← Tools

Google Antigravity

Google

★★★★☆

Google's desktop command center and CLI for parallel coding agents, browser work, and automation.

Category coding-assistant
Pricing Individual plan free; Google AI Pro $20/month; Google AI Ultra $100/month or $200/month with higher limits; Google Cloud organization access uses consumption-based pricing
Status active
Platforms macos, windows, linux
google antigravity coding-agent agent-command-center cli subagents worktrees browser mcp skills scheduled-tasks
Updated July 10, 2026 Official site →

Overview

Freshness note: AI products, plan limits, and agent permissions change quickly. This profile is a point-in-time snapshot last verified on July 10, 2026.

Google Antigravity is Google’s agent-first development platform. The current centerpiece is Antigravity 2.0, a standalone desktop command center for launching, monitoring, and coordinating agents across coding and knowledge-work tasks. It runs on macOS, Windows, and Linux and is separate from the older Antigravity IDE.

That distinction matters. Antigravity 2.0 has no built-in editor; Google recommends using it beside the IDE of your choice. The earlier Antigravity IDE remains available during the transition, but Google says its Agent Manager will eventually be removed. A separate Antigravity CLI exposes the same agent harness for terminal-first work.

Key Features

Antigravity agents can work with local project files, run commands, search the web, operate a dedicated Chrome profile, use skills and MCP servers, and return plans, screenshots, recordings, and other reviewable artifacts. Projects can include multiple folders, carry project-specific permissions, and use Git worktrees to isolate background work.

The multi-agent layer is more substantial than a chat sidebar. A parent agent can create asynchronous subagents for research, testing, or implementation, monitor their state, and receive permission requests through the main interface. Scheduled tasks can start recurring conversations, while the Ultra-only teamwork preview adds more automated coordination and recovery around agent groups.

The Antigravity CLI shares the core harness and settings with the desktop app. Google is steering Gemini CLI users toward this newer CLI, while the Antigravity SDK and managed-agent integrations expose the same underlying runtime to custom or hosted workflows. Gemini 3.5 Flash is the default optimized route, but current plans also list other Gemini models, Claude models, and gpt-oss-120b with plan-dependent quotas.

Strengths

Antigravity is strongest when the hard part is coordinating several kinds of work around a repo: research, implementation, browser verification, tests, and review. Native worktree support and artifact-based feedback are sensible choices because they make parallel work easier to inspect than one giant agent transcript.

The shared desktop, CLI, AI Studio, SDK, and enterprise harness also gives Google a coherent developer-agent story. A prototype can start in Google AI Studio, move into local Antigravity work, and later use Google Cloud agent infrastructure without pretending those surfaces are identical.

Limitations

The product family is still in transition. Individual access is generally available, and organization access to Antigravity 2.0 and CLI is now available through Gemini Enterprise Agent Platform under Google Cloud terms and consumption-based billing. The old IDE and new command center still overlap temporarily, while the SDK, automated teamwork, centralized enterprise controls, and planned IDE extensions carry separate preview or coming-soon boundaries.

Security controls require deliberate setup. Desktop terminal sandboxing is not a universal default, browser and MCP access widen the attack surface, and permissive project presets can expose more of the machine than the task needs. Google’s individual terms also say interactions may be used to evaluate and improve Google and Alphabet products unless the user changes that preference; enterprise access follows a different Google Cloud terms and data-governance path.

Practical Tips

Start with one trusted repo, the default or strict permission profile, terminal sandboxing enabled where supported, and browser access limited to named domains. Use worktrees for parallel agents, keep scheduled tasks draft-only, and avoid full-machine or unrestricted presets unless the workflow genuinely requires them.

Review the actual diff, commands, tests, screenshots, and browser recordings—not just the final artifact summary. Multi-agent coordination can reduce waiting time, but it also creates several simultaneous ways to be confidently wrong.

Verdict

Google Antigravity is a credible agent command center for developers who want local repo work, browser verification, subagents, schedules, and a terminal surface under one Google-backed harness. It is most useful for experienced users who will configure permissions and review every change; it is not a shortcut around engineering ownership.