Notes on setting up Gemini CLI and Gemini Code Assist VSCode extension and checking they are using Google AI Pro subscription

Last updated on 14 Mar 2026
13 March 2026

I had explored using AntiGravity some days back. One thing I noticed is that it was slower to startup than VSCode. So I wanted to have VSCode setup to use Gemini with associated extension and thus be able to benefit from my Google Pro Plan in VSCode. Also I wanted to Gemini CLI to be setup.

Gemini Code Assist VSCode extension

Extension ID: google.geminicodeassist and publisher is Google.
Took long to install and failed to install first time. Airtel conn. quota got probably exhausted.
Switched to Jio. Trying again. Started at around 20:08 . Now it is 20:11 ... It got done a little before 20:24! Speed seemed to be 4 to 6 Mbps typically.

In Gemini Code Assist (GCA), I signed in with my rG Google a/c.

It showed a message about not being able to access pro-snowfall-xhnws (which turned out to be some Google Cloud project). I ignored it.

It does not show me Google 3.1 Pro model (it seems to be defaulting to Google 2.5 Flash). [IIRC, it did show me Google 3 Pro preview but I expected to see Google 3.1 Pro model.]

--------

Me (to Gemini chat panel in GCA): I am enrolled in a Google AI Pro plan. So I have access to Google 3.1 Pro model. I don't see it in the model selection list. Are you using my Google AI Pro plan?

GCA: As Gemini Code Assist, I operate differently than the consumer-facing Gemini versions you might access through a web browser. Here is how the subscriptions and models typically work regarding your question:

Distinct Subscriptions: The "Google AI Pro plan" (often referring to the Google One AI Premium plan or Gemini Advanced) is a consumer subscription. It grants access to advanced models (like Gemini Ultra or the latest Pro versions) via the gemini.google.com web interface and Google Workspace apps (Docs, Gmail, etc.).

Gemini Code Assist, which runs inside your IDE (like VS Code or IntelliJ), is generally part of the Gemini for Google Cloud portfolio. It is an enterprise/developer-focused tool. Access to it is typically governed by a Google Cloud Project configuration or a specific Gemini Code Assist license, not your personal consumer Google One subscription.

Model Selection: Because Gemini Code Assist is optimized specifically for software engineering tasks (code completion, generation, and explanation), it uses specific models tuned for coding (typically variants of Gemini 1.5 Pro). The dropdown list in your IDE—if present—shows the models enabled for the Code Assist service, which often differs from the general-purpose models available in the consumer chat interface.

Regarding "Google 3.1 Pro": If you are seeing specific version numbers like "3.1 Pro" in your consumer account, that is specific to the rapid release cycle of the consumer platform. The IDE extension prioritizes stable, code-optimized models and may not reflect the exact same version numbering or "experimental" models available on the web.

In summary: I am likely not using your consumer "Google AI Pro" plan. I am using the entitlement associated with your Google Cloud credentials or IDE login specifically for Code Assist
-----
Gemini web chat: Gemini Code Assist in VS Code

The above message led me to believe that it was not using my Google AI Pro subscription. I discussed it with Gemini web chat (WGem). WGem said GCA was having a "stale logic" issue. WGem specifically said, "As of late 2025, Google officially linked the Google One AI Premium (Pro) plan to the Gemini Code Assist extension to give individuals higher quotas and access to the "Pro" models without needing a complex Google Cloud setup."

WGem made some settings suggestions to fix the issue. I wanted the docs pages for it.

As per web Gemini (WGem), ref. pages for this issue.
I did not want to get into these settings changes at this stage and so did not make any such changes.

Gemini CLI

I decided to check if I had installed Gemini CLI.
CurrProj> gemini -v
gemini : The term 'gemini' is not recognized as the name of a cmdlet, function, script file, or operable program.
...

CurrProj> npm list -g @google/gemini-cli --depth=0
C:\Users\{Username}\AppData\Roaming\npm
`-- (empty)

CurrProj>
----

So it was not installed. Installed it.

CurrProj> npm install -g @google/gemini-cli
npm warn deprecated prebuild-install@7.1.3: No longer maintained. Please contact the author of the relevant native addon; alternatives are available.
npm warn deprecated node-domexception@1.0.0: Use your platform's native DOMException instead
npm warn deprecated glob@10.5.0: Old versions of glob are not supported, and contain widely publicized security vulnerabilities, which have been fixed in the current version. Please update. Support for old versions may be purchased (at exorbitant rates) by contacting i@izs.me

added 618 packages in 4m

173 packages are looking for funding
  run `npm fund` for details
CurrProj>

It took some time but not as long as the VSCode extension.
BitMeter Statistics graph shows it took around quarter of the download time for VSCode extension, which makes it around 4 minutes (Extension took 16 minutes to finish download & installation).

CurrProj> gemini -v
0.33.1
CurrProj>

cd'd to ssw-astro project
Ran gemini
Logged in with rG Google user credentials
Restarted gemini (it prompted me to do so).

Running: /status
[It is taking some time.]
Eventually it said: The Gemini CLI is currently at version 0.33.1, utilizing Gemini 3 as its default model engine. Key features include Plan Mode (/plan) for complex
  orchestration, an Experimental Browser Agent, and support for Gemini 3.1 Pro Preview via intelligent routing.
--------

 > /model
...
Select Model
 ● 1. Auto (Gemini 3)
      Let Gemini CLI decide the best model for the task: gemini-3.1-pro, gemini-3-flash                             
    2. Auto (Gemini 2.5)
      Let Gemini CLI decide the best model for the task: gemini-2.5-pro, gemini-2.5-flash
    3. Manual
      Manually select a model
...
------

 > /stats model
...
Auto (Gemini 3) Stats For Nerds  
Auth Method:  Logged in with Google (--snipped--)
Tier: Gemini Code Assist in Google One AI Pro 
---snipped---

The tier says: Gemini Code Assist in Google One AI Pro 
That confirms that CLI is recognizing my Google One AI Pro  plan. Am I right?

WGem agreed.

Note that I did not change any settings so far.

/quit - quits CLI

Back to Gemini Code Assist VSCode extension

Reopened VSCode
GCA icon in status bar shows me logged in with my Google a/c but does not show manage subscription
GCA left sidebar's top ... menu shows 'Manage Subscription'. Chose it which opened web page of one.google.com - 'Manage your Google One membership'. Change membership plan button on it -> showed current status: Google AI Pro.
So looks like even GCA extn. has linked to my Google AI Pro plan.

Under models it does show 3 Flash Preview and 3 Pro Preview (IIRC, it showed them earlier on too).
The confusion is why it gave me that response about GCA subscription being different when I asked it about Google 3.1 Pro model not being shown.

-----
In GCA:
/stats model

It gave a different than expected response. But it also said:
Responding with 3 Pro Preview

Initial part of its response: 
It looks like you are looking for code related to a statistical model. In the Python software engineering ecosystem, the term "stats model" almost always refers to the statsmodels library, which is the standard for estimation, statistical tests, and data exploration.
...

WGem: ... In the Gemini CLI, commands like /stats model are intercepted by the application logic to show your quota. In the VS Code Chat Panel, the chat bot doesn't have that same "internal command" mapping. Instead, it treated your input as a standard prompt and—ironically—tried to be helpful by teaching you about the Python statsmodels library! ...

WGem suggested some GEMINI.md settings.

Me: I will get into project level personalization of the GCA later on. For now, I wanted to get it setup right (and using my Google AI Pro plan). I think that's done. Am I right?
[So I did not make any settings change for VSCode GCA extension too.]

Gem: Yes, you are 100% correct. You’ve successfully navigated the maze of modern Google licensing and authentication.

Why you are "Done"

  • Billing/Plan Verification: Your CLI /stats model command explicitly returned "Tier: Gemini Code Assist in Google One AI Pro". This is the highest level of confirmation you can get. It means the backend is correctly recognizing your personal Pro subscription.
  • Model Access: Your VS Code extension already shows 3 Pro Preview and 3 Flash Preview in the dropdown. This confirms that the "Individual Pro" handshake worked, giving you access to the Gemini 3.1 Pro models that are normally reserved for paid users.
  • Authentication Sync: Since you used the same account for both the CLI and VS Code, they are sharing the same authentication tokens. Your terminal "knows" you're Pro, and your IDE "knows" you're Pro.

Why the confusion happened (The "Why")

The chat bot gave you a confusing answer because its "system prompt" (the set of instructions it's born with) is still largely based on the old Enterprise/Google Cloud model. It basically has a "script" that says: "If they ask about Pro, tell them it's for consumers only." It didn't realize that Google updated the rules to allow your specific plan to work in the IDE.

The bot's words were wrong, but the software's settings are right.

What to watch for moving forward

Since you are on a 4G connection (4-6 Mbps):

  • Keep an eye on "Auto" routing: If the response feels slow, manually switch to 3 Flash Preview for small questions. Use 3 Pro Preview only when you need it to solve a complex bug or write a whole function.
  • Ignore the "Project" errors: If you ever see a pop-up about "pro-snowfall-xxxx" project errors again, you can continue to ignore them. Those are just "ghost" projects Google Cloud uses to route your individual traffic.

You’re all set to start coding in ssw-astro. ... 

============================
14 Mar 2026

Me: I have opened ssw-astro project in VS Code. The GCA (Gemini Code Assist) extension is active. Now CoPilot is also active. How do I ensure that for code completion and chat in VS Code, GCA comes into play and not CoPilot (btw code completion quota for CoPilot has been exhausted)?

[I initially explored disabling only the code completion and chat of GitHub CoPilot but retaining some other features, as per WGem, like commit message suggestion. But that did not work out.]

Me: Copilot icon in the bottom right status bar. does not show 'Enable Completions'
But a settings button in it takes me to: @id:github.copilot.enable @id:github.copilot.nextEditSuggestions.enabled
I can change that from true to false. Should I do that?

[WGem said yes and so I did that.]

Me: Now I see that these settings are mapped to "Inline Suggestions" All Files, and Next Edit suggestions. both of which are shown on mouse click on CoPilot in Status bar. OK I have turned both off.
...

Me: But how do I turn off CoPilot chat
...

Me: I will go for disabling the extension. When I try that, I get this message:
Cannot disable 'GitHub Copilot Chat' extension alone. 'GitHub Copilot modernization' and
'GitHub Copilot modernization - upgrade for Java' extensions depend on this. Do you want
to disable all these extensions?
...

Me: I chose 'Disable All'. Now the GitHub icon in the status bar itself has gone grey with a diagonal strike through on it. Clicking the icon, shows an 'enable ai features' button. About sparkle icon, I will check next time I do a commit using VS Code UI.

Now I have to test GCA code completion and chat from VS Code.

WGem: That "strike-through" on the Copilot icon is exactly what we wanted to see—it confirms Copilot is now in a "passive" state where it won't interfere with your typing, but its background services are still there if you need to manually trigger that commit helper later. ...

[WGem suggested a chat prompt which I used.]

Me: @workspace How is the layout handled in this Astro project?
For the above (prompt), first I got this response (in GCA chat):
Gemini 2.5 generated this response due to high traffic on Gemini 3

[GCA chat does not seem to have a copy button for its response like Gemini web chat. So I copy-pasted its response to PrettyHTML.com left pane, picked up the HTML from its right pane, copied it to VS Code temp HTML file, edited it to snip out some stuff and then copy-pasted it below (using Edit HTML mode of Blogger).]

GCA chat:

Based on the files provided, here is an overview of how the layout is handled in this Astro project:

The project uses a component-based approach for its layout, integrating React components for dynamic and interactive UI elements within an Astro framework.

Key Components of the Layout System:

  1. Component-Driven UI: Instead of a single monolithic layout file, the UI is broken down into smaller, reusable components. The files you've provided show a distinction between layout-specific components (src/components/layout/) and general UI components (src/components/ui/).

  2. Mobile-First Responsive Navigation --snipped--

    • It provides a slide-in menu for mobile screen sizes. --snipped--
    • --snipped--
    • Icons --snipped-- are used for the menu open/close buttons.
    • The menu automatically closes if the window is resized to be larger than a mobile viewport, which is handled in --snipped--
  3. --snipped--
  4. --snipped--

    In summary, the project's layout is a flexible and modern setup that combines Astro's static site generation capabilities with the interactivity of React. It emphasizes a component-based architecture, handles responsive design gracefully with a dedicated mobile menu, and robustly manages URL paths for different deployment targets.

    [I shared the above (unsnipped version) with WGem.]

    WGem: That response confirms that your @workspace request worked perfectly! Even though you saw a "high traffic" warning, you've successfully forced the AI to index and analyze your specific codebase.

    1. Decoding the "High Traffic" Message

    The message "Gemini 2.5 generated this response due to high traffic on Gemini 3" is a temporary fallback mechanism.

    • The Good News: Because you have the Google AI Pro plan, the system tried to give you Gemini 3 first.
    • The Reality: When Google's "Preview" servers for Gemini 3.1 Pro are at capacity (very common during peak hours in India/Asia), it automatically drops back to Gemini 2.5 Pro.
    • The Quality: As you can see from the detailed breakdown of your MobileMenu.tsx and withBase function, Gemini 2.5 Pro is still a massive upgrade over the free models. It clearly "read" your files to give that answer.

    2. Analysis of the GCA Response

    The response you received is highly accurate for a modern Astro/React project:

    • Tailwind & Headless UI: It correctly identified that you are using @headlessui/react and md:hidden (Tailwind) for responsive design.
    • Astro-Specific Logic: It spotted your withBase utility. In Astro, handling the base path is one of the most common "gotchas" when deploying to GitHub Pages, and Gemini correctly highlighted this as a "crucial" part of your layout logic.
    • Component-Based: It recognized your separation of concerns between src/components/layout/ and src/components/ui/.

    3. How to avoid the "Fallback" to 2.5

    If you want to ensure you are strictly using Gemini 3.1 Pro (the "Thinking" model), you can try these two things next time:

    • Wait and Retry: If you see the "high traffic" message, wait 30 seconds and ask a follow-up. Often the capacity clears immediately.
    • Force the Model: In the chat dropdown, instead of "Auto," specifically select "3 Pro Preview". This tells the extension: "I am a Pro user; I'd rather wait a few extra seconds for the 3.1 engine than fall back to 2.5."

    4. Why no "Copy" button?

    This is a known quirk in the current GCA version for VS Code.

    • Workaround: Hover over the code blocks (like the <a href={withBase... part). You should see a small "Copy" or "Insert at Cursor" icon appear in the top-right corner of just that code block. For the text, a standard mouse-drag highlight is currently the only way.

    Your Work Log Summary:

    • Verification: @workspace command is active and successfully scanning your project.
    • Plan Status: Confirmed as Pro (hence the attempt to use Gemini 3 and the high-quality 2.5 fallback).
    • Project State: Gemini now "knows" your ssw-astro layout uses React components for mobile menus and handles base-path routing.
    ...
    Me: Free tier users would probably be getting limited amount of Gemini 2.5 Pro 
    Or do they simply not get that model?

    Gem: In the world of Gemini Code Assist (GCA), the Free Tier is actually quite generous, but it is architecturally "fenced off" from the newest Pro engines like 3.1 Pro.

    Here is how the models are distributed between you and a free user as of March 2026:

    1. Do Free Users get Gemini 2.5 Pro?

    Yes, but it's their "ceiling."

    • Free Tier: They primarily use Gemini 2.5 Flash for speed, with Gemini 2.5 Pro acting as their high-end model for complex tasks.
    • Your Pro Plan: For you, Gemini 2.5 Pro is actually your fallback model. Your primary engine is Gemini 3.1 Pro.

    2. The Model "Tiers" Comparison

    Feature Free Tier (Individuals) Your Google AI Pro Plan
    Primary Model Gemini 2.5 Flash Gemini 3.1 Pro
    Secondary/Fallback Gemini 2.5 Pro (Limited)   Gemini 2.5 Pro (Unlimited High-Speed)
    Daily Request Limit   1,000 requests / day 1,500 requests / day
    Context Window ~128k - 200k tokens 1 Million+ tokens
    Agent Mode Basic Advanced (can edit multiple files)

    3. Why the Free Tier doesn't get Gemini 3.1 Pro

    Google reserves the Gemini 3.x series for paid subscribers because those models require significantly more "compute" (GPU power) to run.

    • A free user will almost never see the "3 Pro Preview" option in their dropdown; it simply won't appear as an available choice.
    • If they try to use @workspace (the command you used), their index is much smaller. While you can index your entire ssw-astro project, a free user might only be able to "see" a few files at a time due to their lower token limit.

    4. The "High Traffic" Irony

    When you saw that "high traffic" message earlier, the system essentially said: "I know you paid for the 3.1 Pro engine, but the 3.1 servers are full. I'm moving you to the front of the line for the 2.5 Pro engine instead." A free user during that same high-traffic period would likely have been dropped all the way down to Flash-Lite, which is much faster but significantly less "intelligent" when explaining complex React layouts.

    Summary for your work log: You are essentially paying for priority access to the "3.x" brain and a 10x larger memory (context window). This is why it was able to explain your withBase utility and mobile menu state so clearly—it had enough "room" in its head to read all those files at once.
    ...

    Me: OK. That confirms that Gemini chat is operational now.
    Now give me a separate test with a brand new file - a .ts file - where we test code completion.
    BTW click on Gemini icon on status bar opens a drop down at top of VS Code saying Automatic Code Completion with status as On

    [Some exchanges. Then as per WGem suggestion, I created a src/lib/school-utils.ts file and in that file I entered this comment line:
    // Type definition for a School Event (e.g., Sports Day, Annual Function)
    ---
    The test was that after the above paste, on the new line, I should see a suggestion from GCA.]

    Me: Yes. It gave this code suggestion:
    export type SchoolEvent
    ----
    Good. That now confirms both Gemini code suggestion as well as chat are in play and that CoPilot is not interfering with it.

    WGem: Excellent. That confirms the "handover" is complete. You now have a clean, high-performance environment where your Google AI Pro plan is the primary driver, and Copilot is safely "muted" to prevent interference and bandwidth waste.

    Why this is a win for your workflow:

    • Precision: By seeing export type SchoolEvent correctly (and not a blog interface), you've confirmed that Gemini's Project Awareness is successfully scanning your ssw-astro folders.
    • Bandwidth Efficiency: On your Jio 4G (4-6 Mbps), having only one AI "talking" to the cloud makes a noticeable difference in latency.
    • Intelligence: Even if it falls back to 2.5 Pro during high traffic, you are still getting a model that understands modern Astro/TypeScript patterns far better than a standard free model.

    A Final Tip for your "Static School Website"

    Since this is a static project, you likely rely heavily on the Astro content layer or local JSON/Markdown files for the school data (staff lists, event calendars, etc.).

    If you want to test the true power of the 3.1 Pro engine's reasoning, try highlighting a complex Astro component and using the Chat to ask:

    "Rewrite this component to use Astro's <Picture /> component for better image optimization, ensuring the school's logo is correctly sized for mobile."

    ...

    [I did not want to go that far in my test, as of now.]

    Comments

    Post a Comment