I’d like to clarify whether the official agy -p CLI can be integrated into a local, single-user multi-agent framework using a consumer subscription.
The architecture works as follows:
A Host Agent submits a bounded task to an independent local Python task coordination service, usually without specifying a Worker.
The service selects an eligible execution configuration (e.g., harness, model, and reasoning effort). When multiple candidates exist, a separate Router Agent selects among them but does not launch or control any Worker.
The Python service handles scheduling and launches a separate execution process, which runs the selected harness through a native-protocol adapter.
The Host reviews and accepts the Worker’s result.
The adapters handle execution controls, permissions, sessions, events, and results using each harness’s supported interfaces and the user’s existing configuration.
For Antigravity, the proposed integration would use the official agy -p CLI without modifying its binary, extracting credentials, or proxying backend requests.
Would this architecture be permitted under consumer subscription terms, or would it be classified as prohibited third-party agent orchestration?
In particular, does it matter that the local Python service—not the Host or Router Agent—starts and manages the execution process?
I’d appreciate clarification before implementing this integration.
As noted in the FAQ, Antigravity CLI headless mode (agy -p) is intended for direct local scripting and CI workflows. Wrapping or embedding agy -p as a worker backend inside third-party coding agents, multi-agent orchestration frameworks, or standalone applications is not supported under consumer plans and can trigger automated Terms of Service enforcement.
Having an intermediate local Python coordination service spawn and manage the agy -p process (rather than the Host or Router Agent directly) does not change this restriction, as agy is still being used as an execution backend for an external multi-agent framework under a consumer sign-in.
If you would like to build or run a local multi-agent framework in Python, we recommend using a Google AI Studio API key (GEMINI_API_KEY) or Gemini Enterprise (Vertex AI) instead of a consumer subscription login. You can also use the official Google Antigravity Python SDK (pip install google-antigravity) directly with your API key or Vertex AI credentials.
Thank you for the clarification. I’d like to ask two follow-up questions about the boundary between permitted direct scripting and third-party orchestration.
Suppose the framework no longer launches Antigravity Workers automatically, but instead requires explicit user action for every execution:
1. Manual execution via the local UI
The framework generates a task-specific execution script. The user reviews it and clicks a “Run” button in the local console, which then executes the official agy -p CLI. No automatic AGY invocation occurs.
2. Manual execution via the user’s terminal
The framework only displays the generated script and its executable command. The user copies the command and runs it manually in their own terminal. The framework itself never launches agy.
In both cases, the execution results would still be returned to the local task service for subsequent Host review, but no further AGY execution would be triggered automatically.
Would either approach be permitted under consumer subscription terms?
If neither is permitted, could you clarify whether the restriction comes from the framework generating the execution script, or from the results being integrated back into a multi-agent workflow, even when the user manually invokes the official CLI?
I’d appreciate your guidance on this specific boundary. Thank you!