Does Gemini Code Assist train on your code? What changed when the free tier closed

The Morse teamPublished November 10, 202610 min read

As of November 2026, Gemini Code Assist doesn't train on your code, because the only editions Google still serves are Standard and Enterprise, and Google says those never use your prompts or responses to train its models. The free Gemini Code Assist for individuals, which did use your code for training unless you opted out, stopped serving requests on June 18, 2026 (Google). Google moved its users, along with Google AI Pro and Ultra subscribers, to Antigravity, whose terms say Google uses your code to improve its models and lets employees and contractors review it unless you turn off Enable Telemetry (Google).

We read Google's Gemini Code Assist deprecation pages, its data governance, FAQ, and security pages for Standard and Enterprise, the Gemini CLI terms and configuration docs, the Gemini API terms and abuse monitoring page, and Antigravity's terms, settings, and pricing pages on November 10, 2026. Google's old privacy notice for individuals now redirects to a deprecation page, so we read it in the Internet Archive's copy from May 18, 2026. Google deserves credit on the business side, where its Standard and Enterprise pages say plainly what's sent, what's kept, and that none of it trains Gemini.

Gemini Code Assist training at a glance

Is Gemini Code Assist still available?

Only to organizations, and only for a while. Google's deprecation table lists Gemini Code Assist consumer accounts as deprecated on May 18, 2026 and shut down on June 18, 2026, when the IDE extensions and Gemini CLI stopped serving Gemini Code Assist for individuals, Google AI Pro, and Google AI Ultra (Google). The consumer version of Gemini Code Assist on GitHub stopped taking installations the same day and stopped reviewing code on July 17, 2026 (Google).

The paid editions are winding down too. Google stopped selling new Standard and Enterprise subscriptions on October 9, 2026 (Google Cloud). Annual subscriptions can renew until January 31, 2027, monthly ones until December 31, 2027, and Gemini Enterprise subscriptions keep their Code Assist entitlement until March 31, 2027. Google points both groups to Antigravity.

Did the free Gemini Code Assist use your code for training?

Yes, unless you opted out. Google's notice for individuals said it collected "your prompts, related code, generated output, code edits, related feature usage information, and your feedback to provide, improve, and develop Google products and services and machine learning technologies" (Google, archived May 18, 2026). Human reviewers could "read, annotate, and process" that data after Google disconnected it from your account, and Google kept those disconnected copies "for up to 18 months."

The opt-out lived behind a link. In VS Code, you opened the Gemini Code Assist chat pane, clicked More, and chose Privacy settings, which led to a page where you could opt out of Google using your data to improve its models (Google, archived May 24, 2026). Google's setup page added that "these privacy settings are stored at the IDE level," so opting out in VS Code didn't carry over to IntelliJ.

The guides ranking for this question on November 10, 2026 still describe that toggle as the way to protect your code. The toggle's page is gone, and Google's current pages don't say what happens to the reviewed copies made before June 18, which the old notice said could be kept for up to 18 months.

Is Gemini Code Assist the same as Antigravity?

No. Antigravity is the product Google moved personal users to, and it runs under its own terms. Antigravity's pricing page lists a $0 plan for individuals, plus Google AI Pro and Ultra tiers with higher limits (Antigravity). On any of them, with a personal Google Account, the Antigravity Additional Terms of Service apply:

• What's recorded. Google says it will "record and store your user data, interaction data pertaining to your usage of the Service, related metadata connected to the Service, and any feedback you provide," which it calls Interactions (Antigravity).
• What it's used for. "We use Interactions to evaluate, develop, and improve Google and Alphabet research, products, services and machine learning technologies."
• Who reads it. "Google employees and contractors may access, view, review and use Interactions." Google asks you not to submit "confidential information or any data you wouldn't want a reviewer to see."
• How long. The terms give no period. Interactions are used "unless and until you request deletion," and the way to request it is an email to antigravity-support@google.com.

To opt out, open Antigravity's Settings, go to the Account section, and turn off Enable Telemetry; in Antigravity CLI, the same switch is under /config (Antigravity). The CLI's version "is saved with your account instead of in settings.json," so unlike the old per-IDE setting, your choice follows you between machines. Turning it off has limits, in Google's words: "Even when your Telemetry is toggled off, Google still uses your data to provide the service, process any feedback and help keep Antigravity safe" (Antigravity).

If you choose a non-Google model as Antigravity's main agent, you're also bound by that model maker's terms, which for Anthropic means its commercial terms (Antigravity).

Does Gemini Code Assist Standard or Enterprise use your code?

To answer you, yes; to train Gemini, no. Google says "Gemini doesn't use your prompts or its responses as data to train its models" (Google Cloud). The one opt-in exception is the Gemini for Google Cloud Trusted Tester Program, where shared data is "used for product improvements, not for training Gemini models."

What goes to Google in each request is more than the line you're typing. Google's security page lists your prompt, the response, the conversation history, "snippets of files that are open in the IDE, snippets of files that are stored adjacent to the open file, and cursor location" (Google Cloud). Because the service is stateless, Google says it doesn't store prompts and responses unless your organization sets up a Cloud Logging bucket. Engineers see telemetry, such as whether you accepted a suggestion and its length, but "not the contents of the request." Processing happens at the nearest data center, though "regionality is not guaranteed."

Does Gemini CLI train on your code?

It depends on how you sign in, and the CLI's own terms page hasn't caught up. Google's Gemini CLI terms list three methods: a Google account through Gemini Code Assist, a Gemini Developer API key, and a Vertex AI key (Gemini CLI). As of November 10, 2026, that page still says signing in with a Google account sends you to the sign-up for Gemini Code Assist for individuals, and links to a privacy notice that now redirects to a deprecation page. The Gemini CLI README says the CLI now serves organizations with Standard or Enterprise licenses, "plus legacy paid API / Vertex keys" (Gemini CLI).

Antigravity CLI also accepts a Gemini API key (Antigravity), so the key's own rules follow it into either CLI:

• Gemini API key. An unpaid key falls under terms that let Google improve its models with your code and have reviewers read it, and a paid key doesn't; we cover the details in is the OpenAI API private, which has a section on the Gemini API (Google AI).
• Vertex AI, now Gemini Enterprise Agent Platform. Google "won't use your data to train or fine-tune any AI/ML models without your prior permission or instruction" (Google Cloud).

Separately, Gemini CLI sends usage statistics by default. Google says these cover tool names, model names, and timings, not prompt or file content, and you can stop them by setting privacy.usageStatisticsEnabled to false in settings.json (Gemini CLI).

Who can use your code, by how you sign in

Every row comes from Google's own pages, read November 10, 2026.

Across Google's tools, the code of a person paying nothing, or paying through Google One, can train Google's models unless they opt out, and the code of a company paying through Google Cloud can't. We found the same split on Gemini's chat side, covered in is Gemini private, and GitHub draws a similar line in does GitHub Copilot train on your code.

How Morse keeps your code questions away from your Google Account

Every rule above follows from where your code lands. It reaches Google under your Google Account, and the account's plan decides whether Google may train on it, who may read it, and for how long. On a personal account, finding the switch is your job, and Google moved it when the free tier closed.

Morse is a chat app in your browser. When you want help with a function, a stack trace, or a config file with the secrets removed, you paste the part you need into a chat. Morse doesn't train AI on your content, so there's no telemetry switch to find. The Free plan includes coding, file, and other chat tools. Pick a mode per chat:

• Anonymous, to ask Gemini without your Google Account behind the question. Gemini is one of the proprietary models in Anonymous mode, with GPT and Claude. Our proxy removes your account, email, and IP address and sends the message under Morse's own account. Google sees the code without knowing it's you, and may keep it for a limited time to check for abuse, under its own policy, so strip out keys and internal hostnames before you paste.
• Private, when you don't want the model's host holding a copy of your function. New chats start here. DeepSeek, Kimi, GLM, and other open-weight models, plus xAI's Grok, get your question with your identity stripped, and Morse has zero data retention agreements with its inference, search, and other providers, so they keep nothing. Where a provider confirms zero retention on each reply, Morse checks it and refuses a reply that doesn't. Here's what zero data retention means.
• E2EE, for source only a verified enclave should read. E2EE runs open-weight models like GLM, Kimi, and DeepSeek, without web search or image generation. Before your message leaves the device, your browser verifies the enclave's proof that it's genuine, then seals the message so only that enclave can open it, and our servers pass along ciphertext they can't read. The guarantee holds as long as the enclave hardware and attestation work as designed. It's part of the $19-a-month Pro plan; here's how end-to-end encrypted AI chat works.

Whichever mode you pick, your chat history is stored in your browser, encrypted, and not on our servers. It locks when you sign out, and opening it takes two keys: one that never leaves your device and one our server gives you at sign-in. Attached files go with the message and are never stored by Morse. Morse doesn't sell your data or show ads, and because we don't store chats, we usually have very little to give authorities (Morse privacy policy). Clearing your browser deletes your chats for good, and we can't recover them, so copy any answer you'll want later into your own notes.

Which Google setting to check, by tool

• Antigravity on a personal account: turn off Enable Telemetry under Account, and email antigravity-support@google.com to delete what's already been collected.
• Gemini CLI or Antigravity CLI with an API key: use a paid key, or Google's unpaid terms apply.
• Gemini CLI usage statistics: set privacy.usageStatisticsEnabled to false.
• Code Assist Standard or Enterprise: ask your admin whether Cloud Logging stores prompts, and plan for the subscription's end date.
• A coding question you'd rather not tie to your Google Account: ask it on Morse, in Anonymous mode for Gemini, or in E2EE for source only a verified enclave should read.

Try a chat from the Morse home page without an account, or start free with no card; E2EE chats come with Pro.