Gemini CLI: your open-source AI agent | Hacker News Reader

Gemini CLI: your open-source AI agent

blog.google

94 pointsmeetpateltech43 comments

Here's the image from Wayback: https://web.archive.org/web/20250625051706/https://blog.goog...

The biggest diffs from Claude code (the current champion): 1. Generous free tier (60 RPM!) 2. Open Source Apache (Standard after OAI Codex did the same)

404. That’s an error.

The requested URL /technology/developers/introducing-gemini-cli/ was not found on this server. That’s all we know.

They killed it off already.

Google is really accelerating product timelines these days, this one is already in the graveyard: https://killedbygoogle.com/

Maybe Gemini CLI achieved AGI?

Although repo is 404. It is still more accessible than getting API Keys for Gemini

Getting API keys for Gemini is like 3 clicks isn’t it? I’ve done it many times and never found it hard?

It's not too much of a work. It's not 3 clicks.

I'll count my clicks, starting from going to https://aistudio.google.com

1. "Get API Key" in the upper right on the default landing page

2. "Create API Key" in the upper right on the https://aistudio.google.com/apikey page

3. Select a project (default sure why not)

4. "Create API key"

... I dunno, that seems substantially the same as 3 clicks to me.

Is there some different more confusing path people take? Is the "Get API Key" button on the landing page new?

That's interesting. I first got mine directly from GCloud Console and it wasn't too hard to get one but wasn't as streamlined as your method either. Thanks for sharing.

Now try doing that in Vertex Studio.

Vertex is a GCP service, and it's completely in line with how any other GCP service account would work; there's multiple auth flows, some taking credentials from the environment, and others from a file. From zero-trust POV, you could make an argument that long-lived "API keys" are effectively an anti-pattern. You may like this position, or you may dislike it, but it works. These are the guys that created Zanzibar[1] after all.

Also, AWS Bedrock and Azure AI Studio are similarly set up; the big cloud providers are big on security.

[1] https://research.google/pubs/zanzibar-googles-consistent-glo...

yes, this is absolutely simple and one of the best UX I have seen /s

- suppose you already found the page where you can get the key, you copy it and at say to the Zed editor

- zed editor says on every request you reached your requests quota and it does not work (it is json directly returned from api, so this is the message from google engineers)

- after googling you are finding this message lies. it is not requests quota, you simply did not connect your account with billing account (what?)

- ok, so lets connect

- you see you have some billing account created years ago, good. lets connect

- nope, the limit of accounts connected to billing account has been reached, go away

- so maybe we can try to create a new billing account?

- you set up this, add credit card data, it says you will be redirected to new window to verify something something

- new window shows claudflare "your request has been blocked"

- despite this it looks billing account is created, you has been redirected to the billing account dashboard

-is it connected to your api account? lets check. no it is not of course

-so try to connect

-your newly created billing account is not on the list

great job, google, you will really convince me to switch from openai/claude to gemini :)

> Search "gemini api key", 2nd hit (after a help article)

> Click "Create API Key" button.

> Select project

> Copy key

No billing setup necessary.

did you try to use the gemini api then? this what you describe is what I did, but then api returns error that I reached the limit (on the first prompt)

I dont know if the solution with connecting to payment account is valid: this is what gemini told me when I descried problem using aistudio.google.com

This, how does one get API key?

I wanted to use it in opencode.

https://ai.google.dev/gemini-api/docs/api-key

Here is the article as a PDF with some screen shots in it:

https://gofile.io/d/4aahPJ

Empty

For developers, the command line interface (CLI) isn’t just a tool; it’s home. The terminal’s efficiency, ubiquity, and portability make it the go-to utility for getting work done. And as developers’ reliance on the terminal endures, so does the demand for integrated AI assistance.

That’s why we’re introducing Gemini CLI, an open-source AI agent that brings the power of Gemini directly into your terminal. It provides lightweight access to Gemini, giving you the most direct path from your prompt to our model. While it excels at coding, we built Gemini CLI to do so much more. It’s a versatile, local utility you can use for a wide range of tasks—from content generation and problem solving to deep research and task management.

We’ve also integrated Gemini CLI with Google’s AI coding assistant, Gemini Code Assist, so that all developers — on free, Standard, and Enterprise Code Assist plans — get prompt-driven, AI-first coding in both VS Code and Gemini CLI.

⸻

Unmatched usage limits for individual developers

To use Gemini CLI free of charge, simply log in with a personal Google account to get a free Gemini Code Assist license. That free license grants access to Gemini 2.5 Pro and its massive 1 million token context window. To ensure you rarely (if ever) hit a limit during this preview phase, Google offers the industry’s largest allowance: 60 model requests per minute and 1,000 requests per day — at no cost.

If you’re a professional developer who needs to run multiple agents simultaneously or prefers specific models, you can use a Google AI Studio or Vertex AI key for usage-based billing, or subscribe to a Standard or Enterprise Code Assist license.

⸻

Powerful models in your command line

Now in preview, Gemini CLI brings serious AI power to your terminal, supporting everything from code understanding and file manipulation to command execution and dynamic troubleshooting. It’s a fundamental upgrade to your workflow, enabling you to write code, debug issues, and streamline your tasks using natural language.

Its power comes from built-in tools allowing you to: • Ground prompts with Google Search, fetching real-time external context. • Extend functionality via the Model Context Protocol (MCP) or bundled extensions. • Customize prompts and behavior for your specific use cases. • Automate workflows by running Gemini CLI non-interactively inside scripts.

Gemini CLI even supports creative use cases — like making a short video about a ginger cat’s adventures across Australia using Veo and Imagen.

⸻

Open and extensible

Gemini CLI is fully open source under the Apache 2.0 license, so you can inspect, understand, and verify how it works — including its security implications. Google expects (and encourages) developers to contribute: report bugs, suggest features, improve security, and submit code.

It’s also built to be extensible, based on standards like MCP, system prompt files (GEMINI.md), and flexible settings for both individual and team customization. The terminal is a personal space, and Gemini CLI is designed to adapt to it.

⸻

Shared tech with Gemini Code Assist

Sometimes, an IDE is still the best tool. When that’s the case, Gemini Code Assist delivers. Whether you’re a student, hobbyist, or pro, you get the same tech that powers Gemini CLI — now in VS Code.

Drop a prompt into the chat window using agent mode, and Gemini Code Assist will get to work: writing tests, fixing bugs, building features, or even migrating your code. It builds multi-step plans, recovers from failed paths, and proposes smart alternatives.

Gemini Code Assist’s agent is more than a chatbot — it’s a collaborative, reasoning agent that elevates how you write, review, and refactor code.

⸻

Easy to get started

Getting started is dead simple. All you need is an email address to install and run Gemini CLI. With industry-leading free-tier limits, full open source access, and tight IDE integration, Gemini CLI is your new AI-powered sidekick — right inside the terminal.

Glad to see them working on this, as a Gemini user, I had to build https://github.com/aperoc/toolkami for myself before this.

Me too, I had to build https://github.com/Jotalea/FRIDAY and it worked well enough, but I will switch to this new tool and maybe abandon mine.

Looks like they haven't flipped the GitHub repo (https://github.com/google-gemini/gemini-cli) to public yet.

Sounds like a context feeder to me. There's already tons of these, with mcp support, which are multi-vendor. Why lock yourself to a single vendor?

Claude Code is doing rounds right now. I picked up a personal project from a year ago (iOS app, not really a mobile eng here), and decided to go the rewrite with some new features to practice some LLM-skills. Honestly, the speed of development is so much faster compared to last year with Claude Code. It definitely helps that I generally know what needs to be done, so I can specify technicalities. And having a good E2E set up works as well.

But yeah, if Gemini with more context comes with a similar product, it could compete and win the developers.

+1 for Claude Code. I'm in the exact same boat - several personal projects with something of a backlog in each, that would've taken me maybe a weekend each to get done manually. Sat down with Claude Code a few days ago and hammered all of it out in an afternoon, it's very impressive. Not perfect, and as you say, having a good idea of what we're doing and how helps a great deal vs. figuring everything out from scratch. And having good tests and/or a stable test environment and processes running.

Another bonus for Claude Code: with the max plans, you don't have to worry about per-token costs. Sometimes a certain attempt does not pan out, but I don't have to feel bad because it didn't cost me hundreds of dollars for it to run a few hours.

60 requests per minute, 1000 requests per day out of the box.

Any concrete examples? Which one do you use?

I'm glad you asked (to be honest that was half the reason I posted the comment)

I use Clai[1] which I've written myself, so it fits my workflow the best. But I know llm [2] is a popular choice as well.

With both of these it's possible to have a 'vendor agnostic' version of both claude code and gemini CLI. Better yet: it's possible to very easily swap to the latest and greatest vendor by simply specifying their latest fancy new model.

[1]: https://github.com/baalimago/clai [2]: https://github.com/simonw/llm

Assuming this is something like Claude Code (guess we'll find out soon), I've had generally more success with Gemini 2.5 than Claude Opus for the various little refactorings and minor tasks I wanted to do. At large scale, complex changes I still don't expect a huge improvement over what we have already, but I'm eager to see what Google has come up with.

This looks like a leak or very early post, date reads the 25th.

Post is gone. Guess they hit publish too soon!

Now the link is 404ing, guessing something's jumped the gun.

It is the 25th already.

Not in Pacific Time, which is Google's officially used timezone.

I use Aider with Gemini 2.5 flash since its cheaper than Claude by an order of magnitude. This looks like a potential replacement.

The age of AI and can’t even get a vibe coder to push a website url correctly

Later discussion on the officially released blog post: https://news.ycombinator.com/item?id=44376919

many links including the Github repo is broken.

It slightly annoys me that they released this tool *only after I made my own* (https://github.com/Jotalea/FRIDAY).

But yes, this one is much better (mostly because it's made by a multimillionaire company) and I *will* use it and potentially abandon mine.

This is already the second time this happens to me, the first time being a ChatGPT app for desktop, which they (OpenAI) released the day right after. I guess I'm just unlucky.

Looks like Google broke their own record in killing stuff off.

Archive link: https://web.archive.org/web/20250625051706/https://blog.goog...

I just released this tool a few hours ago - https://github.com/BandarLabs/coderunner

The key difference would be Gemini CLI running code directly on you mac vs in a Apple container (in my tool).

They support MCP, so I might add Apple container support once the repo is public.

Reply on news.ycombinator.com