Google AntigravityGoogle Antigravity
Google Antigravity 是 Google 的 agent-first 开发平台:桌面指挥中心/CLI/IDE 扩展/Python SDK 四面共用一套 harness,命令执行策略与工作区边界是官方显式给的三档控制面。
Google Antigravity is Google's agent-first development platform: desktop command center / CLI / IDE extensions / Python SDK — four surfaces on one harness, with command-execution policy and workspace boundaries as three explicit control switches in the official docs.
适合与不适合When to use it, when not to
适合:Google 系模型+多面协同、定时任务与后台子 agent、执行策略显式可调。 不适合:大陆直连场景(主域直连 000)、要产物脱离 Google 账号。
Fits: Google-family models plus multi-surface coordination (desktop command center / terminal / IDE extensions / SDK on one harness), scheduled tasks and background subagents, teams wanting explicit command-execution policies. Not a fit: mainland direct access (main domain measures 000 direct); artifacts living outside the Google account system.
中文生态China availability 国产 / 国内可直接用 / 备案状态 · 快照domestic / direct access / CAC filing · snapshot 2026-10-21
- 出品属地Origin
- 国外出品foreign-made Google(美国)。Google (US).
- 国内使用Direct access
- 直连不可达direct blocked antigravity.google 大陆直连实测 000(2026-10-21,代理下 200)——直连不可达,需网络环境。antigravity.google measured 000 direct from mainland (2026-10-21; 200 via proxy) —unreachable without a network path.
- 备案状态CAC filing
- 未见于备案名单not in filing list:未见于网信办公示名单(2026-10-21 检索);该产品未面向大陆公众提供服务,备案不适用。not in the CAC public registry (searched 2026-10-21); the product does not serve the mainland public, so filing does not apply.
备案信息以网信办《生成式人工智能服务已备案信息》公示名单为准(按批次滚动更新,本站只抄编号不解读);「大陆可直连」是本站从大陆网络直连实测的结果,不是厂商承诺。GitHub 类站点当日实测有波动,口径写在依据里。Filing status is quoted from the CAC public registry (updated batch by batch; this site copies the number and does not interpret it). "Direct access" is this site's own reachability measurement from a mainland network, not a vendor promise. GitHub-hosted sites fluctuated on the measurement day — the note says so.
固定坐标系Fixed coordinate system
8 个维度dimensions, 与同分区其他对象逐项可比each comparable item by item against others in the same section.
- 模型与开放条件Model & openness
官方 overview 页的界面截图直接列出模型选择器:Gemini 系与 Claude 系、GPT-OSS 等并列可选 (截图所示清单,未逐条核验为完整列表);CLI 会话头显示当前模型(如 Gemini 3.8 Flash High)。 模型面由 Google 侧上架决定,第三方模型可用性随官方调整。
The UI screenshot on the official overview page lists the model picker directly: Gemini-family and Claude-family, GPT-OSS etc. shown side by side as selectable (a screenshot-shown list; not verified one by one as a complete list). The CLI session header shows the current model (e.g. Gemini 3.8 Flash High). The model surface is decided by Google-side listing; third-party model availability changes with official adjustments.
- 运行位置Where it runs
四个官方面(overview 页原文口径):Antigravity App(桌面端,v2.5+,从 Agent Manager 演进的 独立指挥中心)、Antigravity CLI(终端客户端,本地 shell/远程云环境/SSH 都能跑)、 Editor extensions(5 个 IDE 的扩展)、Antigravity SDK(Python 编程运行时)。 App/CLI/IDE 扩展的执行都在你本机环境;SDK 是「几行 Python 跑 agent harness」。
Four official surfaces (overview page wording): Antigravity App (desktop, v2.5+, a standalone command center evolved from the Agent Manager), Antigravity CLI (a terminal client that runs on local shell / remote cloud environments / SSH), Editor extensions (extensions for 5 IDEs), Antigravity SDK (a Python programming runtime). App/CLI/IDE-extension execution all happen in your local environment; the SDK is "run an agent harness in a few lines of Python".
- 本地文件Local files
官方 settings 文档明示边界:默认 agent 只能访问活动项目文件夹与 ~/.gemini/antigravity-ide/ (产物、知识项、配置都在这个本地目录);「非工作区文件访问」是显式开关,默认关。
The official settings docs state the boundary explicitly: by default the agent can only access the active project folder and
~/.gemini/antigravity-ide/(artifacts, knowledge items and configuration all live in this local directory); "non-workspace file access" is an explicit toggle, off by default.- 关机后的任务Tasks after shutdown
官方叙事同时覆盖同步 pair-programming 与异步后台 agent:App 支持按项目归档对话、 定时任务(scheduled messages/tasks)自动处理日常工作;CLI 可调度后台子 agent。
The official narrative covers both synchronous pair-programming and asynchronous background agents: the App supports archiving conversations per project and scheduled messages/tasks that handle routine work automatically; the CLI can schedule background subagents.
- 工具与扩展Tools & extensions
产物机制(Artifacts:实现计划/验证报告/提议变更)是核心面;浏览器在环(browser-in-the-loop) 做前端自动化验证;MCP 支持状态未逐条核验(官方文档导航未见独立 MCP 页,本次未逐页读)。
The Artifacts mechanism (implementation plans / verification reports / proposed changes) is a core surface; browser-in-the-loop does front-end automated verification. MCP support status: not verified one by one (no standalone MCP page found in the official docs navigation; not read page by page the 2026-10-08 pass).
- 上下文与记忆Context & memory
知识项(knowledge items)落 ~/.gemini/antigravity-ide/ 本地目录(settings 文档口径); 跨会话记忆的产品化程度未逐条核验。
Knowledge items land in the local
~/.gemini/antigravity-ide/directory (settings docs framing). How productized cross-session memory is: not verified one by one.- 权限与限制Permissions & limits
官方 settings 文档给出三层控制面:终端命令自动执行(请求审查 vs 始终继续+可配白/黑名单)、 非工作区文件访问开关(默认关)、严格模式(限制外部资源与敏感操作)。账号走 Google 体系。
The official settings docs give a three-layer control surface: terminal-command auto-execution (request review vs always continue + configurable allow/deny lists), the non-workspace file access toggle (off by default), and strict mode (restricts external resources and sensitive operations). Accounts go through the Google system.
- 适合什么任务Best for
适合:要 Google 系模型+多面协同(桌面指挥中心/终端/IDE 扩展/SDK 同一 harness)、 要定时任务与后台子 agent、要「命令执行策略」显式可调的团队。 不适合:大陆直连场景(主域直连不稳)、要把产物放非 Google 账号体系的。
Good fit: wanting Google-family models plus multi-surface coordination (desktop command center / terminal / IDE extensions / SDK on one harness), scheduled tasks and background subagents, teams wanting an explicitly tunable "command execution policy".
Not a fit: mainland direct-connection scenarios (the main domain's direct connection is unstable), wanting artifacts kept outside the Google account system.
头号误解Top misconception
- 四个面(App/CLI/IDE 扩展/SDK)是同一 harness 的不同入口,不是四个独立产品——别按 Gemini Code Assist 或 Firebase Studio 的旧认知套它(官方文档有 Firebase Studio 迁移页,那条线并入本平台叙事)The four surfaces (App/CLI/IDE extensions/SDK) are one harness with four entry points, not four separate products — don't map old Gemini Code Assist or Firebase Studio notions onto it (the official docs carry a Firebase Studio migration page; that line folds into this platform's narrative)
- 「始终继续」模式=agent 自动执行 shell 命令(官方原文标注自主性高、风险高)——默认是请求审查,别顺手改完就忘'Always continue' mode = the agent auto-executes shell commands (official wording flags it: high autonomy, high risk) — the default is review-on-request; don't flip the switch and forget
- 模型选择器截图里的清单是界面快照,不是承诺——第三方模型可用性随官方上架变化The model list visible in the overview page's selector screenshot is an interface snapshot, not a promise — third-party model availability changes with official listings
未知项清单Known unknowns
- MCP 支持状态(未逐页读官方文档导航)
- 免费/付费边界与配额细节(D9:价格数字不采;能力边界未逐条核验)
- 跨会话记忆产品化程度(知识项机制细节未逐条核验)
- 定时任务的执行位置(本机 vs 云,未核验到官方声明)
- MCP support status (official docs navigation not read page by page)
- Free/paid boundaries and quota details (D9: price figures not collected; capability boundaries not verified one by one)
- How productized cross-session memory is (knowledge-item mechanism details not verified one by one)
- Where scheduled tasks execute (local vs cloud; not verified to an official statement)
证据来源Evidence sources
判断可回到以下一手源复核。本站核验日Every judgement can be rechecked against the primary sources below. Verified on 2026-10-30,内容更新日content updated on 2026-10-21。
| 类型Type | 名称Name | 链接Link |
|---|---|---|
| official | Google Antigravity · 官网(四产品面叙事)Google Antigravity · official site (four product surfaces narrative) | https://antigravity.google |
| official | Antigravity Docs · Choose your Antigravity(四面对照/模型选择器界面)Antigravity Docs · Choose your Antigravity (four-surface comparison / model picker UI) | https://antigravity.google/docs/overview |
| official | Antigravity Docs · 设置(命令执行/工作区边界/严格模式)Antigravity Docs · Settings (command execution / workspace boundary / strict mode) | https://antigravity.google/docs/settings |
| changelog | Antigravity Docs · Release Notes(changelog)Antigravity Docs · Release Notes (changelog) | https://antigravity.google/docs/release-notes |
实测记录Measured runs
本档未跑端到端编码任务;能力结论来自官方文档,编码向测量协议见 protocols/coding/_protocol.md。
实测建议(三条,都指向官方可读页没给的东西):
- 模型选择器实际完整清单(overview 页截图非承诺清单)
- 定时任务在 App 关闭后是否继续跑(官方叙事有 scheduled tasks,执行位置未逐页读)
- 严格模式各子开关的实际拦截行为
This site has not completed measured runs. The test protocol is at protocols/coding/_protocol.md.
Suggested measurement (three, each pointing at something the official readable pages do not give):
- The actual complete list in the model picker (the overview page screenshot is not a committed list)
- Whether scheduled tasks keep running after the App is closed (the official narrative has scheduled tasks; execution location not read page by page)
- The actual interception behavior of each strict-mode sub-toggle
相关条目Related entries
- Gemini CLIGemini CLI同厂终端 agent 前代线:Antigravity CLI 与它的关系官方未逐页声明。Same vendor's terminal-agent predecessor line: the relationship between Antigravity CLI and it is not declared page by page by the official.
- JulesJules同厂云异步线:Jules 在云 VM 跑,Antigravity 四面都在你本机。Same vendor's cloud async line: Jules runs in cloud VMs; all four Antigravity surfaces run on your machine.
- Chrome DevTools MCPChrome DevTools MCP另一层 · MCP 工具Other layer · MCP tools
本页由This page is generated from the ai-handbook 数据层生成(CC BY 4.0)。data layer (CC BY 4.0).
方法论与坐标系定义见仓库内For the methodology and coordinate definitions, see METHODOLOGY.md.