Google Antigravity release notes: CLI codelab makes the terminal agent concrete

Adam Olofsson Hammare
Google Antigravity release notes: CLI codelab makes the terminal agent concrete

Google has not posted a newer Antigravity app version after 2.1.4 in the public changelog. Still, there is a useful release-note signal today: Google's official Hands-on with Antigravity CLI codelab was updated on June 15, 2026, and it makes the terminal workflow much more concrete.

Google Antigravity is Google's agentic development platform. An agentic IDE or CLI lets an AI agent plan, read files, propose changes and sometimes run tools inside a bounded workspace. Antigravity CLI is the terminal surface for that pattern: a TUI, or text-based terminal interface, for agentic coding and file work.

Source: Hands-on with Antigravity CLI says Last updated: Jun 15, 2026 and describes Antigravity CLI as a lightweight Terminal User Interface surface with multi-step reasoning, multi-file editing, tool calling and conversation history. Google Antigravity Changelog still shows 2.1.4 from June 11, 2026 as the latest app version.

Google Antigravity release notes: the CLI codelab is today's signal

This is not a new 2.1.5 row. It is a documentation update with practical changelog value. The codelab shows how a user installs Antigravity CLI, signs in, trusts a workspace, finds help commands, opens settings, selects a model and uses the CLI for both coding workflows and everyday file tasks.

For Nordic teams, that matters because a terminal agent usually sits closer to real files than a normal chat tab. You need to know the boundary: which folder the agent can see, when commands need approval and how to review a plan before files change.

Source: Hands-on with Antigravity CLI says the codelab teaches installation, basic usage and use cases. It also shows the workspace prompt where Antigravity CLI requires permission to read, edit and execute files in the current folder.

What Google now makes clearer in Antigravity CLI

A few details in the codelab are worth putting into your team's own operating note before people start using the terminal agent more freely:

  • Antigravity CLI installs as agy, and the codelab shows separate install commands for macOS/Linux, Windows PowerShell and Windows CMD.
  • On first launch, the user can choose Google OAuth or a Google Cloud project. The codelab uses a personal account in its example.
  • The CLI asks whether you trust the current workspace folder before it can read, edit and execute files there.
  • /help shows commands and keyboard shortcuts. /quit or ctrl+d exits the interactive session.
  • /config or /settings opens settings such as Tool Permission, Sandbox Mode, telemetry and color scheme.
  • agy models lists available models, and agy --model "Gemini 3.5 Flash (Low)" starts a session with a selected model according to the codelab.
  • The codelab also shows --dangerously-skip-permissions, but describes it as a mode that auto-approves tool permissions. That should not be the default for workflows with customer data, API keys or production files.

Source: Hands-on with Antigravity CLI shows the install command curl -fsSL https://antigravity.google/cli/install.sh | bash, the version check agy --version, first launch with agy, /help, /quit, /config//settings, agy models, agy --model "Gemini 3.5 Flash (Low)" and the warning around --dangerously-skip-permissions.

Human step before you ask the agent to work

Do this yourself, not as an agent prompt:

mkdir agy-cli-projects
cd agy-cli-projects
curl -fsSL https://antigravity.google/cli/install.sh | bash
agy --version
agy

Choose the login method, trust only a folder you actually want to use as the agent's workspace and open /config or /settings before the first real task. Keep permission review as the default until you have written your own rules for what the agent may do.

Source: Hands-on with Antigravity CLI verifies the folder commands, macOS/Linux install command, agy --version, first launch with agy, the Google OAuth flow, the workspace trust prompt and /config//settings.

Short example: use the new Antigravity feature

Once the CLI is installed and started inside a trusted project, paste a small task that exposes the boundaries before code changes:

Inspect this workspace as an Antigravity CLI onboarding check. Identify the main files and likely project purpose, list which commands or file edits you would need approval for, then propose one small README improvement. Do not edit files until I approve the plan.

Good output should:

  • separate what the agent can see from what it assumes
  • name files or commands that would need approval
  • propose one small README improvement, not a full refactor
  • end with a clear question before files change

What Hammer readers should test today

Treat Antigravity CLI as a controlled terminal coworker, not a free-running background process. Start in a separate workspace folder. Keep secrets in env vars or a secret manager, use scoped API keys when tools need access, and make approval gates plus run logs part of the routine.

This is Tool Forge territory: before the terminal agent becomes useful in real work, you need to decide workspace, permissions, model choice, logging and human review. The codelab gives you a decent first draft for that operating card.

Source: An important update: Transitioning Gemini CLI to Antigravity CLI says Antigravity CLI became available on May 19, 2026 and that affected consumer Gemini CLI/Gemini Code Assist flows stop serving requests on June 18, 2026. Google Antigravity Download lists Antigravity CLI as a downloadable terminal surface.

FAQ

Is this a new Google Antigravity version?

No. The public changelog still shows 2.1.4 as the latest app version. The update is Google's official Antigravity CLI codelab, updated June 15, 2026, with practical CLI instructions.

What should teams test first in Antigravity CLI?

Start in a separate workspace folder, review /config or /settings, keep permission review on and ask the agent for a read-only onboarding check before files change.

Where should human control sit?

At workspace trust, tool permissions, model choice, API keys and before file edits or terminal commands run. Use env vars, secret managers, scoped keys, approval gates and run logs.

The Forge newsletter

Get new articles in your inbox

Pick the topics you care about. No noise, at most one email a week.

Get new articles in your inbox

We follow GDPR. Unsubscribe anytime.