the env var: export ANTIGRAVITY_API_KEY=<your-key> (Linux/macOS) or $env:ANTIGRAVITY_API_KEY = '<your-key>' (PowerShell).
3. Run agy -p "your prompt" -- no browser needed.

Remote fleet member (SSH):

1. SSH into the remote machine.
2. Add ANTIGRAVITY_API_KEY=<your-key> to the member user's shell profile (e.g., ~/.bashrc or ~/.profile).
3. while registering the member use llm =agy. Auth will work without any interactive steps.