Terms Clarification: Personal Antigravity CLI Integration with a Multi-AI Orchestrator

Google Antigravity
1

Dear Google Antigravity Team,

I am developing a private, non-commercial AI orchestration system for my own software development projects.

Before integrating Antigravity, I would like to confirm whether my intended architecture complies with the Google Antigravity Additional Terms of Service.

Project overview

The system will be used exclusively by me and hosted on a private Linux server, potentially using Oracle Cloud or Vultr.

It will use three official coding clients:

  • Google Antigravity CLI (agy)

  • OpenAI Codex

  • Anthropic Claude Code

The system will provide a single private chat interface and dynamically coordinate the coding clients.

Its primary objective is to improve development quality, reduce repeated analysis, and maintain workflow continuity across independently subscribed AI services.

How it would work

For example, Codex may implement a feature while Antigravity independently verifies a previous code revision.

The orchestration program may then pass Antigravity’s verification findings to Claude Code for corrections.

Tasks will be allocated dynamically rather than through fixed roles.

Three execution modes are planned:

  • Sequential: one active coding client at a time, with automatic task handoffs.

  • Parallel: multiple official clients executing independent tasks concurrently.

  • Hybrid: dynamically combining sequential and parallel execution based on efficiency.

All clients will access one shared project environment with appropriate file-conflict controls.

Authentication and restrictions

I intend to execute only Google’s official, unmodified agy CLI through supported command-line mechanisms.

I will authenticate using my own Google account and eligible subscription through Google’s official login process.

I will not extract, intercept, share, or proxy OAuth credentials, call undocumented APIs, bypass usage restrictions, or provide access to other users.

Questions

  1. Is invoking the official agy CLI from my privately developed orchestration application permitted under the Antigravity Additional Terms?

  2. Does providing a private web-based interface or running the official CLI on a personal cloud server change that permission?

  3. Is automatic task handoff between Antigravity and other official AI coding clients allowed?

  4. May the orchestration program dynamically select Antigravity based on task suitability, availability, and normal subscription usage limits?

  5. Are sequential and parallel executions of the official CLI permitted within the applicable subscription restrictions?

  6. Does this architecture fall under the third-party software restriction in Section 6, despite using only the official unmodified CLI and official authentication?

  7. If any operation is restricted, could you identify the relevant policy and suggest a compliant approach?

I have reviewed the following discussion, where running the official agy executable from a script was described as permitted:

I would appreciate clarification on whether that interpretation also applies to my proposed remote, single-user orchestration system.

If necessary, please refer this inquiry to the appropriate Antigravity policy or licensing team.

Thank you for your assistance.

Best regards

2

Hi @BuddhaRyu,

The key point: Section 6 forbids extracting, intercepting or proxying your Antigravity sign-in credentials to power other tools. One example is plugging them into OpenClaw, OpenCode or Claude Code. Running the official, unmodified agy CLI in headless mode (agy -p) from your own scripts is exactly what that mode is for.

Your questions:

  1. Running agy from your orchestrator: Yes. Running the official agy program from your app (e.g. agy -p "<prompt>" --output-format json or stream-json) for your own development work is fine, as long as agy handles its own sign-in and agent work.
  2. Private web interface or personal cloud server: That’s fine too, with two conditions:
    • Only you use it. Giving anyone else access, even friends or teammates, counts as account sharing.
    • Sign in on that machine with the official login (for example over SSH port forwarding). Don’t copy sign-in token files between machines.
  3. Handoffs between agy, Codex and Claude Code: Yes. Passing results like diffs or review notes between tools in your own project is fine. Use agy for real coding or review tasks, not as a plain model connection for another tool’s agent.
  4. Picking agy automatically: Yes, as long as you stay within your normal usage limits and don’t switch between accounts or work around rate limits.
  5. One at a time vs. in parallel: Running tasks one at a time, or a few in parallel, is fine. But every agy -p run uses the same 5-hour and weekly quota as the IDE and CLI. Many parallel runs from cloud servers like Oracle Cloud or Vultr can use up quota quickly or trigger abuse protection. On a cloud server, we recommend one agy task at a time.
  6. Does section 6 apply? No. Only the official agy program talks to Antigravity, under your own sign-in, and your orchestrator just sends prompts and reads the output.
  7. What to avoid: Extracting or proxying credentials, letting anyone else use your interface, and large bursts of parallel runs from cloud servers.

For heavier workloads: If you need many parallel or always-on agents on a server, use the Antigravity Python SDK (pip install google-antigravity), or the Gemini API with a Google AI Studio or Vertex AI API key, instead of your personal subscription.