Twelve Pages Google Antigravity & Jules

Google Antigravity & Jules

Agent-first desktop IDE, parallel subagents, skills/MCP customization, and asynchronous GitHub PR automation with Jules.

Updated 2026-11 v1.0 12 min read 12-Page Illustrated Edition AI-Authored & Human-Directed
Google Antigravity and Autonomous Coding Agents Illustration

Core Takeaways

Four Unified Surfaces

Work seamlessly across the Desktop App, VS Code-based IDE, Terminal CLI (agy), and Python SDK.

Jules DNA Built Right In

Includes the complete late-2026 Jules-to-Antigravity migration map for async branches, PRs, and cron runs.

Powered by AI Ultra

Unlock 5x–20x frontier Gemini quotas so multi-agent swarms and overnight /goal runs never stall.

1. The Big Picture: Antigravity + Jules in One Studio

Google Antigravity (antigravity.google) is Google's flagship agent-first software creation platform—designed not just to autocomplete lines of code, but to plan, build, test, browse, and ship entire projects alongside you.

Quick Navigation

Visualization: Four Ways to Drive One Shared Agent Brain

SURFACE 1 • DESKTOP Antigravity 2.0 App Mission control for chat, artifacts, swarms & cron automations. SURFACE 2 • EDITOR Antigravity IDE VS Code-based editor with Supercomplete Tab, Cmd+I & visual diffs. SURFACE 3 • TERMINAL Antigravity CLI (agy) Lightning-fast TUI inside any local or SSH shell session. SURFACE 4 • CODE Python SDK pip install library to spawn & stream agents in scripts/CI. Shared Antigravity Engine (Gemini Frontier Models + Jules Async Primitives) Reads the exact same AGENTS.md / GEMINI.md rules, .agents/skills/, Plugins, MCP servers & Sandbox policies
Important Late-2026 Update: Jules Is Now Folding Into Antigravity!
Google's standalone asynchronous GitHub agent, Jules (jules.google), pioneered background cloud-VM bug fixing and automated Pull Requests. In late 2026 (read-only Nov 11, 2026; full sunset Dec 31, 2026), Google consolidated Jules's best async capabilities—branched background subagents, scheduled cron tasks, AGENTS.md memory, and GitHub PR automation—directly into Google Antigravity. Page 8 covers both Jules's workflow and your exact migration map!

2. First-Time Setup & Your Google AI Ultra Advantage

You don't need to be a senior software engineer to use Antigravity. Non-coders use Antigravity 2.0 to generate guides, websites, and automations, while engineers pair it with the IDE and CLI.

3-Step Setup (Under 3 Minutes)

1 Install Your Surface

Download Antigravity 2.0 or Antigravity IDE from antigravity.google, or install the terminal CLI (agy) and Python SDK.

2 Sign In w/ AI Ultra

Authenticate with the Google account tied to your Google AI Ultra plan to unlock frontier models and 5x–20x rate limits automatically.

3 Open or Import Config

Select a workspace folder. Coming from Cursor, Claude Code, or Codex? Ask the agent: "Migrate my agent settings into a plugin."

Which Antigravity Surface Should You Open Today?

Surface Best For Everyday Superpower
Antigravity 2.0
Desktop App
Creators, PMs, researchers & multi-agent orchestrators Clean workspace without code clutter. Delegate research, build apps from scratch, inspect rich Markdown/HTML Artifacts, and schedule daily cron jobs.
Antigravity IDE
Standalone Editor
Hands-on coding, refactoring & debugging Full VS Code compatibility plus single-keystroke Tab Supercomplete, Cmd+I inline edits, inline Code Lenses, and red/green visual diff review.
Antigravity CLI
agy in Terminal
Terminal lovers, SSH servers & quick repo tasks Type agy inside any folder to launch an interactive terminal agent with full slash commands and subagent support (~/.gemini/antigravity-cli/settings.json).
Python SDK
google-antigravity
Custom automation pipelines & CI test suites Spawn programmable Agent() instances inside Python scripts, stream live thought/tool events, and wire custom local tools.

What Your Google AI Ultra Plan Unlocks Here

1. Unthrottled Multi-Agent & Swarm Runs
Features like /teamwork (parallel agent swarms), /boost (multi-perspective deep verification), and /goal (long overnight runs) consume heavy tokens. AI Ultra's 5x–20x quota multiplier keeps them humming without hitting mid-task caps.
2. Frontier Models + Cloud Developer Credits
Ultra gives you priority access to flagship reasoning models (Gemini Pro / Carbon / Deep Think) inside the Model Selector, plus Google Developer Program Premium monthly Google Cloud & Gemini API credits for deploying what you build.

3. Antigravity 2.0 Desktop App: Annotated Visual Tour

Antigravity 2.0 separates high-level thinking from raw code files. You converse in the center Chat Canvas while live subagents, terminals, artifacts, and file diffs update in the right-hand Auxiliary Pane.

Antigravity 2.0 • Workspace: acme-web-app
2 Model: Gemini 4 Carbon Sandbox: ON
NAVIGATION 1
+ New Chat
Projects (4)
Scheduled Tasks
Plugins & Skills
Settings & Policy
You: /plan Add Stripe checkout & verify mobile layout in @src/checkout
ARTIFACT • implementation_plan.md 3
1. Create webhook route in api/stripe/route.ts
2. Launch Browser Subagent to verify checkout UI
Comment Proceed →
Type / for commands, @ to attach files/MCP, or drop screenshots... 4
Subagents (2) Tasks Diffs (3)
Browser Agent5
Testing localhost:3000/checkout
Bg Task: npm test6
18/18 passed (1.4s)
Files Changed (3)7
+84 -12 across 3 TypeScript files

Numbered Interface Reference

1 Left Sidebar: Switch workspaces, start fresh chats, manage cron Scheduled Tasks, toggle Plugins & Skills, or tune security rules.
2 Model & Sandbox Bar: Pick Gemini Flash (speed), Pro/Carbon (deep coding), and inspect active terminal sandbox status.
3 Interactive Artifacts: Plans, reports, Mermaid diagrams, and diffs open as rich documents—click Proceed to approve execution.
4 Smart Composer: Trigger workflows with /, pull exact context with @, or drag UI mockups and PDFs straight into chat.
5 Subagents & Swarm Monitor: Watch parallel research, coding, or live Webview Browser agents work without cluttering main chat.
6 Background Tasks & Terminals: Dev servers and long test suites run asynchronously and automatically wake the agent when done.
7 Files Changed & Sidecars: Inspect every added/deleted line across your repo, or open plugin UI panels (like Android Device Hub or Compose Preview).

4. Antigravity IDE: Three Speeds of AI Pair Programming

Built on VS Code so all your favorite extensions and themes carry over, Antigravity IDE gives you three distinct gears depending on whether you are editing one line, one function, or twenty files.

Gear 1: Passive (Tab)

Millisecond Flow State. Predicts your next edit, multi-line diff (Supercomplete), missing import, or next cursor jump.

Press Tab to accept
Press Cmd+→ word-by-word
Press Tab again to jump!
Gear 2: Inline (Cmd+I)

Surgical Single-Block Edit. Highlight a function, press Cmd+I (Ctrl+I), and ask for a refactor, docstring, or type fix.

Edits only highlighted lines.
Shows red/green inline diff.
Zero risk to rest of file.
Gear 3: Sidebar Agent

Multi-File Autonomous Builder. Reads codebase, edits across folders, runs tests, checks browser, and fixes errors.

Toggle Planning Mode first
for complex architecture
before touching code.

Essential Antigravity IDE Features & Shortcuts

Feature / Action Shortcut (macOS / Win) How to Use It Effectively
Supercomplete & Tab-to-Jump Tab / Esc After renaming a prop or parameter, wait 0.2s—Antigravity highlights the next callsite and lets you teleport there with Tab.
Tab-to-Import Tab Type a component or utility name from another file; pressing Tab automatically inserts the clean top-of-file import statement.
Inline Command Cmd+I / Ctrl+I Place cursor on an empty line to generate a new helper from scratch, or select messy code and type "Extract into pure async helper."
Inline Code Lenses Click lens above symbol One-click buttons floating above classes and functions: Explain, Refactor, or Write Unit Tests scoped to that exact symbol.
Diagnostic Auto-Fix Hover red squiggle / Problems Click Fix with Antigravity on any TypeScript, Python, Rust, or linter error to patch root causes using surrounding context.
Pro Workflow: Run Antigravity IDE + Antigravity 2.0 Side by Side
Both apps share the exact same workspace and .agents/ rules! Many Ultra users keep Antigravity 2.0 running long research or multi-agent swarms on a second monitor while making fast Tab and Cmd+I tweaks inside Antigravity IDE.

5. Steering Agents with / Slash Commands & @ Mentions

Typing a single slash command at the start of your prompt shifts Antigravity into specialized operating modes—from interactive design interviews to unstoppable overnight execution.

Complete Built-In Slash Command Cheat Sheet

Command What Mode It Activates When to Reach for It
/plan Step-by-Step Blueprint Mode Before any multi-file feature or refactor. Writes a reviewable plan artifact before editing code.
/grill-me Interactive Design Interview When your idea is still fuzzy—the agent interviews you on edge cases and trade-offs until aligned.
/goal Relentless Autonomous Mode Long or overnight tasks (e.g., "Get test coverage to 90% and fix all build errors")—won't stop early.
/boost Deep Multi-Perspective Mode Hard architecture bugs or research requiring deep reasoning, competing hypotheses, and strict proof.
/teamwork Multi-Agent Swarm Mode Large projects where independent slices (frontend, API, tests, review) can be built in parallel.
/browser Live Webview Browser Agent Testing localhost web apps, clicking through UI flows, or interacting with live JS-rendered sites.
/schedule Recurring Cron Automation Setting up daily/weekly background runs (also accessible via /automation), like morning PR digests.
/learn Persistent Memory Capture Right after you correct a mistake or solve a tricky setup—saves the lesson as a reusable rule/skill!
/compact Context Window Optimizer Summarizes a very long chat history so your session stays sharp, fast, and uncluttered.

Precision Context with @ Mentions

@ Files & Folders

Attach @package.json or @src/auth/ so the agent reads exact targets immediately without guessing.

@ Terminals & Chats

Reference a failing build with @terminal or carry decisions forward from @previous-conversation.

@ Rules & MCP Tools

Explicitly invoke a style guide via @rule or query connected databases/GitHub via @mcp.

6. The 6-Step Agentic Build Loop (Idea to Verified PR)

Whether you are building a personal dashboard from scratch or adding a feature to an existing repo, this 6-step loop keeps you in full control while the agent handles the heavy lifting.

1 Align w/ /plan Describe goal + @ files; agent researches repo & drafts plan. 2 Approve Artifact Read plan artifact, tweak edge cases, and click "Proceed". 3 Build & Run Tests Agent edits files, runs terminal test suites, and self-heals bugs. 4 Live Browser Check Browser subagent clicks through localhost UI & screenshots proof. 5 Inspect Diffs & /learn Verify Files Changed pane; save any new repo gotchas via /learn. 6 Ship Clean PR Ask agent to commit & open a GitHub Pull Request via MCP/gh.

Concrete Example: Building a Personal Expense Tracker Web App

Turn Exact Prompt You Type What Antigravity Does Behind the Scenes
Turn 1: Plan /plan Build a clean Next.js + SQLite expense tracker with CSV import and monthly charts. Creates a task checklist & architecture plan artifact with tech choices and schema; pauses for feedback.
Turn 2: Build Click Proceed on the plan artifact (or reply: "Add dark mode toggle too, then proceed.") Scaffolds project, writes database & chart components, runs npm run build and unit tests until green.
Turn 3: Verify UI Start the dev server and open localhost:3000 in the browser to test uploading a sample CSV. Starts background server, delegates to the Browser Subagent, uploads a test CSV, and verifies charts render.
Turn 4: Ship Initialize git, write a clean commit, and push to a new private GitHub repo via GitHub MCP. Commits tested code and creates your remote repository + clean README automatically.
Zero-Polling Reactive Architecture
When Antigravity starts a long test suite, dev server, or subagent, you never need to babysit or ask "is it done yet?" Background tasks automatically wake the agent the instant they complete.

7. Google Jules: Async Pioneer & Antigravity Migration

Google Jules (jules.google) introduced "fire-and-forget" asynchronous engineering—cloning GitHub repos into cloud Linux VMs, proposing a plan, running tests, and opening Pull Requests while you worked elsewhere.

Interactive Workstation vs Asynchronous Cloud Agents
[LEFT: INTERACTIVE MODE] Live pair-programming in IDE/App —— Unified in Antigravity (Late 2026) → [RIGHT: JULES ASYNC MODE] Parallel branched workers & overnight cron PRs

The Late-2026 Timeline & 1-to-1 Jules → Antigravity Map

Rather than splitting your quota between two separate coding products, Google folded Jules's async architecture into Antigravity (Jules entered read-only mode on Nov 11, 2026 ahead of full sunset on Dec 31, 2026). Every beloved Jules workflow now has a direct—and more powerful—home inside Antigravity:

Classic Jules Feature (jules.google) How You Do It Better in Google Antigravity Today
"Give Me a Plan" Approval Gate Use /plan (or IDE Planning Mode). Antigravity writes a structured Markdown plan Artifact with an interactive Proceed button.
60 Parallel Cloud Tasks (AI Ultra)
jules remote new --parallel 3
Use /teamwork (Swarm) or spawn subagents with workspace: "share" / "branch" so each agent edits its own isolated git copy in parallel.
Repo AGENTS.md Memory File 100% Native! Antigravity automatically discovers and obeys your existing root and subdirectory AGENTS.md (and GEMINI.md) files with zero changes.
GitHub Issue / PR Automation Connect the built-in GitHub MCP Server (or gh CLI skill) + /goal to pull an issue, fix it on a branch, run tests, and open a ready-to-merge PR.
Scheduled Scans & Audio Recaps Use /schedule (or /automation) to run nightly dependency/security audits or morning changelog digests automatically.
Used Jules Previously? Export Your History Before Dec 31, 2026
Run npx @google/jules in your terminal and authenticate via jules login to export past session patches, setup scripts, and repository memories straight into your local Antigravity .agents/ folder before standalone jules.google retires.

8. Multiplying Yourself: Subagents & Native Swarms

With a Google AI Ultra plan, you aren't limited to a single single-threaded assistant. Antigravity can dispatch specialized specialist agents—or an entire collaborative team—that work simultaneously.

Lead Coordinator Agent (Your Chat) Plans slices, assigns briefs, merges verified branches Research Subagent Read-only sandbox. Scans 50+ files without bloating your main context window. Worker A (share copy) Builds backend API slice in its own git worktree— zero file edit collisions. Worker B (share copy) Builds React UI slice in parallel & coordinates on the Swarm message board. Browser + Image Agents Generates UI illustrations & clicks through live web pages to verify flows.

Subagents vs. Swarms: When to Use Each

A Focused Subagents (Automatic)

Antigravity automatically delegates focused jobs—like surveying a huge repo (research), clicking around a web app (browser), or generating assets (image-generator)—and receives a concise summary back.

B Native Swarms (/teamwork)

For ambitious builds or cross-checking critical code, type /teamwork. Multiple named agents (author-1, author-2, verifier) collaborate on a shared public message board!

Understanding Workspace Isolation Modes

Workspace Mode How It Works & Why It Prevents File Conflicts
inherit (Default) Works directly inside your active folder. Ideal for read-only researchers, code reviewers, or single-author tasks.
share (Worktree) Creates an instant lightweight working copy sharing .git storage—letting 3+ authors edit and build simultaneously without overwriting each other!
branch (Clone) Creates a completely isolated clone for risky refactors or competing architectural experiments (just like Jules's parallel --parallel 3 runs).

9. Customizing Antigravity: Rules, Skills, Plugins & MCP

Out of the box, Antigravity is a brilliant generalist. With five modular customization building blocks, you turn it into a specialist tailored to your exact voice, codebase, and third-party apps.

Rules Blueprint, Skills/Plugins Gears, and MCP Universal Hub
1. Rules (GEMINI.md / AGENTS.md): Always-on guardrails, tone & repo conventions.
2. Skills & Plugins (SKILL.md): On-demand playbooks & 1-click Marketplace bundles.
3. MCP & Hooks: Live connections to GitHub, Drive, Figma, Databases & scripts.

The 5 Customization Building Blocks Compared

Type Where It Lives What It Does
Rules AGENTS.md, GEMINI.md, or .agents/rules/*.md Enforces coding standards, tech stack choices, or response style. Loaded hierarchically from edited subfolder up to repo root.
Skills .agents/skills/<name>/SKILL.md (or ~/.gemini/config/skills/) Step-by-step SOPs (like PDF creation, security audits, or TDD). Uses progressive disclosure: full text loads only when relevant!
Plugins Marketplace or ~/.gemini/config/plugins/<name>/ Packages skills, rules, subagents, and MCP servers into one toggleable unit (e.g., Google Docs/Drive/Sheets, Vercel, Figma, Android).
MCP Servers mcp_config.json (or inside a Plugin) Model Context Protocol bridges giving the agent live tools for GitHub PRs, SQLite/Postgres, Chrome DevTools, Linear, or Canva.
Hooks hooks.json Shell scripts triggered automatically at lifecycle events (e.g., auto-formatting or secret scanning before/after tool calls).
Team vs. Personal Scope (Priority Order)
Put team-shared rules and skills in your repo's .agents/ folder and commit them to Git—Workspace configs always take top priority! Put personal preferences (like your preferred writing voice or global plugins) in ~/.gemini/config/ so they follow you across every project on your Mac.

10. Headless Power: agy CLI, Python SDK & Cron Jobs

When you want agentic power directly inside your macOS terminal, automated Python scripts, or recurring morning briefings, reach for these three headless workflows.

1 Antigravity CLI (agy)

Open Terminal in any project folder and run agy:

$ cd ~/my-project && agy
> /help         # list TUI slash commands
> /plan Fix flaky auth test
> /exit         # or Ctrl+D twice

Settings live in ~/.gemini/antigravity-cli/settings.json.

2 Scheduled Automations (/automation)

Run standing jobs on a cron schedule—even while away from your desk:

  • Type /automation (or /schedule) in Antigravity 2.0.
  • Pick cadence (Hourly, Daily, Weekly, or custom cron).
  • Antigravity writes a background sidecar config that wakes a fresh agent session on schedule (e.g., "Every weekday at 8:30 AM, check open GitHub PRs and Google Calendar and write a priority brief").

3. Programmatic Orchestration with the Official Python SDK

Install via pip install google-antigravity (includes the compiled local runtime binary) to embed full Antigravity agents inside your own Python pipelines, CI checks, or data workflows:

import asyncio, sys
from google.antigravity import Agent, LocalAgentConfig, CapabilitiesConfig

async def run_audit():
    # Default is safe read-only mode; pass CapabilitiesConfig() to enable file edits/commands
    config = LocalAgentConfig(
        system_instructions="You are a concise release-readiness auditor.",
        capabilities=CapabilitiesConfig(),
    )
    async with Agent(config) as agent:
        response = await agent.chat("Run pytest and summarize any failing edge cases.")
        async for call in response.tool_calls:  # inspect live tool calls!
            print(f"[Tool] {call.name}")
        async for token in response:
            sys.stdout.write(token)

asyncio.run(run_audit())

11. Safety Controls, Troubleshooting & Quick Cheat Sheet

Autonomous agents are only fun when you trust their boundaries. Antigravity gives you granular global and per-project guardrails under Settings → Permissions & Sandbox.

The 4 Permission Dials You Control

1 Tool Execution Policy

Choose proceed-in-sandbox (recommended sweet spot: fast autonomy inside an OS sandbox), request-review (ask before shell commands), or strict.

2 Workspace Boundary Policy

Set Non-Workspace File Access to ask or deny so an agent working on one project never touches unrelated folders without permission.

3 Command Allowlist / Denylist

Pre-approve safe daily commands (npm test, git status) while hard-blocking destructive ones (rm -rf, production deploy scripts).

4 Browser & Network Allowlist

Restrict which external domains the Browser Subagent can visit; sensitive login and payment pages always pause and hand control back to you.

Quick Troubleshooting Fixes

Common Situation Fastest Fix
Agent repeats a mistake you already fixed Type /learn or add a 1-line rule to your project's AGENTS.md so it sticks forever.
Long conversation starts feeling slow or fuzzy Run /compact to distill context, or start a fresh chat referencing @previous-conversation.
Need Google Docs, Figma, or Linear integration Ask: "Search the Marketplace for Figma / Google Docs" and click Add on the plugin card.
Want Jules-style isolated parallel attempts Ask Antigravity to "spawn 2 branched subagents comparing Approach A vs. Approach B."
Power-User Pre-Flight Checklist: 6 Habits for Effortless Sessions
☐ 1. Drop an AGENTS.md in repo root with build/test commands.
☐ 4. Enable proceed-in-sandbox for safe, interruption-free speed.
☐ 2. Start multi-file work with /plan or /grill-me.
☐ 5. Verify web UI changes with the live /browser agent.
☐ 3. Tag exact targets with @file or @folder.
☐ 6. Use /goal or /teamwork on AI Ultra for big builds!

Save the 12-Page Magazine PDF

Keep a print-ready copy of Google Antigravity & Jules for offline reading, desk reference, or sharing with teammates (3.4 MB).