Antigravity (AGY): Pair-Programming & Subagent Orchestrator

---
title: "Antigravity (AGY): Pair-Programming & Subagent Orchestrator"
description: "Google Antigravity CLI and IDE assistant: multi-subagent orchestration, background task management, tool groups, and brain transcripts."
okf:
  source_origin: "clawdie/www:harnesses/antigravity.md"
  source_of_truth: "sites/agents/src/content/docs/harnesses/antigravity.md"
  source_revision: "sha256:8955520b5ee006ba9357d26cddae7f3556528b54ae1c03336d153419764793f0"
  last_verified: "2026-09-25"
  classified_by: "typesafe-ai/jev"
  verified_by: "pending-human-review"
  disposition: "staged-for-review"
  status: "review-required"
---

# Antigravity (AGY): Pair-Programming & Subagent Orchestrator

**Antigravity** (AGY) is an advanced agentic assistant designed by the Google DeepMind Advanced Agentic Coding team. Built for deep pair-programming, interactive engineering, and multi-subagent orchestration, it bridges interactive developer workflows with autonomous execution.

---

## Architectural Pillars

### 1. Multi-Subagent Background Execution
Antigravity can define, invoke, and supervise concurrent subagents in isolated workspaces:
- `invoke_subagent`: Dispatches specialized subagents (`research`, `self`, or custom-defined roles) to run in the background.
- **Reactive Wakeups:** The primary agent does not poll in wasteful loops; the system automatically delivers waking notifications when a subagent finishes or sends a message.

### 2. Reactive Background Task Engine
Commands launched via `run_command` execute with synchronous timeouts or seamlessly detach as supervised background tasks. Agents monitor task lifecycle states, inspect log URIs, and receive completion notifications automatically.

### 3. Sacred Trajectory & Brain Transcripts
Every decision, thought, and tool execution is recorded in structured JSON Lines transcripts (`transcript.jsonl`) under the agent's application directory:
- Truncation-safe logging with full-transcript parity.
- Markdown artifacts generated into dedicated conversation directories for user review.

### 4. Interactive Operator Alignment
- `ask_question`: Renders interactive modal questions with selectable options and default write-in inputs, halting execution until the human operator responds.
- Slash commands (`/goal`, `/plan`, `/grill-me`, `/boost`): Shortcuts triggering specialized behavioral modes and verification gates.

---

## Role in the Clawdie Ecosystem

While `zot` handles lightweight system turns and `hermes` manages the multi-platform gateway, **Antigravity serves as the primary workstation pair-programming orchestrator**:
- Conducting multi-agent audits and code migrations across worktrees.
- Running live System-1 decision benchmarks (like Jev triage).
- Synthesizing, verifying, and deploying documentation and web surfaces.

This content is not available in your language yet.

Antigravity (AGY): Pair-Programming & Subagent Orchestrator

Antigravity (AGY) is an advanced agentic assistant designed by the Google DeepMind Advanced Agentic Coding team. Built for deep pair-programming, interactive engineering, and multi-subagent orchestration, it bridges interactive developer workflows with autonomous execution.

Architectural Pillars

1. Multi-Subagent Background Execution

Antigravity can define, invoke, and supervise concurrent subagents in isolated workspaces:

invoke_subagent: Dispatches specialized subagents (research, self, or custom-defined roles) to run in the background.
Reactive Wakeups: The primary agent does not poll in wasteful loops; the system automatically delivers waking notifications when a subagent finishes or sends a message.

2. Reactive Background Task Engine

Commands launched via run_command execute with synchronous timeouts or seamlessly detach as supervised background tasks. Agents monitor task lifecycle states, inspect log URIs, and receive completion notifications automatically.

3. Sacred Trajectory & Brain Transcripts

Every decision, thought, and tool execution is recorded in structured JSON Lines transcripts (transcript.jsonl) under the agent’s application directory:

Truncation-safe logging with full-transcript parity.
Markdown artifacts generated into dedicated conversation directories for user review.

4. Interactive Operator Alignment

ask_question: Renders interactive modal questions with selectable options and default write-in inputs, halting execution until the human operator responds.
Slash commands (/goal, /plan, /grill-me, /boost): Shortcuts triggering specialized behavioral modes and verification gates.

Role in the Clawdie Ecosystem

While zot handles lightweight system turns and hermes manages the multi-platform gateway, Antigravity serves as the primary workstation pair-programming orchestrator:

Conducting multi-agent audits and code migrations across worktrees.
Running live System-1 decision benchmarks (like Jev triage).
Synthesizing, verifying, and deploying documentation and web surfaces.