At a glance
A proprietary terminal interface to Google's shared Antigravity agent harness, with a managed multi-provider model catalog, asynchronous subagents, structured headless output, SSH use, browser actions, MCP, plugins, skills, hooks, granular permissions, and optional native OS sandboxing.
Good choice if
- Individual Gemini CLI users moving to Google's supported consumer terminal product
- Terminal, SSH, or headless work that benefits from parallel background agents
- Teams using a managed model catalog through Google AI plans or the Gemini Enterprise Agent Platform
Check before choosing
- Model access is managed by Google plans or Google Cloud; the documentation does not offer BYOK endpoints or local-model execution
- Native OS sandboxing is available but off by default, while workspace file reads and writes are auto-allowed by the permission engine unless stricter policy is configured
- Since 1.1.9, a permission pattern accepted at a prompt persists for the rest of the conversation rather than one turn, so an interactive approval can become a session-scoped rule
See 5 more considerations
- Always-proceed modes and one-command sandbox escapes can remove review or containment boundaries
- Conversation rewind and fork operate on session history, not the local Git checkout, so they do not provide file rollback or worktree isolation
- Telemetry is enabled by default and must be changed in settings when that collection is not desired
- The public GitHub repository distributes releases and support material but does not expose the core harness source, engineering tests, security policy, or evaluation suite
- Antigravity CLI 1.1.11 fixes zero-word command allowlists matching every command, auto-approval in strict or request-review modes, and startup MCP admin-control caching; older binaries should not be treated as having those corrections