§ 01 · The Hook

The bait, then the rug-pull.

Three differentiators in one breath — open source, one-million-token context, and a thousand free queries a day — before the title card appears. David Ondrej's cold open on the Gemini CLI GitHub README sets the frame fast: this is a structural threat to Claude Code, even if the polish isn't there yet.

§ · Stated Promise

What the video promised.

stated at 00:18“I'll show you how to set it up, how it compares to Claude Code, and how to get the most out of it.”delivered at 11:30

§ · Chapters

Where the time goes.

00:00 – 00:42

01 · Cold open — three advantages in 12 seconds

Gemini CLI README with animated >GEMINI logo. Open source + 1M context + 1K free queries/day stated immediately.

00:42 – 01:45

02 · Install + auth options

npm install -g @google/gemini-cli. Node.js 18+ prerequisite. Three auth paths: Google login (free, data training), API key (paid, no training, always Pro), Vertex AI.

01:45 – 02:45

03 · Free tier graphic + context comparison

Full-screen card: 60 req/min, 1K req/day, Open Source, Free. Model comparison chart showing Gemini 2.5 Pro context is 5× larger than Claude Opus/Sonnet.

02:45 – 04:50

04 · Running Gemini in Cursor

Opens Gemini CLI inside Cursor terminal. Theme selection, /auth flow. Running on real Vectal production codebase (React/TypeScript/Supabase).

04:50 – 06:10

05 · GEMINI.md vs settings.json + API rate limit

Default GEMINI.md is vague 20 lines. Hits 429 rate limit errors with API key mid-demo. Switches back to Google login. Navigates docs to find settings.json configuration trick.

06:10 – 07:15

06 · AGENTS.md redirect trick

Create .gemini/settings.json with contextFileName: AGENTS.md. Lets you reuse a single highly-optimized prompt file across Gemini, Codex, and Claude Code. Shows his 400-line AGENTS.md now powering Gemini CLI.

07:15 – 08:51

07 · Live coding task — Gemini fails

Prompt: replace emoji thumbs buttons with Lucide React icons in ThumbsFeedback.tsx. Gemini reads file twice, gets downgraded to 2.5 Flash, request cancelled, no changes applied.

08:51 – 09:52

08 · Vectal sponsor segment

Demo of vectal.ai AI task manager — project-level system prompts, team context, Gemini 2.5 Pro inside the app.

09:52 – 11:18

09 · Claude Code one-shots the same prompt

Same prompt, Claude Code. Reads file, applies Lucide React icons with correct green/red coloring, done in one shot. Direct side-by-side contrast.

11:18 – 12:28

10 · Verdict — today: Claude Code. Tomorrow: watch Gemini

Claude Code clearly better now. Gemini CLI's open-source fork-ability, free tier, 1M context are structural advantages worth tracking. Less than 24 hours old.

12:28 – 13:58

11 · Gemini CLI tools demo + web search

Lists all available tools: ReadFolder, WriteFile, Shell, Memory, GoogleSearch, WebFetch. Runs web search — Gemini searches for news about itself using GoogleSearch tool.

13:58 – 16:18

12 · Extended Vectal team pitch + part-two CTA

Team plan demo with per-person AI context. CTA: comment 'part two' for advanced video on MCPs, memory, web tools.

§ · Storyboard

Visual structure at a glance.

hookopen — GitHub README00:00

promisefree tier card01:45

valuerunning in Cursor02:45

valueAGENTS.md trick06:10

ctaVectal sponsor08:51

valueClaude one-shot09:52

ctapart-two CTA13:58

§ · Frameworks

Named ideas worth stealing.

02:30model

Auth Tier Matrix

Google login (free, data training on)
API key (paid, data training off, always Pro)
Vertex AI (enterprise)

Which Gemini CLI auth to use depends on data privacy needs and budget — free users risk getting downgraded to 2.5 Flash at peak times.

Steal forAny tool comparison video that has a free vs paid tier decision

06:10concept

AGENTS.md Redirect Trick

Create .gemini/settings.json with contextFileName pointing to your existing AGENTS.md. One optimized prompt file works across Gemini CLI, Codex, and Claude Code.

Steal forImmediate workflow improvement for anyone using multiple coding agents

01:18concept

Context Window as Codebase-Fit Signal

Gemini 2.5 Pro's 1M context vs Claude's 200K means Gemini is structurally better for large monorepos, even if current quality lags.

Steal forFraming AI tool comparisons — structural advantages vs execution quality are separate axes

§ · Quotables

Lines you could clip.

02:30

“1,000 queries per day. Most people will not even get close to 1,000 prompts a day, which means Gemini CLI, for most of you, will be completely free.”

Concrete number + permission to not worry — eliminates the cost objection in 10 seconds→ TikTok hook

11:10

“Google open-sourcing Gemini CLI — whereas Claude Code and Codex are not fully open-sourced — is a huge power move.”

Framing that positions Gemini as a platform play, not just a product→ IG reel cold open

10:20

“Cloud Code was able to one-shot it. No problem.”

Four words. No setup needed. The entire comparison in a sentence.→ TikTok hook

11:30

“It's been less than twenty-four hours since they released it. So for now, Claude Code is clearly better, but I would keep testing Gemini CLI.”

Honest hedge that keeps the door open — drives part-two engagement naturally→ newsletter pull-quote

§ · Pacing

How they spent the runtime.

Hook length42s

Info densityhigh

Filler8%

Things they pointed at.

00:10linkGemini CLI GitHub ↗

06:10linkGemini CLI docs — configuration.md ↗

07:15productNew Society (Skool community) ↗

08:51productVectal.ai ↗

01:18toolAI model comparison tool (artificialanalysis.ai)

§ · CTA Breakdown

How they asked for the click.

15:50next-video

“If you want me to make another video on it, comment below 'part two' and I'll make it happen.”

Soft engagement CTA seeded earlier at 11:18 and closed out at end. Works because the first video intentionally leaves MCPs, memory, and web tools uncovered.

§ · The Script

Word for word.

HOOKopening / re-engagementCTAthe pitchmetaphoranalogystory

00:00HOOKSo Google just released Gemini CLI, their own autonomous coding agent. It's completely open source. It has 1,000,000 contacts window, and you get 1,000 queries per day for free. This is a serious competitor to both Codex and Clothcode. And in this video, I'll show you how to set it up, how it compares to Clothcode, and how to get the most out of it. So stay tuned. Alright. So just type in Gemini CLI into Google and click on the first GitHub link. This will take you to the official GitHub page from Google. So let's scroll down, and I'll show you how to set up step by step because it's actually very simple. Right? You don't have to be afraid if you don't know how to use the terminal or if you're not a programmer. First step, what we have to do is n p m install dash g at Google slash gemini CLI. So just copy this command, open any terminal, and simply paste it in and hit enter. This will start installing Gemini on your system. Now as they say, the only prerequisite is that you have Node. Js version or above installed. So here is the link. Again, I'm gonna link the GitHub repo below the video. But if you don't have Node. Js on your system, which you definitely should have, just click on this link and download Node. Js, again, version 18 or higher for your system. Once you have Gemini installed with this command, it runs for a few seconds. So then if you type Gemini into your terminal, boom. There it is. Gemini is launched. This command as well, you should probably run it, uh, every morning because that's also how you keep Gemini CLI updated. But now let's move on to cursor, and we're gonna be using it in an actual production ready code base. Gemini 2.5 Pro,

01:28which is what Gemini CLI is built upon, the best model from Google, has a five times higher context window than Cloth for Opus and Cloth for Sonnet. And these are the only two models you can use in cloth code. So if you have a large code base, this might be the best coding agent for you. Anyways, once you're in cursor, just press command j to open the terminal or control j if you're Windows. Type in Gemini. Boom. Just like that, it will open. And, again, if you haven't authenticated yet, you'll have to choose a theme, which you can also change anytime

01:59with, um, slash theme. You here you have a lot of options, actually. To be honest, I like the default light. So I'm gonna select that. And, also, it will ask you to authenticate. So if you do slash off, this is what Gemini will look like when you first run it. Right? You have three different options. Log in with Google, Gemini API key, and Vertex AI. The simplest one is obviously log in with Google. However,

02:19there is different things you should consider. Now as I said before, you get literally 1,000 queries per day for free. Thousand queries per day. Most people will not hit that. Let's be honest. Most people will not even get close to 1,000 prompts a day, which means Gemini CLI, for most of you, will be completely free. However, there are a few things to consider. With the login with Google, you get a thousand messages for free, but Google will train on your data. If you use your own API key basically, if you pay, Google will not use your data.

02:51So, again, if you use a big code base that is on a business that you do not own and you cannot share the data, then you should probably be using API key. Plus, you get the added benefit of not being randomly downgraded. If you log in with Google and you're using the 1,000 free messages per day, a lot of the times, especially during the peak periods of usage, you get downgraded to Gemini 2.5 Flash. So,

03:12hopefully, that doesn't happen. But if you have your own startup or you just wanna vibe code something from scratch, log in with Google is the simplest way. If you wanna make sure you're using the best model, Gemini 2.5 Pro. And if you don't want Google to train on your data, make sure to set up your API key in Google AI Studio. Make sure you fill out your billing details. That way, you're actually paying. And by the way, Gemini models are some of the most cost effective AI models

03:35in the world. So I would recommend going with the Gemini API key unless you really want to save money. In that case, just log in with Google. Anyways, let me first close all the other cloth codes here because I use a lot of cloth codes. K? And then, also, you can just rename it to Gemini CLI and pin the terminal. That way, you don't accidentally close it. So this is how your workspace honestly should look like inside of Cursor. One autonomous agent here and then maybe the file here, or if you wanna be try hard like me, a second autonomous agent on the right. So, literally, most of my time, I don't even have a single file open. Obviously, you can still switch to the files at any time and see the code yourself.

04:13But the beauty of these autonomous coding agents is that when they are changing any of the files, they show you the difference. So, really, all you need to do is have Gemini on one half of the screen, cloth code on the other half, and just vibe code with both of them. So since we've created a Gemini dot m d file, you can see above chat, it's saying using one Gemini MD file. So it's using these instructions, but this is this is very vague. Right? This is, like, 20 lines of prompts. This is not good at all. So what we need to do is we need to tell it locate the dot Gemini folder.

04:45And the reason for that is if we go back to the documentation, we can actually see uh-uh. If we scroll down a bit okay. So I found it. You need to go to docs, CLI,

04:55and then configuration MD. It's a bit, um, you know, hidden, but okay. Why are we getting errors here? Wait a second. Too many requests. Resource exhausted. Okay. So, uh, the bad news is that they seem to be overloaded. I don't know what's happening, but my API key is not working. Let me know, guys, if you encountered this in the comments, but I had to switch it off back to my Google login, right, which does seem to be working. So I don't know. I don't know what's wrong with the API method. Maybe they'll patch it in the next update. Who knows? But, basically, I wanted to run this prompt, locate Gemini folder.

05:31And the reason for that is if we go into the docs, into Gemini CLI doc CLI configuration, here, we'll see a very important section. Right? So I read a lot of the docs. And here, this section basically shows how to use the settings dot JSON. So let's see if it located that. Okay. We don't see a Gemini folder. Let's see settings dot JSON. Okay. Yeah. We don't have a Gemini folder, so we need to create one. So let's go back to the web,

05:58and this is what we need to create. So say, okay. Then create this file and folder. Boom. Settings dot JSON. The reason for that is it allows us to reuse

06:12some prompts. Here's the example that I tweeted yesterday evening. By the way, if you wanna get my thoughts in real time on all of these new AI updates that just come out, just follow me on Twitter at david Andre one. It's the easiest way to stay up to date with all the other tips. And, yeah, here is the tip. Okay. So when you create Gemini okay. So we need to allow this. Yes.

06:33Dot Gemini folder, and it wants to write this file. And we need to add the following, codec filename agents dot m d. Why? Because this makes Gemini CLI use your hopefully optimized agents dot m d. Again, if you're watching my videos, you will have an agents dot m d file, especially if you're in the new society. Because in the new society, in the classroom,

06:53I have an entire step by step workshop on the ultimate codecs guide, which takes you from the codex setup, GitHub fundamentals, my codex workflow, codex variations PR review, codex Internet access, and my new codex workflow with cloth code and cursor. So, yeah, this is available in the classroom inside of the new society. Make sure to join. But the whole point of this tweet is so that you don't have to write a new prompt file because it's it's getting, you know, overwhelming with so many files. So Gemini CLI created this settings dot JSON, so you can just open it settings dot JSON

07:27in the Gemini folder. Let's put it to our second half. So instead of, um, you know, going for the Gemini dot m d file, which is fine. This is better than nothing for sure. Most people will not even bother creating any system pro file, which terrible. The better method, however, is to delete this file. So let me just delete that. Boom. And

07:46put this in into your dot Gemini slash sharing JSON. Actually, make sure to remove this HTTP. I don't know how that got copied over, but the JSON needs to look like this. Context file name agents dot m d. And if we restart, let's say, we kill this, boom, clear, and type in Gemini. But you can easily start it from any terminal just by typing Gemini. There it is using one h s dot m d file. So, basically, this redirects Gemini telling it, hey. Instead of using the Gemini dot m d file, here is where we have the context. This is the context file name. That way, you can use the same file and keep improving your file, you know, whatever you're using for codex or for cloth code.

08:26CTAAnd here, like, I have 400 lines of instructions. Right? So instead of me having to craft a new Gemini dot m d file from scratch, I can use my highly optimized 400 line prompt for codecs that I use every single day, and I can instantly use that for Gemini CLI, making it way more powerful and way more useful for my own code base. So let's get to building with Gemini CLI. Of course, I'm gonna be building my AI startup, Vectal dot ai, which, by the way, if you're not using Vectal,

08:56CTAthis might be the easiest way to increase your productivity. Why? Because we have all of the best AI models, including Gemini 2.5 pro, powering our AI agents that can create tasks, help you complete tasks, do web research using Perplexity Pro, and so much more. If you're still using outdated task management tools like Todoist, ClickUp, Notion, just go to Vectular AI, Sign up. You can get started for completely for free. Take ten minutes to move your tasks over, and you will never look back. The ability to chat with the best AI models that know everything about your tasks, that know what projects you have active, which, by the way, in each project, you can set a custom system prompt, which this is not available in any other productivity or task management tool. These AI agents are aware of that, are aware of your user context, which allows them to be a lot more useful than just the default chat GBD or default cloth. So, again, go to vectel.ai,

09:49CTAcreate an account, move your tasks over, and you will never look back. So what I want to do with Gemini, I want to improve the icons right here. So, again, let's go back to chat. Let me refresh. So I'm adding these icons for the feedback of the chat. Right? That way, we can have a clear feedback loop of how good of a job the agent did. So for example, here, it did exactly what I wanted, so I would give a thumbs up. But let's say you ask create four tasks and it only creates two. Well, that would be a thumbs down. So, basically, I'm adding this feedback mechanism, but right now the buttons are a bit too visible, I would say. So I wrote a prompt. Your task is to improve the design of the thumbs up, thumbs down buttons in and then I tag the file, which, by way, the tagging of files in Gemini CLI is very fast. Instead of using these yellow emojis, which are way too visible, let's use the thumbs up icons from Lucid React. Let's see how good

10:40HOOKGemini CLI is at front end design. Okay. So we got downgraded to 2.5 Flash, which is unfortunate because, uh, I would like to be using 2.5 Pro, but I don't know why the API key method isn't working. Uh, maybe, uh, maybe it's just overloaded. I mean, they must have some crazy demands right now because they just released it. And, I mean, let me know guys if you've encountered this error with using the API key or how to solve it. I I would be more than happy to do that. So comment below if you know how to solve that because I really don't wanna be using 2.5 Flash. Okay. So 2.5 Flash failed. I mean, I'm trying to compare it to Cloth code, but it's difficult because Gemini 2.5 Flash is nowhere near good as Cloth for Opus or Cloth for Sonnet. I cannot seem to copy what? When I copy my prompt, it copies the response.

11:25HOOKOkay. Let's just do up arrow. Oh, this box appears always. So, yeah, this is one of the tools that Gemini has access to, which is the read many files tool, and it still didn't update it. I'm waiting for instructions. Let's try Clothes Code. Like, I'm I'm guaranteeing you Clothes Code will one shot this. Keep in mind, Clothes Code has been out for, like, over, what, almost two months,

11:49HOOKand the Gemini CLI released literally less than twenty four hours ago. So, yeah, I'm I'm trying to give an honest review. I'm not trying to overhyp. You know? Gemini CLI has some massive advantages, and there it is. Cloud Code was able to one shot it. No problem. Like, to be fair, 1,000,000 contacts window, as I showed earlier, is much better than 200,000.

12:08Also, fact that you get 1,000 free queries per day is huge for a lot of people. The fact that Gemini CLI is open source is also huge because you can literally fork it and build your own coding agent, maybe using different models, maybe customizing the problems, whatever. So this is a huge power move from Google, open sourcing Gemini CLI, whereas ClothCode and

12:28CTACodex are not fully open sourced. Right? However, at the end of the day, what matters is how good the agent is. And when it struggles doing a simple prompt, it's nothing crazy, guys. Like, changing the icons is a simple prompt. And when it struggles to do even that, whereas ClothCode, its biggest competitor, one shots that, no problem. Yeah. Google needs to do some polishing. Again, I still believe Gemini CLI has major potential due to the fact that I listed, but it's been less than twenty four hours since they released it. So for now, Cloth Code is clearly better, but I would keep testing Gemini CLI. And maybe, like, in a week, when it gets through, like, a bunch of updates, maybe they'll be back, uh, especially if we can get the API method working so that we can always use Gemini 2.5 pro, not Flash, because Flash is a much worse model. So, again, comment below if you've been to solve that API error. But for now, Clothcode is the clear winner. If you're going to choose which to use, use Clothcode. If you don't have any money, Gemini CLI is a great alternative. Now if you want me to make a more in-depth video because this was largely about the setup and the differences compared to other tools, there's a lot stuff to cover such as the tools within Gemini CLI. It has a lot of tools. Also, MCPs, how to connect MCPs and run it with MCP servers. Also, the memory. Right? It has an interesting way of managing memory, which is like stored notes or anything else about your project or your preferences. Also, the web search. So Gemini CLI has the ability to browse the web. So I'm gonna say browse the web

13:58CTAabout latest news about a new coding agent from Google.

14:05CTALet's see if it finds info about itself. This is a major advantage because, obviously, Google owns Google Search. Right? So they built it in into Gemini CLI so it can get documentation faster, check the web. And there is also a custom separate tool called web fetch where you can give it a website, and it gives you it basically crawls it completely. See, it's using the Google search tool right here, latest news for coding agent Google.

14:28CTASo, yeah, there's a lot more stuff to cover. I could go on for another hour hour hour and a half. If you watch my previous Clotco videos, you know that that's true. So if you want me to go even more in-depth into Gemini CLI, comment part two. If I see a lot of comments saying part two, I will make a very in-depth advanced video on Gemini CLI and post it in the next week or so. With that being said, thank you guys for watching. If you haven't already, go to vector.ai and sign up, especially if you have a team. If you are a small or medium sized business,

14:59CTAjoin the team plan. Everybody on your team will have access to the cutting edge AI models in the app. Let's be honest. It's very hard to get your employees to use AI tools like ChatGPT, Cloth, Perplexity if they're not built in into the software that you're using. This is the problem that Vector solves. Plus, as I showed you, you can have custom system prompts for each project. So let's say let's say you have a software project right here, and you say, our tech stack is this this this. The current focus of q three is blah blah blah. And you put in focus for each project to save it. And when you select that project, which, by way, you can easily favorite it, the AI agents will know what the focus of that project is, and they will adapt the responses to that for all people in that project. Plus, if you go with the Vectal team plan, you could also give a custom system prompt to every person on your team so that Vektel knows, okay. This person is a video editor. This person is a developer, and it adapts the responses to them. Moving your team to Vektel has

15:54CTAto be the easiest way to instantly boost the productivity of all the people on your team. So simply go to vektal.ai, get the team plan, move your team over, and suddenly, you became an AI first company. With that being said, thank you guys for watching. Let me know what you think of Gemini CLI. And if you want me to make another video on it, comment below part two, and I'll make it happen. With that being said, thank you guys for watching, and have a wonderful productive week. See you.

§ · For Joe

The AGENTS.md redirect is immediately stealable.

JoeFlow / Claude Code playbook

One settings.json file lets you point Gemini CLI at the same AGENTS.md you already use for Claude Code — no rewriting prompts, instant upgrade.

Create .gemini/settings.json with {"contextFileName": "AGENTS.md"} in your project root.
Your existing Claude Code or Codex AGENTS.md now powers Gemini CLI too.
For JoeFlow or any project where you're comparing agents, run both in Cursor split-screen — same codebase, same prompt, judge the diff.
The open-source angle is a product story: Gemini CLI can be forked and shipped as a custom coding agent. Worth watching for MCN+ tooling.
Use the auth framing (free/paid/enterprise tier) as a template for any tool comparison video — it positions the tradeoffs cleanly without FUD.