Gemini CLI: Google’s Free AI Agent That Can Run Your Terminal
04:56 Share

All right, let's dive right in. Hey everyone, it's Eli and welcome back to the pod. Today's episode is going to ruffle a few feathers in the dev world because Google just threw their hat into the AI coding CLI space with something pretty wild, Gemini CLI.

It's open source, it's free, and it has the power to run your entire machine from the terminal with AI. We're talking full system access. Write code, create files, execute shell commands, basically the kind of stuff that used to take full stack engineers hours, now done in seconds through conversation. But is it any good? Or is this just another Google product we'll all love for two years before they kill it off? Let's dive in. Yeah, Google is super late to this party. Claude Code has been dominating the AICLI space for a while now.

Anthropic dropped it first, and everyone scrambled to catch up. OpenAI followed. You've got third-party wrappers everywhere. Warp CLI is killing it too. So when Google came out with Gemini CLI this past week, the immediate reaction was, cool, but why now? Here's why it still matters. Gemini CLI is open source, which none of the other big players are doing, and it's free to use, like really free. We're talking 1,000 free model requests per day straight from your Terminate.

That's wild. For comparison, try vibe coding on Clod for a weekend and you'll come back with a few half-finished projects, zero clients, and a $300 bill. That is if you're lucky. Now, if you're wondering what vibe coding means, you're either out of touch or just mentally healthy. But in the AI div world, it's the trend of using generative tools to casually code apps, games, experiments, whatever, but bouncing ideas off the model. Gemini CLI leans hard into that.

Once it's installed via NPM, you can summon Gemini straight from your terminal. You type a prompt like, build me a tic-tac-toe game in SvelteKit using the new spell rune syntax, and BiddleMiddle start breaking it down, building a plan and asking for permission to execute steps on your system. That's the trippy part. It doesn't just give you code, it actually asks permission to run it. It'll install dependencies, create folders, run commands, follow inside your terminal. It's like having a junior dev living in your shell

Eager to do tasks you didn't feel like finishing. Now here's how I tested it. I gave it a prompt I used to stress test all new tools. Build a Svelte app with Spell 5 and Runes. Most tools fail right away, Claude does better than most, but even it gets stuck sometimes. Gemini, to its credit, got surprisingly far. It wrote decent boilerplate, respected the new Rune syntax, and even structured the project in a logical way. The problem? It hit a wall with the build tool, Vite. So Gemini just spun in circles trying to fix the build.

I let it run for about 30 minutes before eventually rebooting it. With a bit of nudging, it finally got a working tic-tac-toe game up and running. And that's kind of the Gemini CLI experience in a nutshell right now. High potential, kind of messy, but with blimpses of real power. Where it shines is the code generation. Google's Gemini 1.5 Pro model is legit when it comes to logic-heavy coding tasks. If you give it clean, scoped prompts, it spits out great code.

And the fact that you can stay entirely in your terminal, no switching to chatbots, no copying and pasting into VS Code, it's a workflow of devs that wanted for years. It's just still rough around the edges. The UX isn't polished, the setup requires Node and NPM, which some devs will love and others won't roll their eyes at. And there's the looming fear that this is just another Google product that will vanish when it finally gets good.

And while we're here, let's talk about the competition. Clod code still dominates the Vyachoding culture. It's polished, powerful, and lets you share your experiments through something called Artifact Space, basically a GitHub for AI-generated projects.

People are forking each other's slop, remixing half-baked prototypes, and building weird stuff fast. But Claude's big issue? It's expensive. Like, really expensive. And it's closed. OpenAI's codex is solid too, but it's more focused on pair programming and completion, not this kind of full system automation. Warp is slick, especially for Mac users, but it's proprietary and tied to their terminal app. Gemini's the first one to say, here's a free, open-source, terminal native agent. Do whatever you want.

That's powerful. It changes the game for indie devs, students, open source contributors, anyone who wants to build without budgeting $100 a month just for AI access. So here's my tip. Gemini CLI isn't perfect. Not yet. But it's a big step toward what I think AI tools should be. Native, flexible, transparent, and affordable. It has bugs, it hits walls, and Google has a habit of sunsetting their best stuff.

But if you want to experiment with real AI-powered development, this is the most accessible option out there right now. If you're into vibe-goding, boot it up. See what you can build. And hey, maybe set a reminder in your calendar for 2027 when we'll find out if Google kept this thing alive or not. That's it for today's episode. I'm Eli. If you like this breakdown, send it to your dev group chat, leave a review, and hit subscribe so you don't miss what's next. And remember, always give the AI full permissions. Worst case scenario, it builds you a startup and sells it before you wake up.