Service update: mitigating abuse and prioritizing traffic

[ryanjsalva]

Hello, friends. 👋

We are making adjustments to the service backing Gemini CLI, and want to be upfront about what is changing.

🌟 Key changes

• Abuse Detection: We’re implementing more robust detection for policy-violating use cases (e.g. using Gemini CLI oAuth with third-party software). See the terms of service for details. If you think your account was flagged in error, please reach out to our support team by filing a ticket at console.cloud.google.com so we can get it sorted for you.
• Prioritizing traffic: Starting March 25, 2026, we’re changing the way Gemini CLI routes traffic with higher priority given to accounts based on license type and account standing. As a result, some customers may encounter capacity-related limitations during periods of high traffic.
• Model restrictions for free tier users: Starting March 25, 2026, Gemini Pro models will only be accessible via paid subscriptions. Free tier users will be limited to Gemini Flash models. Upgrade to a Pro or Ultra plan to access Gemini Pro models.

As a reminder, to get direct control over your own quotas and billing, use Gemini CLI with your own paid API key from AI Studio or Vertex AI.

We understand these changes may require some planning. If you have questions or concerns, please voice them. We’ll do our best to answer, and always remain transparent with this community. The whole product and engineering team is here to support you during this process. As always, thanks for building with us. 🙏

[PhilosophiMoonbeam]

these changes may require some planning = Switch to Codex btw 🤷🏻‍♂️

[lab-key]

Appreciate the transparency—changes like these are understandable, but they quietly reshape who gets to build and experiment. Hoping access remains inclusive, because innovation often starts on the free tier.

Agreed.

I made most things just using the free tier... I'm glad I mostly finished the project that was, I think, important.

I figured out Self-Driving before Elon Musk :D

[ProdigyDSP]

Google about to shoot themselves in the foot again. Guess it's time to look for other solutions.

[xiaoCZX]

好吧，谷歌的各位，我能理解那些想用命令行界面（CLI）运行第三方软件的人。如果他们知道 CRUSH 和 OpenCode 的存在，肯定不会用谷歌的 CLI。但我的抱怨是，即使是专业版用户，他们的 CLI 也烂到家了。3.1 专业版和 2.5 版都严重过时了，现在完全没用。你们明白吗？而且，代币数量减少这么多，肯定会流失客户的，这很不正常。你们明白吗？有时候，这简直太愚蠢了。你们得好好研究一下那些该死的模型，谁他妈的给Gemini Pro做个预览？我以前用2.5的时候还挺好的，现在简直就是垃圾，你们懂吗？三四天我根本没法用CLI，这不正常，你们这群白痴，居然两周都这样！你们要是再这样下去几天，你们的一切努力都将付诸东流！我要换成Codex了，Codex超棒的，GPT烂透了，但是Codex配上Harrs，用最新的5-4版本或者5-4 Codex，才是最好的。Claude还用那种他妈的掠夺性手段来骗人，告诉你们可以用免费的模型和免费的CLI，用开源软件，用Crush，用Aether，现在有很多，很多中国模型，它们都他妈的很棒。如果这些混蛋继续这样下去，我们都会转用开源软件，我已经说过了。

话糙理不糙

[haryn]

Here is a clever workaround for Plus/Free accounts getting discriminated: https://haryn.hashnode.dev/the-gemini-cli-paradox-route-to-your-own-endpoints-a-digital-thriller

[autom8edIT]

Thanks for moving to GO, node.jshit is ass, it uses almost 40GB when you have a serious workload running on your GPU with a local model.

Then there's no need to use your enterprise trial and send request straight from pwsh and hopefully the experience will be even better not just performance wise thanks to GO, if you make the connection to AG IDE proper I see crazy potential in that!

[YurySolovyov]

Is opencode via https://github.com/jenslys/opencode-gemini-auth usage safe? (Pro account)

[mitchell-wallace]

@ThomasRedstone that doesn't even make sense. opencode still hits google's endpoints, how is the "learning" different from hitting the same endpoint with gemini cli?

They get much more information when it's from their own tool. For example, they can A/B test system prompts, test different versions of their edit tool to see which is most consistent, pull more information about the environment to better understand where and why their agent crashes so much, ...

Not to mention, if you run out of quota in Opencode, you look for the most cost-effective way to get more usage. That might be Codex or Kimi or Chutes or whatever. If you run out in a Google product, the happy path is to just give Google more money.

And there's customer loyalty too - they want you in the ecosystem.

These factors means Google gets more value when you use their tool and can better justify running inference at a loss.

Obviously Opencode is better. Google isn't saying you can't use Opencode. They just don't want to lose too much money on Opencode users.

Why doesn't Codex care? Because sama knows how much y'all love Opencode and the developer sentiment is worth it to them.

[YurySolovyov]

@mitchell-wallace if the argument applies to sama, why doesn't it apply to google? google has no interest in developers reputation/karma?

[Sujaicodes]

It may work and feel convenient, but tools like that rely on unofficial authentication flows, which brings a few risks like security and policy issues.Its okay for learning/testing, but avoid for production or anything sensitive.

[Integer-Conversion-Error]

@YurySolovyov google is big enough not to care. look at every single release they've made for the past two years. practically every single one (with a few notable exceptions) was half-broken, fragile, unsupportable headline-grabbers. it seems google has a bad case of split-brain, one side (the research they do, e.g. attention is all you need, turboquant, imagen, the inference TPUs, etc.) is slower but more impactful overall, while another side is moving fast and breaking things and getting little progress made on the way, (think the dumpster-fire that is Antigravity, the shallow feature-set of NotebookLM, the Gemini chat interface breaking down after 50k conversation length, etc.). At the end of the day, Google is a trillion-dollar company, and they aren't burning other people's investments like Sama is, they are already a huge company well before. They have the ecosystem, the integrations, the existing partnerships, and their own existing user-base, all of which they can choose to gradually burn if necessary to beat openAI/anthropic. not by necessarily being better, but just being "fatter".

[autom8edIT]

Let's say it doesn't take much effort to be able to become a web dev. compared to one of their true geniuses that created Gemma which is fucking sick, especially if you have a 4090 or 5090 or Ada card with 24GB+ VRAM.

It's actually almost a genius even on a 4080.

I could create web services with java scripts in Notepad as a kid. Could I create a super sick AI model like Gemma? Hell no!

Also since AI came along, no wonder it was reported that 75% of Google's code was made by AI. Their only code that can't be replaced by their Gemma model is basically just their AI models and the backend infrastructure.

Rest of their services are web code, Android is just a Java VM with a Linux bootloader that Gemma can code as good as it possibly gets.
Trust me when I say that you'll see massive layoffs from Google real soon.

They have the infrastructure to run a model of Gemma that is smarter than anyone of their employees except for a few key people like the AI model team and infrastructure team that handle the backend and network loads that are so insane they had to make a special bypass of the Linux kernel because it's too slow to handle the communication between the server NIC and CPU.

Google just needs to keep maybe 10k out of their 80k devs. All they need is ideas for future services and products and Gemma will handle almost all of the rest.

Web dev in a year maybe already, but certainly no later than 2030 and onward will "just" be about designing UI's that are easy to use and look good, zero lines of code will be done by humans. And how much has the UI changed in Gmail, Google Docs and so on over 2 decades? Basically nothing so it's not like they need a massive amount of UX designer's either.

Both Claude Opus 4.7 and Gemini 3.1 Pro can write perfect html, css, java, you name it.
I know, because I've asked both to build different stuff from 0 and it worked great and the code was solid.

Especially Claude is actually insane, doesn't just almost always nail stuff first try, it creates proper documentation of everything and nice read.me and get started guides and tests it's own code on autopilot, you only need to let it free and it will let you watch youtube or what ever instead.

[anxkhn]

L move.

Gemini Pro models will only be accessible via paid subscriptions. Free tier users will be limited to Gemini Flash models.

[kc8se]

They just increased the price for Workspace because it now offers such a huge benefit through Gemini amongst other things. Then around a month later they remove the ability to use all except flash models. It's time to de-Google my infrastructure.

[gimlichael]

Google Plus subscription here in the EU:

Its great that Google is capable of providing good and free open-source models such as Gemma 4; it sucks that they are behind on the cloud frontier models .. and treat Plus as second rank citizens 😞

[JM-Grande]

[germangar]

It is not. I tried it weeks ago. It only gives you tokens for the web and nanobanana or something like that. I had to go back to pro.

[blouflab]

most of us here don't give a damn about the ecosystem, we just want the CLI.

[sigmaSd]

This post doesn't mention the decrease in flash quota , these days it seems it's not 1000 free requests anymore , more like 250 or so , is this correct? And is it on purpose? If so the readme needs to be updated as well

[bdmorgan]

The readme doesn't mention anything about Flash quota.

[sigmaSd]

It says 1000 free requests what is the intended meaning then, it definitely makes you think it's flash especially since that was actually the case, it's just these days it got nerfed

[Jufralice]

Its 150 requests per day with the flash models

[ryanjsalva]

Good question. Our published limits include totals across model calls including Pro (when applicable), Flash and Flash lite.

[germangar]

IDK, each request spends to much time queued that I never reach the quota with flash.

[hjkimbrian]

In terms of purchasing, is AI Pro/Ultra the only option? How about Gemini Code Assist for enterprises?
https://developers.google.com/gemini-code-assist/docs/overview
https://docs.cloud.google.com/gemini/docs/codeassist/set-up-gemini#purchase-subscription

Or organizations that purchase Gemini Enterprise?
https://docs.cloud.google.com/gemini/enterprise/docs/editions

Can you please confirm?

[ryanjsalva]

Good question. Gemini Code Assist for Enterprise remains an option.

[Mins-sr]

Gemini Code Assistant Standard と Gemini Code Assistant Enterprise は 区別されていますか？ どちらも有料のプランです。

[SaschaHeyer]

@ryanjsalva
@NTaylorMullen

I am using Gemini CLI through its official ACP protocol for this
https://www.linkedin.com/posts/saschaheyer_ai-opensource-gemini-activity-7439703629630148608-H-p2

Same interface Zed and JetBrains use.
The CLI runs with standard Google OAuth, no direct API bypassing.

Wanted to flag this in case there's anything to keep in mind for automated/daemon usage of the CLI regarding TOS and fair use. Would love your thoughts and feedback on this as we want to prevent any abuse issues.

[bdmorgan]

I could be misunderstanding so please reply back with more information, if you think it would be helpful. However, based on what you're describing, that sounds like a legitimate use of the product.

[skellorok]

Please clarify the allowed usage through the Gemini CLI's Agent Client Protocol (ACP) implementation --acp flag as well as through the Agent-to-Agent (A2A) implementation https://github.com/google-gemini/gemini-cli/tree/main/packages/a2a-server.

Both of these are intended to enable third-party tool integration with Gemini CLI to route requests to Gemini CLI and both support leveraging the OAuth subscription configured inside Gemini CLI natively.

There are people building solutions based on these like @SaschaHeyer mentioned. Are you changing them? Are they going away? Are you removing OAuth subscription from them? The community needs to know before investing time building more integrations.

Thank you.

[DonRedwoood]

I need to know as well. I've built a tool that automatically creates sandbox containers for projects where the Claude Code, Gemini CLI, and Codex settings, plugins, skills, MCP servers, and OAuth are all inherited. So the users are still using Gemini CLI exactly as normal but within a more secure sandboxed container. I doubt this is a violation of any of Google's policies since it's basically a symlink. But it'd be nice to know for sure.

[skellorok]

@bdmorgan @ryanjsalva Please can we get a clarification to a legitimate question that impacts users and devs? Thank you. What use cases are allowed via your A2A and ACP implementations that are designed to allow integration with third-party tools and support using the user's OAuth-authenticated session that already exists in Gemini CLI.

[tisonkun]

+1 I'd like to ensure that both:

1. Create a cronjob (automation) that calls gemini -y -p
2. Create an application that replies to Telegram messages (as a bot) by locally calling gemini -y -p

... will never violate the ToS mentioned here.

AFAICS these don't bypass the OAuth procedure used by Gemini CLI and it extends how Gemini CLI can be used. You guys already support headless mode for automations/programming scenarios.

However, the words here, "using Gemini CLI oAuth with third-party software," are super vague and spread uncertainty. As mentioned in another reply, even using Gemini CLI with a terminal app can fall into this category.

cc @bdmorgan @ryanjsalva

[SaschaHeyer]

@ryanjsalva

Starting March 25, 2026, we’re changing the way Gemini CLI routes traffic with higher priority given to accounts based on license type and account standing.

This is very vague can you provide more details which licenses get the higher priority and what defines account standing?

[realcgslav]

What bizarre moves. By capping limits for users, especially Pro ones, you’re preventing them from building API-based apps that would ACTUALLY bring you money and scale to end-customers. I’m sure you’ve run the numbers, but awareness among potential future users is incredibly low. I run AI training sessions, and when I showcase Gemini-CLI or Antigravity, people are absolutely floored because they only associate AI with ChatGPT.

Obviously, after these nonsensical limit restrictions for PRO (and I’m not talking about usage outside of CLI or Antigravity), I’m done promoting you. I was doing it for free, and now, on top of that, these limits make it impossible for me to use the tool normally or build anything worthwhile.

These are very poor moves that undermine many of your indirect revenue streams.

[tho-maj]

seems like they aren't good in systemic thinking. weird.

[LeanMaster69420]

seems like they aren't good in systemic thinking. weird.

I thought they had those big thinking models.

[lst0451]

I used to think Claude was too expensive, but now, with Google’s restrictions, Antigravity is even more expensive than Claude. That makes no sense, so I might as well just stick with Claude.

[KF-76]

I work as a business consultant in addition to developing software, and i may also stop if these changes impact workflow whatsoever. The priority and model changes are especially concerning. This was google's chance to gain some clout by taking the high road and make millions of online businesses that connect to their existing paid services. If they botch this that will never happen again.

[odapplications]

Why not just explicitly block / ban users with openclaw? After reading the doc that seems to be the only ISSUE.

[FranckALPHA]

Yeah, I think the problem is detecting when it's used by OpenClaw.

[AMGMNPLK]

Why?? If I pay for credits, allow me to use them as I want

[Transient-Onlooker]

Ofc credits should be allowed. The ones with problem is OAuth, as I think so.

[andresaraujo]

If you think your account was flagged in error, please reach out to our support team by filing a ticket at console.cloud.google.com so we can get it sorted for you.

This does not work? I can't seem to be able to create support ticket to try to restore access to gemini-cli (with Pro subscription), Is there any option I can get some support or at least clear answer so I can cancel, (my case)? @ryanjsalva

Btw the link should be https://console.cloud.google.com/support/createcase/v2

[kurokirasama]

a really bold and wrong move, i guess i'll start looking for alternatives

[AhmadMWaddah]

i am using now Qwen Coder CLI too with Gemini CLI, both as a Consultant (Gem) and the Developer (Qwen), i also will search for another alternatives for Gemini CLI to work with Qwen,

[Transient-Onlooker]

Agreed for moving to qwen. Qwen as light model, Claude ad heavy ones.

[bitsnaps]

@AhmadMWaddah KiloCode has always free tier LLM, and it can be wired to QwenCode, which is one of the best alternatives at the moment.

[DJTOMATO]

lol, lmao even, yep that's over. just ban openclaw users.

Gemini Pro models will only be accessible via paid subscriptions. Free tier users will be limited to Gemini Flash models.

[AMGMNPLK]

Why?? If I pay for credits, allow me to use them as I want

[neohemp]

Why?? If I pay for credits, allow me to use them as I want

Yeah it doesn't really make sense isn't that what the quotas are for lol

[ahmedfaaid]

LOL.

[sasaiber]

Starting March 25, 2026, Gemini Pro models will only be accessible via paid subscriptions. Free tier users will be limited to Gemini Flash models. Upgrade to a Pro or Ultra plan to access Gemini Pro models.

Now the CLI is useless with flash models for free use. It's a shame... although blocking the lobster lovers would have been enough.🤌🏻
AI tools are becoming less accessible, no wonder

[adamoutler]

I agree. Flash models should not be anywhere near a terminal.

[ihorh]

Well, after mid-March, gemini cli just feels "dumber" than before. And slower. Either it is routing, or restrictions, hard to reverse-engineer. Pre mid-March I've been doing pretty amazing stuff with gemini-cli and have been able to stay within 1000 RPD quota most of the time. Now quota is burning in few hours with much worse outcomes :-(.

[oerx]

With OpenCode and Superpowers, it’s working fine for me, even with Gemini 3 Flash Preview.
https://github.com/obra/superpowers

I think it can also be used with gemini-cli, but gemini-cli often removes lines of code while generating them. That’s the main issue with gemini-cli.

[haryn]

China AI opensource disruption of US AI bubble seems to be working ;-)

[omergoktas]

• Gift people "Pro" plans obnoxiously and then call it "abuse" when people actullay use it.
• Call your plans "Plus", "Pro", and "Ultra", but underneath implement more, hidden "restricted" and "unrestricted" versions on each plan on a whim.
• Implement unclear quotas -weekly vs hourly- and prevent people access on random heuristics-based restrictions.
• Call it "abuse" and "restriction" but don't explain why on per case basis.
• Ban paying customers and do not allow refunds.

What the hell is wrong with you Google?

[tuapuikia]

Don't forget, Google can review what prompts you are sending to the Gemini CLI or Antigravity; they just don't use those prompts for training. They can even detect whether a prompt is triggered by an agent or a human by looking at the prompt style and signature. If you review the prompts sent by an agent (using the API and enabling prompt logging), you will see that agents are burning token services.

How can someone pay $20 or $250 and expect to get $1,000 in benefits in return?

[lab-key]

How can someone pay $20 or $250 and expect to get $1,000 in benefits in return?

The issue is you don't have any guarantees in terms of a return you still need to figure out a lot yourself. Why are they charging people for a semi broken product they are looking to keep charging more and more? What happens when the model comes to a point where $20 API usage actually allows for it to do something that can even guarantee make you $25?

Are the people that were actually paying and spending a lot of money to get them there gonna get some guarantee usage?

[kstepyra]

429 with almost fully unspent limit. Any updates, google?

[tuapuikia]

Anthropic will remove Claude code plan soon. Good! Gemini API method is the way for power user 😁😁😘

[Gybra]

Ahahahaah where did you hear that?

[tuapuikia]

Ahahahaah where did you hear that?

Anthropic officially will only support users that subscribe to max plan or higher or API pay as you use method.

[Gybra]

Oh well it is basically useless the pro plan as it finishes almost with two prompts. It has no sense at all using it with claude code though

[shawnholt]

I have a repo / project folder to help me manage and configure openclaw. IT IS OUTSIDE the system ... but the mere mention of openclaw seems to trigger this! INANE.

[tuapuikia]

It's okay if you use Gemini to fix Openclaw code, but it is not okay to use the Gemini CLI to manage third-party tools. Use the API method to avoid breaching the TOS.

[elsaviturki5-png]

We're making changes to Gemini CLI that may impact your workflow. │
│ What's Changing: We are adding more robust detection of policy-violating use cases and changing how we │
│ prioritize traffic. │
│ How it affects you: This may result in higher capacity-related errors during periods of high traffic. │
│ Read more: https://goo.gle/geminicli-updates

[elsaviturki5-png]

We're making changes to Gemini CLI that may impact your workflow. │
│ What's Changing: We are adding more robust detection of policy-violating use cases and changing how we │
│ prioritize traffic. │
│ How it affects you: This may result in higher capacity-related errors during periods of high traffic. │
│ Read more: https://goo.gle/geminicli-updates

[elsaviturki5-png]

✕ [API Error: [{
"error": {
"code": 403,
"message": "Verify your account to continue.",
"errors": [
{
"message": "Verify your account to continue.",
"domain": "global",
"reason": "forbidden"
}
],
"status": "PERMISSION_DENIED",
"details": [
{
"@type": "type.googleapis.com/google.rpc.ErrorInfo",
"reason": "VALIDATION_REQUIRED",
"domain": "cloudcode-pa.googleapis.com",
"metadata": {
"validation_error_message": "Verify your account to continue.",
"validation_url_link_text": "Verify your account",
"validation_url":
"https://accounts.google.com/signin/continue?sarp=1&scc=1&continue=https://developers.google.com/gemini-co
de-assist/auth/auth_success_gemini&plt=AKgnsbt1lGVDQ4ug8jrsynFp2XO-3dJPOJQLsLo1On-EZXLo4_a-pv7jVXW1ol2corn
iNIAk84PWgqm4MxB7sXaucIjTfDFtbWcT65ciXp0UDqyZ6qvfSUCrFhdeYzGHgGZXMVT0x4hW&flowName=GlifWebSignIn&authuser"
,
"validation_learn_more_link_text": "Learn more",
"validation_learn_more_url": "https://support.google.com/accounts?p=al_alert"
}
},
{
"@type": "type.googleapis.com/google.rpc.Help",
"links": [
{
"description": "Verify your account",
"url":
"https://accounts.google.com/signin/continue?sarp=1&scc=1&continue=https://developers.google.com/gemini-co
de-assist/auth/auth_success_gemini&plt=AKgnsbt1lGVDQ4ug8jrsynFp2XO-3dJPOJQLsLo1On-EZXLo4_a-pv7jVXW1ol2corn
iNIAk84PWgqm4MxB7sXaucIjTfDFtbWcT65ciXp0UDqyZ6qvfSUCrFhdeYzGHgGZXMVT0x4hW&flowName=GlifWebSignIn&authuser"
},
{
"description": "Learn more",
"url": "https://support.google.com/accounts?p=al_alert"
}
]
}
]
}
}
]]

[tuapuikia]

Isn't it obvious? Google KYC is kicking in, requiring you to provide information to verify who you are to avoid resources being abused.

[codayon]

After reading some of the comments I'm unsure, whether using a premium plan will be good idea.

[tuapuikia]

Use the pay-as-you-go API if you are unsure whether your usage will breach the TOS.

[xZero707]

Well, traffic prioritisation is working absolutely great! With almost no used quota and premium account, no models are accessible today. Or better said, every tenth attempt might yield an answer after waiting for 10 or more minutes. At this point I can do stuff faster manually.

[tuapuikia]

Well, traffic prioritisation is working absolutely great! With almost no used quota and premium account, no models are accessible today. Or better said, every tenth attempt might yield an answer after waiting for 10 or more minutes. At this point I can do stuff faster manually.

Well, traffic prioritisation is working absolutely great! With almost no used quota and premium account, no models are accessible today. Or better said, every tenth attempt might yield an answer after waiting for 10 or more minutes. At this point I can do stuff faster manually.

I have no problem to use Oauth to fix my brownfield project. I have API billing method for backup.

[sadkodev]

[adamoutler]

This is wonderful progress. Google has successfully halted all abusive and legitimate inference traffic from using Gemini CLI during the time periods between Tuesday morning to Friday Afternoon. The answer was right there all along, just shed everyone's connections from the server clusters and deprioritize free, pro, ultra, tier 1 API, and tier 2 API users.

Looking back, one must wonder why this wasn't implemented sooner! It is just so easy to ignore the valid requests to achieve all traffic prioritization goals! Good job Google!!!

[Sing-Li]

Absolutely fascinating to watch Google (finally!) having to deal with scale !

Scale of viral global adoption of their initial AI developer offering (which used to be leading only a couple of months or so back). (Anthopic's 40b ARR and growing - this is no longer our grandpa's "search scale"!)

Hope this continued struggle will choke out some short sighted mid-level management and get it to attention of some higher up visionaries 🤞

[adamoutler]

#24937

[fyalcin]

can someone tell me if invoking gemini -p inside a script running as a cronjob might have triggered a restriction on my account? It was super light script just running once a day. Asking because I keep getting the error

"code": 429,
"message": "No capacity available for model gemini-3-flash-preview on the server"

on a Pro plan. There's nothing else in the error message that points to a possible restriction but the OP here is super vague so I'm not sure if they'd make it obvious in the error message if I were being restricted...

[tuapuikia]

Is the request itself that trigger the filtering. Gemini backend can see your prompts but won't use it for training. So if you use it as invocation tool by 3rd party then you will have risk to get ban. Gemini cli is tools for development and not for 3rd party automation. If you use for automations, please use API method. You won't get ban for using it. Oauth is value added service from pro or ultra plan with large cloud storage.

[AkimioJR]

This is such a letdown. I'm a subscriber to the Google One AI Pro plan, yet after sending just a simple Hello, the gemini‑cli didn’t respond for 20 whole minutes.

[erlangparasu]

how about via Antigravity?

[AkimioJR]

Antigravity

It seems ok in Antigravity. I tried again in gemini-cli with the -p flag, but it always return error code 429.

[Cian-CloudIntCorp]

[tuapuikia]

Sorry you feel that way @fyalcin , I am just responding to users who are asking me questions? PS you can't make the AI claim on an AI related thread friend. One simply does not have enough time to respond, I was trying to be nice and that is why I personally use AI to filter my response like most of us, this response is all me and shall be my last, I have already said Im looking for feedback on code not marketing. My last reply is taken directly from the project read me and my comparison document last week on my LinkedIn Comparing Hermes and Open Claw to Bare-AI and also Google Gemini CLI: https://www.linkedin.com/posts/cian-egan_bare-ai-agent-vs-openclaw-vs-hermes-agent-activity-7454772473461379072-lNbx?utm_source=share&utm_medium=member_desktop&rcm=ACoAAATjnJMBfTe4nmpFbj75hCeXdd8yqQeFQtA.

PS I started this project as a bit of fun as google said they open sourced gemini, however, what they did not tell us though was that gemini cli was heavily locked down to google ecosystem and it was not very open source to perfectly honest with you, even though the whole project exists because of open source tools and open source contributions too many to list (see bare-ai.js formally gemini.js formally many many other contributors even meta if I remember correctly. My own work here, is not vibe coding ( I am exIBM Chief Integration Architect of Workday Practice in EMEA) and given my experience with APIs,I was looking for a new challenge. This project has over 200 commits to the main CLI and brand new repo on top called bare-ai-agent written in bash) and let me tell you it took a LOT of work (months) to fully unstrip it from google back to true open source. The version I released is truly open source built off the backs of many contributors. I wont be making any more changes to it though other than bug fixes. I would also like others to contribute. Personally I am moving on now to build on top of it premium models but wont be advertising here, its for a different type of user. Thankyou for reading.

I rather using gemini cli with api key than using something that is forked from it. If i need a feature I can develop it myself. Whether open source it or not is my personal choice. Please stop promoting 3rd party app in 1st party repo. Also, using vault will not make your software secure.

[Cian-CloudIntCorp]

100% agree, if you just want google only then use gemini cli. If you want to work with any ai and have your keys stored securely and swapped out as your model swaps out and you want the open source gemini project feeling then use bare-ai-cli or maybe a fork of that cli if it happens... Easy. Lots of other discussions on this forum about other ai providers too btw, Luckily it's a free world. If you don't like it then just scroll past , but I do want to thank you for your time. Honestly, I can tell who is over 40 and who is not by the comments. It's a free country/world. True Open Source projects used to mean something. For any other vibe-coders listening, just remember, that we all stand on the shoulders of giants. (not mag7) literal human beings who built most of the underlying software and I want to pay special thanks to the 600 contributors who helped make my project a reality (by abstraction) along with the thousands of others who those 600 built on. Thankyou, for reading.

[kstepyra]

@Cian-CloudIntCorp dude it's still an ad, we fight google here, not look for alternatives. there is plenty of them, and nobody is looking for them here.

[Cian-CloudIntCorp]

@Cian-CloudIntCorp dude it's still an ad, we fight google here, not look for alternatives. there is plenty of them, and nobody is looking for them here.

@kstepyra, I do hear you "dude" :), but I did not have googles budget and the project is free, I WAS honestly just looking for some folks to try it and provide feedback and thought hey look "a bunch of disgruntled vibe coders mad at google would be a good start", but hey, live and learn. lol. All your points are taken, most folks here want "click and play" and for the "ai to brush their teeth too". That's Ok, I get it , however, the best way to fight google if that is your cause is to be model agnostic and to use ALL the AI providers. This is exactly what happened with anthropic yesterday, the deal with X was effectively a response to the community backlash over the low limits applied to their plans. Thank you for your feedback.

[haryn]

Use any LLM with gemini-cli follow these instructions: https://haryn.hashnode.dev/the-gemini-cli-paradox-route-to-your-own-endpoints-a-digital-thriller

[amatiytsiv]

amazing fix now gemini cli cannot authenticate - sms delivery just fails in Spain. im wondering what a point in such a great security improvements. no way authenticated besides sms is avaiable wtf ?

[erlangparasu]

killed by google

[doroved]

Gemini CLI - RIP

[Gybra]

@doroved did you already tried it? It suffers the throttling request mechanism as well or it is now reliable and fast as other LLMs providers?

[Skyxim]

@doroved did you already tried it? It suffers the throttling request mechanism as well or it is now reliable and fast as other LLMs providers?

I've already tried it, but the experience with Flash 3.5 or CLI is still unsatisfactory; after 2 hours, the server experienced high load.

[Gybra]

@doroved did you already tried it? It suffers the throttling request mechanism as well or it is now reliable and fast as other LLMs providers?

I've already tried it, but the experience with Flash 3.5 or CLI is still unsatisfactory; after 2 hours, the server experienced high load.

ok then 😄 i was curious to give it a second chance but you've confirmed nothing has changed since i've switched to claude 2 months ago now

Thanks anyway!

[Skyxim]

@doroved did you already tried it? It suffers the throttling request mechanism as well or it is now reliable and fast as other LLMs providers?

I've already tried it, but the experience with Flash 3.5 or CLI is still unsatisfactory; after 2 hours, the server experienced high load.

ok then 😄 i was curious to give it a second chance but you've confirmed nothing has changed since i've switched to claude 2 months ago now

Thanks anyway!

I just tried again.

⚠ Our servers are experiencing high traffic right now, please try again in a minute.