Antigravity — Quick reference

Google's agent-first dev platform · hand whole tasks to AI agents · desktop app + CLI + SDK · Mac/Windows/Linux · free For Individuals tier

At a glance

What it is: instead of one assistant in a chat box, Antigravity gives AI agents their own space to work. A Manager Surface spawns, orchestrates and watches many agents in parallel; an Editor View is a full AI IDE with tab-completion and inline commands. You hand over a whole task — plan, code, run, self-correct — rather than copy-pasting snippets. Ships as a desktop app, a CLI, an SDK, and a managed agent service. Runs: Google's platform — needs a Google account, not self-hosted. Pricing one-liner: For Individuals = $0/mo with basic weekly rate limits; paid tiers raise the limits.

Setup & first task

1 · Install
Get the desktop app
Download Antigravity for Mac / Windows / Linux from antigravity.google.
2 · Sign in
Google account
Sign in with a Google account. Pick the For Individuals plan — $0/month, no subscription needed.
3 · First task
Hand it to an agent
On the new-agent screen, pick a project, leave the model on Gemini 3.5 Flash, paste a task, and send. The agent plans, writes code + a test, runs it, and iterates until it passes.
# Paste into the new-agent prompt box, then send:
Write a Python script that fetches a UniProt entry by accession,
prints the protein name, length, and organism, then add a test and
run it — fix any errors until the test passes.

Key surfaces & controls

ControlWhat it does
@In the prompt box: mention files / context for the agent
/In the prompt box: run an action
mic buttonDictate the task instead of typing (live voice transcription)
Model selectorSwitch the model per task (Gemini / Claude / gpt-oss — see below)
Local / New WorktreeChoose where the agent runs — your branch, or an isolated Git worktree
Manager SurfaceControl room: spawn, orchestrate & observe multiple agents in parallel
Editor ViewFull AI IDE — tab-completion + inline command requests beside the running agent
ArtifactsAgent's task lists, screenshots & browser walkthroughs — review & comment like a doc

Antigravity CLI

The same agents, in your terminal. On first run it greets you with "Welcome to Antigravity CLI!" and a colour-scheme picker (light, dark, tokyo night and more). Type a task in plain English; the agent (shown as AGY) replies with the change as a reviewable code diff before you accept.

you: add a greeting function
AGY: Here's the change   → shown as a diff to review

Beyond the app and CLI, Antigravity also ships an SDK and a managed agent service — the same agents can run inside your own scripts and pipelines.

Prompting patterns

Outcome + check
Give it something to verify
Name a goal and a way to check it — "add tests and run them until they pass". Because the agent can run code, a verifiable instruction lets it self-correct instead of guessing.
What + where
Be specific and located
"add a second function that fetches a sequence by accession" beats "make it better". Specific, located asks win.
Follow up
Build in the same agent
Send the next step as a follow-up in the same agent so it builds on work already done instead of starting over.
Parallel
Independent tasks only
"Agent 1: refactor into modules with type hints. Agent 2: write pytest tests for the normalisation step." No shared file, no "do A then B" — so they run at once.
Scheduled
Hand off recurring work
"Every night, pull the latest commits, run the full test suite, and write me a summary of any failures." Set it as a cron-scheduled task.
Maintenance
Async bug-fix loop
Hand over a maintenance job — the agent reproduces the issue, generates a test case, and implements the fix across editor, terminal and browser.

Context, models & modes

NeedHow
Add contextType @ in the prompt box to mention files / context for the agent
Isolate experimental workPick New Worktree — the agent gets its own Git worktree, main stays clean; review before merging
Trusted small editsPick Local — runs the agent on a chosen branch (e.g. main)
Inline IDE workEditor View — tab-completion + inline command requests beside the agent
Many agents at onceManager Surface — spawn/observe parallel agents; split a big job across subagents
Work while awayRun an agent in the background, or a cron-scheduled task, and review later
Verify the workRead the agent's Artifacts — task lists, screenshots, browser walkthroughs — and comment

Model-agnostic. From the model selector, switch between Gemini (3.5 Flash, 3.1 Pro, 3 Flash), Anthropic Claude Sonnet & Opus 4.6, and gpt-oss-120b — all inside one platform, all on the free plan. Use a stronger model on a hard task, a fast one for routine edits.

Pricing & limits

TierCost & what you get
For Individuals$0/mo — Gemini 3.5/3.1/3, Claude Sonnet & Opus 4.6, gpt-oss-120b; unlimited tab-completions & command requests; basic weekly rate limits
AI Pro / AI UltraPaid tiers add higher rate limits & a flexible AI credit pool — see antigravity.google/pricing for current pricing

Parallel, background and scheduled runs all draw from the same weekly quota — they finish a multi-part job faster but don't make it cheaper. See antigravity.google/pricing for current plan pricing.

Gotchas & tips

Agents act on your code — supervise. Isolate risky runs in a New Worktree and review the Artifacts before merging; discard cleanly if you don't like the result. Batch related changes into one task rather than many tiny ones — every request draws down the weekly limit, and a real prototyping session can exhaust the free tier quickly. For parallel agents, keep the tasks independent (no shared file). Young & fast-moving: 2.0 was a full rebuild shipped May 2026 — expect change. Docs: antigravity.google/docs.