Antigravity IDE is Google's agent-first code editor, pairing a full development environment with autonomous AI agents that plan, code, run commands, and verify their own work through reviewable artifacts.
Antigravity IDE is Google's agent-first development environment: a complete code editor with an autonomous agent working beside you. You create a project, describe the outcome you want in plain language, and the agent plans the work, edits files, runs terminal commands, and drives a browser when a task needs the web. Rather than approving every individual tool call, you review higher-level deliverables such as implementation plans, task lists, screenshots, and browser recordings, then comment on them to steer the next attempt. Familiar editing features stay available for hands-on work, so Antigravity IDE fits developers who want to delegate an entire feature and those who prefer to stay in direct control.
The wider Antigravity platform is now a family of products rather than one application: a standalone hub for monitoring parallel agents, extensions for Visual Studio Code, JetBrains, and Zed, plus command-line and SDK surfaces for automation. Inside Antigravity IDE, a project can bundle several folders or repositories, so an agent understands frontend and backend context at once, and every project carries its own permissions for file paths, URLs, and tools. Security presets decide whether terminal commands and file edits need your approval first. Pairing that scope control with an Intelligent Terminal means agents run builds and tests exactly where a change lands, and the results stay easy to verify.
The practical gain is throughput without losing oversight. Work that once meant switching between an editor, a shell, and a test runner can be delegated, and the agent returns an artifact you can inspect before anything ships. Because agents operate across the editor, terminal, and browser, they can reproduce a bug, patch it, and confirm the fix instead of stopping at a suggestion. Running several agents in parallel on separate Git worktrees keeps experiments from colliding with your own working tree, while per-project permission sets and URL allowlists limit what an autonomous run is allowed to touch. Reviewing a plan and a screenshot is far quicker than reading raw tool logs, which makes results easier to trust and easier to pass to a teammate. Teams already comfortable in Visual Studio Code, JetBrains, or Zed keep their habits too, because Antigravity IDE extends familiar workflows rather than replacing them.
Key capabilities and highlights.
A complete editing surface with tab autocompletion, natural-language commands, and a dedicated agent sidebar, so you can write code directly or hand a whole task over without leaving Antigravity IDE. The editor keeps familiar shortcuts and file navigation, which makes the move from another code editor a short one.
Launch and supervise several agents at once across different workspaces. Each agent can run in your active folders or inside an isolated Git worktree, so long-running or experimental jobs never disturb the branch you are editing. Progress, plans, and outcomes stay visible in one place, making it realistic to keep three or four tasks moving while you work on something else.
Agents report progress as implementation plans, task lists, screenshots, and browser recordings rather than an opaque stream of tool calls, and you can comment on any artifact to correct course before more code is written. This review layer keeps autonomy accountable: you approve the approach first, then judge the finished result against the plan you already read.
Agents can open Chrome, navigate pages, and debug web behaviour on their own, which lets them test a fix in a real browser and return evidence rather than a guess. That matters for anything front-end facing, where a change has to be confirmed visually and interactively before you would call it finished or hand it over for review.
Conversations can run on Google's Gemini models as well as Anthropic's Claude models and open-weight alternatives, so you can match the model to the difficulty and cost profile of each task. Switching mid-project is straightforward, which helps when a routine refactor does not need the same reasoning power as an architectural change.
Define a project around several folders or repositories for full cross-repository context, then set isolated rules for allowed file paths, permitted URLs, and which tools an agent may call. Each project inherits sensible defaults but can override them, so a sensitive backend repository can demand stricter approvals than a scratch folder used for quick experiments.
Connect local or remote Model Context Protocol servers, including ready-made integrations for Google Cloud services, to ground agent work in your own tools, databases, and documentation. Because tool access is configured per project, an agent working on internal reporting can reach the right systems without carrying every integration you have ever installed.
Queue agent instructions to run once or on a recurring schedule, and capture repeatable procedures such as code review or release checks as skills the agent can follow again on later projects. Scheduled work and reusable skills turn Antigravity IDE from a place you visit into a background teammate that keeps handling the routine parts of a codebase.
Browse historical releases and previous builds.
Common questions about Antigravity IDE.
It is Google's agent-first integrated development environment. You work in a familiar code editor while autonomous agents handle multi-step tasks such as implementing a feature, fixing a bug, or running a test suite, reporting back through plans and other reviewable artifacts.
A typical assistant suggests code inside your current file and waits for you. Antigravity IDE delegates at the level of a whole task: agents plan, edit multiple files, run commands, and check their own results, while you supervise progress through artifacts rather than approving each individual action.
Artifacts are an agent's verifiable deliverables, such as an implementation plan, a task list, screenshots, or a browser recording. They show what the agent intends to do and what it actually produced, and you can leave comments on them to guide the next step.
Yes. Agents in Antigravity IDE can work across the editor, the terminal, and an integrated browser. They can run build and test commands, and they can drive Chrome to inspect a page or reproduce an issue. Security presets and project permissions decide whether those actions need your approval first.
Antigravity IDE supports Google's Gemini models alongside Anthropic's Claude models and open-weight alternatives. You choose the model per conversation, which is useful when one task needs a fast, inexpensive model and another benefits from stronger reasoning.
Yes. The agent manager is built for parallel work, so you can supervise multiple agents across projects and workspaces. Agents can operate directly in your active folders or inside isolated Git worktrees, which keeps concurrent experiments from interfering with your current branch.
Scope each project to specific folders or repositories, then use security presets to require approval for terminal commands and file writes. You can also allowlist or block URLs and limit which tools and MCP servers a project's agents may use, so autonomy stays inside boundaries you set.
Yes. It supports local and remote Model Context Protocol servers, including packaged integrations for Google Cloud services, so agents can query your databases, issue trackers, or documentation instead of guessing. Projects can also restrict which MCP tools are available to keep context focused.
Antigravity IDE is Google's agent-first code editor, pairing a full development environment with autonomous AI agents that plan, code, run commands, and verify their own work through reviewable artifacts.
Visual Studio Enterprise 2022 is an integrated development environment for building, testing, debugging, profiling, and maintaining complex applications with advanced tools for professional teams and large codebases.
Visual Studio Professional 2022 is an integrated development environment for writing, building, testing, debugging, and managing applications through a unified workspace designed for professional developers and small teams.
Visual Studio Community 2022 is a free Windows development environment for eligible users, combining code editing, compiling, debugging, and testing tools for desktop applications, web services, and games.
Microsoft Visual Studio Code is a free Windows code editor that combines language-aware editing, debugging, source control, terminal access, testing, extensions, and project navigation in one configurable development workspace.
Comments