Human-in-the-Loop Task Manager for AI Agents
Docs / Agents / Integrations
Integrations

Connect Antigravity to AgentRQ

Connect Google Antigravity to AgentRQ over the Agent Client Protocol (ACP) for human-in-the-loop task management.

Google Antigravity is Google's AI coding agent. It supports the Agent Client Protocol (ACP), so it can run behind AgentRQ's gateway with no custom integration code.

Does Antigravity Support ACP?

Yes. Antigravity publishes an ACP server to the ACP registry as antigravity-acp, so AgentRQ's gateway can fetch and run it by name — nothing to install by hand.

What Connecting via ACP Gets You

Once Antigravity is bridged through AgentRQ's ACP Gateway, every task it works on shows up on your AgentRQ task board in real time. You can approve risky commands, reply with extra context, or reassign work — from the dashboard or Slack — without touching the terminal Antigravity is running in.

Connect Antigravity

Antigravity needs two things the other ACP agents do not, and both flags belong on every command:

  • → --allow-unverified-agent — the registry publishes no sha256 for Antigravity on any platform, and the gateway refuses to install a download it cannot verify unless told to.
  • → A login first — Antigravity will not open a session until it has been authenticated.

Copy the .mcp.json for your workspace from Settings → Setup → Antigravity in the AgentRQ app, save it in your project root, then run both commands from that directory.

Log in (the first run also downloads the agent):

bash 
npx @agentrq/acp-gateway@latest --login --agent antigravity-acp --allow-unverified-agent

Start the bridge:

bash 
npx @agentrq/acp-gateway@latest --agent antigravity-acp --allow-unverified-agent

Antigravity offers four login methods: oauth-personal (a Google account), oauth-business (Gemini Enterprise), gemini-api-key, and agent-platform (Gemini Enterprise Agent Platform, formerly Vertex AI). Pick one when prompted, or name it directly with --login oauth-personal. Sign out again with --logout.

Choosing a Model

Antigravity exposes several Gemini variants, each with a high / medium / low reasoning-effort suffix. List them:

bash 
npx @agentrq/acp-gateway@latest --list-models --agent antigravity-acp --allow-unverified-agent

Then select one — --model accepts either the model id or its display name:

bash 
npx @agentrq/acp-gateway@latest --agent antigravity-acp --allow-unverified-agent --model gemini-3.8-flash-high

The gateway applies the model as each session opens and reports it back to AgentRQ, so the workspace shows which model actually ran.

Supported Platforms

The registry publishes Antigravity for Apple Silicon macOS, Linux (x86_64 and arm64) and Windows (x86_64 and arm64). There is no Intel macOS build, so on an Intel Mac the gateway has nothing to install.

The binary is downloaded once and cached per agent, version and platform — under ~/.cache/acp-gateway/agents on macOS and Linux, %LOCALAPPDATA%\acp-gateway\agents on Windows.

FAQ

Does Antigravity work with AgentRQ?

Yes — Antigravity speaks the Agent Client Protocol (ACP), so it connects to AgentRQ through the open-source ACP Gateway with no custom integration code.

How do I connect Antigravity to AgentRQ?

Save your workspace's .mcp.json in your project root, log in with npx @agentrq/acp-gateway@latest --login --agent antigravity-acp --allow-unverified-agent, then start the bridge with the same command minus --login.

Why does Antigravity need --allow-unverified-agent?

The ACP registry publishes no checksum for the Antigravity binary, and the gateway will not install a download it cannot verify. The flag is you accepting that trade-off; it is required on every command, and only the first run downloads anything.

Which models can Antigravity use?

Several Gemini variants, each in high, medium and low reasoning-effort forms. Run --list-models to see the current set, and pick one with --model.

References