Gemini CLI vs Gemini Code Assist: Which Google Tool

Gemini CLI is an agent you run in a terminal: you type a goal, and it reads the repository, edits files and runs commands until it is done. Gemini Code Assist is an assistant inside your editor, VS Code or a JetBrains IDE, with completions as you type, a chat panel and an agent mode. They are closer than the names suggest. In VS Code, Code Assist’s agent mode is powered by Gemini CLI, the two share quotas, and they read the same ~/.gemini/settings.json. So the real choice is where you want to work, not which engine you want. One more thing changed this year: since 18 June 2026, neither serves Google’s individual and consumer tiers, so for most people starting today the question is licence first, tool second.

The short version

• Gemini CLI: a terminal agent, open source under Apache 2.0, scriptable with gemini -p, configured through settings.json and GEMINI.md.
• Gemini Code Assist: IDE extensions for VS Code, JetBrains IDEs, Android Studio, Cloud Workstations and Cloud Shell Editor, sold to organisations through Google Cloud as Standard and Enterprise editions.
• Agent mode in Code Assist: the multi-step, tool-using mode in the IDE chat. In VS Code it runs on Gemini CLI; in IntelliJ it has its own tools.
• Individuals: the free Code Assist tier and Google AI Pro and Ultra sign-ins stopped working in both products on 18 June 2026. Google points those users to Antigravity.

First: who can still use them

Before comparing features, check that you can sign in. Google’s notice on consumer accounts↗ (opens in a new tab) says that from 18 June 2026 the Code Assist IDE extensions and Gemini CLI stopped serving requests for Gemini Code Assist for individuals, Google AI Pro and Google AI Ultra, and that the Login with Google option no longer works for those tiers. Access through Code Assist Standard or Enterprise subscriptions is unchanged. The same notice says consumer users can move to the Antigravity family of products.

Gemini CLI’s own announcement adds that API key authentication is unaffected, and its authentication page now opens with a note that Gemini CLI was replaced by Antigravity CLI on 18 June for unpaid-tier and Google One users. Some older guides, and parts of the repository’s README, still describe a free personal tier. Trust the dated notices. In practice, Gemini CLI today means a Gemini API key, Vertex AI, or an organisation’s Code Assist licence with a Google Cloud project set in GOOGLE_CLOUD_PROJECT.

Gemini CLI: the terminal agent

Gemini CLI installs with npm install -g @google/gemini-cli or Homebrew, and runs in any terminal. It has four approval modes, from asking before every change to approving everything, a read-only plan mode, an optional sandbox, subagents, extensions and a headless mode for scripts and CI. Those are covered elsewhere: the modes and plan mode in Gemini CLI plan mode, and how it stacks up against Anthropic’s and OpenAI’s agents in Gemini CLI vs Claude Code vs Codex CLI.

What matters for this comparison is that the CLI is the engine. It does not care which editor you have open, it can run on a server with no editor at all, and its source is public, so you can read exactly what it does before it runs on your machine.

Gemini Code Assist: the IDE assistant

Code Assist lives in the editor. Google’s Code Assist overview↗ (opens in a new tab) lists VS Code, JetBrains IDEs such as IntelliJ and PyCharm, Android Studio, Cloud Workstations and Cloud Shell Editor. Its everyday features are the ones you notice without asking: completions and generated code as you type, and a chat panel that works with the code you have open. It also reviews pull requests in GitHub repositories, with behaviour you can customise.

It is bought and administered through Google Cloud, and a licensed user works against a Google Cloud project. That is the point of it for many teams: the same billing, identity and data governance as the rest of their Google Cloud estate.

Agent mode: where the two meet

Agent mode is Code Assist’s answer to the terminal agent: a chat mode that plans a multi-step task, uses built-in tools and MCP servers, and proposes edits across files. Google’s Gemini CLI page for Code Assist↗ (opens in a new tab) says agent mode is powered by Gemini CLI in VS Code, that each Code Assist edition provides quotas for Gemini CLI, and that those quotas are shared between Gemini CLI and agent mode. A long agent session in the IDE and a long session in the terminal draw on the same allowance.

The two IDEs differ. According to Google’s agent mode guide↗ (opens in a new tab), which marks the feature as Preview:

• VS Code: switch the Agent toggle in the Code Assist chat. MCP servers go in ~/.gemini/settings.json, context in GEMINI.md files globally or in the project, and auto-approval is the geminicodeassist.agentYoloMode user setting.
• IntelliJ: open the Gemini tool window and choose the Agent tab. MCP servers go in an mcp.json in the IDE’s configuration directory, context in GEMINI.md or AGENT.md at the project root, and auto-approval is a checkbox under Agent options.

The practical consequence in VS Code: one settings file configures both. An MCP server you add for Gemini CLI is there in agent mode too, and a GEMINI.md you wrote for the terminal steers the IDE agent the same way.

{
  "mcpServers": {
    "fenbs": { "httpUrl": "https://fenbs.ai/api/mcp" }
  }
}

The Gemini CLI Companion is not Code Assist

A third piece is easy to confuse with Code Assist. The Gemini CLI Companion↗ (opens in a new tab) is an extension for VS Code, VS Code-compatible editors and Antigravity that connects a Gemini CLI session running in the editor’s terminal to the editor itself. The CLI then sees your recently opened files, cursor position and selection, and shows its proposed changes in the editor’s own diff viewer, where you accept or close them. You install it by running /ide install inside Gemini CLI, or from the marketplace followed by /ide enable.

So “Code Assist vs the Companion” is really “an IDE assistant vs a terminal agent that can see your IDE”. The Companion adds no completions and no chat panel; the conversation stays in the terminal. If you like Gemini CLI but want diffs where you read code, the Companion is enough.

Editions, by name

Google’s documentation names two current editions of Code Assist, Standard and Enterprise, both bought through Google Cloud. Enterprise adds code customisation, which lets an organisation connect its private repositories so suggestions reflect its own code. Gemini Code Assist for individuals, the consumer edition, is listed under deprecations, together with the consumer version of its GitHub code review. Check Google’s current plans for prices and quotas rather than a comparison table.

Where Antigravity fits

Antigravity is where Google has sent individual developers. Its overview↗ (opens in a new tab) describes Antigravity 2.0 as a standalone desktop app for managing agents, available with a CLI, an IDE and extensions for editors such as VS Code and JetBrains. The CLI, agy, offers to import Gemini CLI settings, skills and MCP servers when you first run it. If you signed in to Gemini CLI or Code Assist with a personal Google account, that is the migration path; the comparison with Anthropic’s agent is in Antigravity vs Claude Code.

Who each suits

• You live in the terminal, script your tools or run agents in CI: Gemini CLI, with an API key, Vertex AI or your organisation’s licence.
• You want completions and chat in VS Code or a JetBrains IDE, under Google Cloud administration: Gemini Code Assist.
• Your organisation already has Code Assist licences: use both. Agent mode in VS Code and the CLI share quotas and configuration, so choose by task, not by account.
• You want Gemini CLI with diffs in your editor, and no second product: the Gemini CLI Companion.
• You are an individual who used the free tier or a Google AI subscription: Antigravity, since neither product serves those accounts any more.

Keeping the work on one board

An engine shared between terminal and IDE still leaves each session’s record in its own place: the terminal history, the chat panel, a plan file. The work list belongs somewhere colleagues can see it. With the entry above in settings.json, both Gemini CLI and agent mode in VS Code can connect to the same fenbs board. Run /mcp auth fenbs in Gemini CLI to sign in through the browser, tick what it may do, and every task it creates, moves or comments on is recorded in History under the assistant’s name, on your behalf, whether the session ran in the terminal or the editor.

A browser sign-in gets an hour-long access token that refreshes itself. A client that cannot open a browser can use a token you issue under Settings, “Connect an AI assistant”, with a name, the scopes it needs and an optional expiry. Revoking either stops that assistant and leaves your own sign-in alone.