Google is transitioning the Gemini CLI to a new Antigravity CLI. If you script prompts, run evals, or wire models into CI/CD, here’s how to migrate quickly and safely.
What’s changing (in plain English)
Per Google’s announcement, Antigravity CLI replaces Gemini CLI as the primary command-line tool. Expect differences in command names, flags, auth flows, and defaults—typical in a major CLI transition. Read the official post first to understand scope and timing.
Source: Google Developers Blog
Fast migration checklist (15–30 minutes)
- Confirm the transition details: read Google’s announcement and note any deprecation windows.
- Install Antigravity CLI via the method recommended in the docs. Verify with: antigravity –version
- Authenticate once: run the login/auth command and confirm your project, scopes, and default region (if applicable).
- Discover commands: run antigravity –help and antigravity help <command> to map your day‑to‑day tasks.
- Smoke test: run a trivial prompt/generation locally and check exit code and output shape (JSON/text).
- Update scripts/aliases: replace old CLI invocations, then run a dry‑run in staging.
- Pin versions in CI/CD: lock a known‑good Antigravity CLI version or container image for reproducibility.
- Update environment variables/secrets: rotate tokens if scopes or names changed; document new variables.
- Monitor after cutover: watch logs, latency, rate limits, and cost metrics for the first week.
Map common tasks (safe discovery pattern)
- List available commands: antigravity –help
- Find subcommand options: antigravity help <subcommand> (repeat for generate, chat, eval, models, etc.)
- Print machine‑readable output: look for a global –json or –format flag in help text.
- Set defaults (model, temperature, system prompts): check config or profile commands in help.
- Streaming vs. non‑streaming: confirm flags and exit codes before swapping in production.
Breaking changes to watch for
- Command/flag names: even small renames can break scripts; search/replace cautiously.
- Output schema: JSON keys, token usage fields, and error payloads may differ. Validate with a contract test.
- Auth scopes and profiles: ensure the new CLI is pointing at the right account/project and has needed scopes.
- Defaults: model IDs, temperatures, safety settings, and timeouts can change across tools.
- Config locations: CLI config/cache directories may move; update Docker volumes and backup paths.
- Exit codes: CI steps relying on specific codes (e.g., 0/1) should be re‑verified.
CI/CD hardening tips
- Version pinning: lock to an explicit Antigravity CLI version and document upgrade cadence.
- Hermetic builds: bake the CLI into a container image to avoid drift across runners.
- Golden tests: keep a small suite that asserts response shape, latency thresholds, and non‑regression prompts.
- Observability: tag runs with build IDs; export logs/metrics to your APM for post‑deploy checks.
Where to learn more
- Announcement: Google Developers Blog
- Docs and APIs: ai.google.dev/docs
Takeaway
Don’t wait for breakage. Install Antigravity CLI, map your core workflows with –help, and pin a working version in CI/CD. A one‑hour rehearsal today avoids outages later.
Enjoy this nugget? Get weekly, no‑fluff updates on practical AI. Subscribe to The AI Nuggets newsletter.
