Back to Gap Map Product / UX / Orchestration & agents

Antigravity

Google
closed / Overall score: n/a

Google's agent-first development platform, shipping a desktop application, a CLI, an IDE and an SDK. It runs agentic development workflows on macOS, Windows and Linux with model optionality across Google, Anthropic and OpenAI models, and integrates with Google Cloud, Firebase and Android. Its CLI supersedes the earlier Gemini CLI.

Openness

1 high confidence
1.0
source
closed
license
proprietary(Google product, free public preview)
self-host
no

Antigravity is a Google service used under Google's own terms, with downloadable apps but no published source and no way to self-host. Its free individual plan and its support for third-party models do not change that, since neither opens the platform's code.

  • https://antigravity.google/product recorded 2026-09-29

    The product page offers the desktop app, CLI, IDE and SDK as downloads, describing "The fully-featured, agentic IDE" and "Simple Python scripts to iterate on agentic applications"; searched the body for source code, repository, open source, self-host and on-prem: none.

  • https://antigravity.google/terms recorded 2026-09-29

    The Antigravity terms page says use of the service is governed by "these Google Antigravity Additional Terms of Service" on top of the Google Terms of Service, a proprietary service agreement with no open source license; searched the body for open source license and found only "third party or open source model" as a model-choice remark.

Adoption

not assessed

Antigravity is a free public preview and Google discloses no user count for it specifically, so there is no honest adoption signal to read. Google's own developer-highlights post describes the Antigravity ecosystem, the 2.0 desktop app and Managed Agents in the Gemini API without giving a download, install, developer or active-user count anywhere.

Capability

4 medium confidence
4.0

Antigravity has broad agentic features but no published SWE-bench score, so it is rated on feature set alone. That leaves it one step below OpenHands, which has a verified benchmark result.

Verified 2026-09-29