vibe-coders

[round solstice]

Wow.. been using this for a few days. Amazing.

Can't wait until they add "bring your own keys" to get around the usage limits

[merry folio]

https://tenor.com/view/sweet-so-sweet-gif-844156360703251045

[round solstice]

This could be the primary interface I use multiple hours per day.

10 years ago, GNU Emacs. 5 years ago, VSCode. Now, this?

[lucid idol]

Really gonna need a response on how Gemini 3 Pro (High) is automatically routing tasks to a combo of 2.5 flash and Gemini 3 Pro (Low) though

https://cdn.discordapp.com/attachments/1365049274068631644/1440952563993149531/Screen_Recording_2025-11-20_at_01.22.52.mov?ex=69200726&is=691eb5a6&hm=312a6106c96eb1161b550f2422de38742877c1e0fc562f005a1579d08a614865&

[wary stump]

they must increase the limit

[lucid idol]

This is not a limit

I can see my quota in the network logs for each model

I had 95% of my Gemini 3 Pro (High) remaining

While it was doing this

Ironically I was down to something like 79% remaining of 3 Pro (Low) by the end of this

Because it was just routing me to flash or 3(Low) all of the time

[wary stump]

i mean quota must be increase

[lucid idol]

oh yeah. i just meant this specific thing i posted was NOT due to using it enough to hit any rate limits

[quaint kelp]

yo

[mighty basin]

unable to add pplx mcp server

[quaint kelp]

these models are great does anyone have any information on when pricing will be available?

[naive sparrow]

Hey antigravity Agent keep failing to load?

[lucid idol]

lol...

[quartz gorge]

Ohhhhhhhhhhhhhhhhhh I love that we've got an anti gravity chat!

because antigravity is sooooooooooooooooo good

[naive sparrow]

What do you like the most with it and what are you working on

[quartz gorge]

I love the transparency, it's planning and the execution. The entire workflow is like working with a very switched on junior who is keen and exceptionally good - the dream junior who codes like a senior but obeys like a junior should 😄

My use case at the moment is developing python notebooks that teach GenAI engineering skills to those entering the industry for the first time. One of my favourite use cases is removing all the code from the workbook except imports or anything defining data. It's so super simple but something we find really time consuming - we essentially have two versions of all of our workbooks - the student version and the solutions version. The agenetic side of Antigravity has been able to make this painless for us. We first create the solutions version, which the agents have been really great at working with - I've struggled with some tools because they over complicate or miss the mark on the whole "this is for teaching" side of things. Antigravity "gets the point" with working on these workbooks. Once we've got a solutions version we're happy with, we'd typically then have to strip the code and save it. Again, antigravity now does this in seconds for us, which is brill. Importantly, it's also been amazing at reflecting changes we make in the workbooks. When we update one of them, it's been great at reflecting the change in the other one for us, which is traditionally something that was a bit of a nightmare. I'm sure this is all stuff we could have built solutions for - but AntiGravity has made it super simple.

[naive sparrow]

😊 fun

@quartz gorge have this happened too you do you how resolve it

[quartz gorge]

Yeah, I think it's so popular that this is happening a lot!
It's just patience and trying again at the moment

[cosmic ferry]

We got a channel for anti gravity?

How did I miss it!

I was looking everywhere for it lol

I tried to customize my Anti Gravity's look. here's how it turned out haha

[lucid idol]

is it confirmed whether or not certain users (ai ultra or enterprise users etc) actually do have access to gemini 3 pro high?

ive tested prompts that take 10+ mins to think (as in it wont route to low thinking on purpose from lack of difficulty) and still per network logs get routed to 2.5 flash and 3. pro low before task runs

as in

the model/user/task is told it doesnt have access to 3 pro high

[cosmic ferry]

How do I know if it's using banana 2?

@lucid idol ur here too?

hmmm

Left input right output by Anti Gravity

[lucid idol]

im everywhere

or so ive been told

[cosmic ferry]

I mean Screenshot to UI

[lucid idol]

[cinder osprey]

Im here too 🙋‍♂️

[naive sparrow]

No

[bright spindle]

This thing is so good, but will we ever get more gemini 3 tokens? Feel like I am hitting the rate limit every day

[wintry viper]

helloo Is fronend cooked

[agile monolith]

Hi everyone

[nocturne venture]

Same here

[agile monolith]

Same , we need more tokens

[ebon sapphire]

is there a way to check current usage in antigravity? i have not looked this up yet anywhere online. I'd love to get more details on that to understand when I'm about to hit the limit

[ebon sapphire]

For Linux users, if you installed antigravity and now you're seeing an error when trying to update like

N: Skipping acquire of configured file 'main/binary-i386/Packages' as repository 'https://us-central1-apt.pkg.dev/projects/antigravity-auto-updater-dev antigravity-debian InRelease' doesn't support architecture 'i386'

you'll need to update the definition to pull the right files for your architecture. if you followed the instructions from the antigravity website, you'll need to add the right arch in /etc/apt/sources.list.d/antigravity.list so that it ends looking something like:

cat /etc/apt/sources.list.d/antigravity.list
deb [arch=amd64 signed-by=/etc/apt/keyrings/antigravity-repo-key.gpg] https://us-central1-apt.pkg.dev/projects/antigravity-auto-updater-dev/ antigravity-debian main

anyone from the antigravity team here to help update the installation instructions to make sure the arch is set right?

[terse jacinth]

Wait, is Antigravity just a customized VSCode fork?

[languid seal]

Yes, Antigravity is based on VSCode Open Source

Per Google Documentation
The primary entry point to Antigravity is our Editor, a surface based upon the VS Code codebase but full of rich AI-enabled features designed to improve your code-writing experience.

[terse jacinth]

Great!, I'm moving away from VSCode right away

[spare quail]

Is antigravity also building you apps with google tools/services like in aistudio?

[languid seal]

Not sure how you are defining like but yes, Antigravity is there to help you build apps.

[bronze grail]

Same

It outperforms replit agent so im sold

[round solstice]

I've contnued to be impressed. But I'm angry when I hit the freebie wall!

[stoic hound]

Hey dev team, I have a suggestion.

You should consider adding a source section like NotebookLM for:
a) inspiration - like a color palette PNG, or screenshots of features you like, or an example of a layout drawn in an image editor.
b) resources - like PDF documents explaining complex business use cases, font files, logo files, partial code, secrets/keys, etc.

P.S. This looks amazing to use as is. I just finished watching the video and if it weren't late I would be jumping on to use it ASAP. I can see so much potential for where you can take this in the future too which is exciting.

[mighty basin]

btw how to globally configure mcp servers in antigravity

[jovial ermine]

nop

[lucid idol]

You realize you can just add them into your folder and use them as context right

[spare quail]

I meant if it locks you in with google services. In aistudio when you ask to build something it will choose their own google services (which is nice too). I was wondering if other people had the same experience but with Antigravity

[agile monolith]

Hi everyone, what are your use cases for antigravity?

[mighty basin]

Inside the editor while coding u feel g = -9.98m/s²

So no diff in dx u code from earth moon or space

[remote widget]

just be sure docker is running

[mighty basin]

Docker why?

[arctic pumice]

Trying to think of startup ideas😂

[drowsy star]

can we please get more tokens if we have pro/ultra subscription ?

You cant use cli or antigravity to help with coding tasks, gemini 3 release just does not feel that good, but the model is great

[steel reef]

alright guys finally found an official antigravity chan lol

they go out of their way to keep this hidden

anyway any idea why on macOS it installed 7 extensions by itself and didn't install any on windows installs? what is supposed to be the default?

[cinder osprey]

You mean when you import from VsCode to Antigravity?

[steel reef]

Didn't do any importing in either install

just installed those 7 extensions on macOS and nothing in Windows - I remember the setupflow maybe mentioned something about extensions but didn't pay attention so my bad

[cinder osprey]

When i installed first time Antigravity i clicked to import from VsCode but did not worked but i had to reinstall it yesterday and then import worked. Dont know why. Im Windows user

[steel reef]

I dont have these extensions installed anywhere else so its definitely not an import and it only installed on macOS

[cinder osprey]

Ahh

I dont know

[agile monolith]

Lmao

[fervent python]

Is there a way to disable the agentic mode in the chat pane and only use mission control for the agectic tasks? Despite using the fast more and instructing the model not to change the code and only answer questions, sometimes it will goes off and starts changing.

[fervent python]

Fork antigravity 😅

[bronze grail]

antigravity is REALLY good with file organization scripts

[patent heart]

super ide

[runic lily]

I've been an AI (for coding) sceptic aside from GPT here and there - used VS Code for years, decided to give this one a go - In an app I've been working on for over a year it's just generated 5 admin dashboards for me in less than an hour - probably would've done it faster if I didn't prompt for each one individually.. as dirty as it feels to be using this stuff directly in my codebase - I'm hooked on this drug 🤦‍♂️ 🤣 earlier I refactored an entire frontend with it.. I'm more a backend dev so this would've taken me hours, possibly days to get where it got to...all of this for free and still having full control over the code!

**bit worried about what might happen when they come out with pricing though.. I mean there are OSS plugins for VS code that would do this, so I mean if you can pay for Gemini 3.0 pro API in the end then one could use one of those I guess...lets see..

[twilit jewel]

hi, need help in solving a challenge

[urban glacier]

did you find anything related to this?

[ebon sapphire]

nope but also didn't looked for long enough so there could be something in the docs that i missed or didn't reach

[astral sorrel]

Hi. Does Antigravity have an always reading guideline file like GEMINI.md, AGENTS.md or CLAUDE.md? I didn't see one in the doc. What's the best practice for this?

[delicate ember]

Can you please add the rate limits to the docs pls

[static wasp]

Hey everyone, I am just curious since there seems to be a lot of excitement around this. What's gives this an advantage over something like AI Studio or other vibe coding apps?

[naive sparrow]

Think it need more work is really not there yet crashing in bigger projects

[stoic hound]

Yes, the suggestion is primarily a UX improvement. If there were predefined and fine tuned pathways for certain types of data, it would make a reliable way to input user or business requirements.

[ebon sapphire]

hitting some funky errors now but looks like other users seen the same in vscode 🫠

Error loading webview: Error: Could not register service worker: InvalidStateError: Failed to register a ServiceWorker: The document is in an invalid state..

[worn trail]

A bit late, but I already have it! 🔥 🔥 🔥

[thick mantle]

Antigravity leverages Gemini's larger context window, allowing it to generate detailed implementation plans and checklist-based task lists for complex software development. The resulting ease and quality with which the LLM (like Gemini 3 or Claude) handles these complicated task plans dramatically improves the AI's ability to execute and sustain a project over a long duration.

[winter crane]

Anyone know how to get Gemini to use Antigravity's browser/chrome integration? I've installed the extension, and have made sure I can manually open it with the chrome button - but Gemini never seems to do anything with it, even if I try to mention it in the prompt

[mighty basin]

i've specified never ask for review and turbo mode for commands execution too still

its asking every time before running a command

[heavy pawn]

google ai pro wont work for antigravity? i bought but i still get limited

[honest forum]

hey guys, I have noticed that I am not getting really good results with antigravity vs AI studio, why is that the case? its rather terrible... also it doesnt seem agentic event tho I have set it to turbo mode, it still requires me to accept commands which makes it slow and still not doing things correctly....

[winter crane]

Daaaaaang, I'm impressed! I asked Gemini in Antigravity to generate a new site logo based on my overall project's main brand logo, and it did of course use Nano Banana to do so - but even more impressive is that when it wanted to implement the new logo, it tried to make it transparent via Nano Banana. That failed and I reported it to the model (NB replaced the background with a checkerboard pattern to make it seem like it was transparent haha) it decided to instead write a quick Python script to make the original (non fake transparency) image into a proper transparent one

That was very crafty of it to do instead of just saying "Ah sorry, I can't make it transparent"

[steel reef]

anyone else who's using AntiGravity on a mac and windows

can you check if you had 7 extensions installed on mac and 0 on windows

[cursive island]

Anyone able to get it to run without crashing or doing mock data...

[static wasp]

TYSM!!

[languid seal]

Antigravity currently in free public preview status with unpublished (as far as I know) rate limits during this phase. Paid plans and limits are TBD.

[buoyant token]

Hi! I’ve been running into limits using AGY with 3.0 high. Can I use my API key from AI Studio?

[heavy pawn]

[winter crane]

Not yet from what I recall

[cursive island]

Is it just me or is gemini 3 super lobotimized

Its not doing 1/2 of the cool stuff it did on release day

[tired bluff]

wouldn't be surprised if it is, AI is expensive to maintain at that scale, so, they'll take a lot of measures to reduce the burn. It's the Burn and Churn cycle

[cursive island]

Constantly corrupting files like NONSTOP

[astral sorrel]

I found the same problem. I didn't see any mentioning for this in doc. I ask here and no one answers.

[thick mantle]

what specific "cool stuff"? also were you accessing gemini 3.0 via ai studio/gemini, or via Antigravity on release day?

gemini.md from gemini-cli /init command?

[thick mantle]

what's the total length of files you're working with? 700,800, 1000+ lines?

[cursive island]

Both

[cursive island]

I got quite a few, longest lines I've had is 2.5k, one is a documentation doc with 50k words (I don't remember line count) but it doesn't call that often unless I ask

[thick mantle]

did the pillow script convert the png to a different format first and then do something something something "alpha mask", or replacing specific colors with transparent ones?. Peaked my interest. I might try setting up a custom workflow in antigravity to repeat something like that as I've had tons of images come out with the model mimicking how a transparent background looks in say, photoshop? And I always have to resort to just going with a "solid green screen background" and manually having to edit the file that way. Always end up with an image having horrible half transparent green outlines =/

[thick mantle]

My guess is that Antigravity's custom "system prompts" it uses were tightened to improve code reliability

[winter crane]

Uhhh, possibly. PIL / Python isn't my speciality, but what I can do is provide the source for the script that it ran since its still saved:

# ~ ❯❯❯ cat $(find ~/.gemini -iname '*.py') --plain
from PIL import Image
import sys

def remove_white_background(input_path, output_path):
    print(f"Processing {input_path}...")
    img = Image.open(input_path)
    img = img.convert("RGBA")

    datas = img.getdata()

    newData = []
    for item in datas:
        # Check if the pixel is white (or very close to white)
        # RGB (255, 255, 255)
        if item[0] > 240 and item[1] > 240 and item[2] > 240:
            newData.append((255, 255, 255, 0)) # Fully transparent
        else:
            newData.append(item)

    img.putdata(newData)
    img.save(output_path, "PNG")
    print(f"Saved to {output_path}")

if __name__ == "__main__":
    input_file = "/home/russjr08/.gemini/antigravity/brain/b37434e7-1bee-464a-9029-77ef5426cede/project_logo_1763897307904.png"
    output_file = "/home/russjr08/dev/project-path/public/project-logo.png"
    remove_white_background(input_file, output_file)

[thick mantle]

oh this is going to save so much time. tyty

[winter crane]

No problem! I'll definitely be keeping it around too for the future as well, since Nano Banana definitely likes to generate stuff with a white background if there isn't an "actual" background (such as logos)

[astral sorrel]

I think so. Is there a following guide file for Antigravity? Nearly every agent has one. But I didn't find Antigravity one.

[stoic citrus]

Whats antigravity?

[cinder osprey]

AI Supported IDE from Google

[honest forum]

why do I still have to manually accept evern tho that I got turbo enabled???

[craggy nebula]

Have this happen to anyone else during set up?

[steel reef]

anyone from Antigravity know why these randomly got installed on my macOS install?

[fast kettle]

its a thing u can chose when u first install

it asks if u want to install like 8 most common extensions or something

[steel reef]

I ran through the same setups on both windows and Mac

And only on Mac I got those dumped and it’s only 7

[fast kettle]

the setting prob got linked to ur google account

[fast kettle]

i didn't remember the exact count

but yea at least for me it asked me to chose

[cursive island]

Ugh

[wheat orbit]

Loving the product but I can get out of control java tasks that slow my machine down

[bronze grail]

Who named antigravity?

[bronze grail]

It's worthit. VS code with gemini 3 pro based ui

And what is the Agent Manager vs editor can someone explain better?

[cinder osprey]

You mean in VS Code?

[bronze grail]

no in antigravity

[languid seal]

Have you read the docs? What's your baseline knowledge of coding, AI, agents, etc.?

Simply
Editor is the code editor
Agent Manager is the management tool for agents

Yes, I understand the redundancy and lack of explaination.

Check the docs and see if that helps. See if you have more specific follow up questions.

https://antigravity.google/docs/home

[spare quail]

Why did the name of the channel changed to vibecoders?

[merry folio]

Antigravity and Jules channel in one

[dapper wyvern]

Do Gemini CLI and Antigravity share the same quota bucket? I want to try Antigravity, but I don't want to use up all of my 3.0 Pro requests quota because I need them for CLI.

[cinder osprey]

There are different Plans for CLI and Antigravity but i cant tell for sure

[dapper wyvern]

Plans? I am on Gemini Pro and I use Google auth for CLI. I just don't know if I use auth for Antigravity if they use a combined quota or are they separate. Is that what you mean by different plans?

[cinder osprey]

You dont have advantages for Gemini CLI if you have Gemini Pro. Its the same like free tier then as i know

These are Plans for CLI:
https://developers.google.com/program/plans-and-pricing?hl=en

[dapper wyvern]

Are you sure? With Pro, I get 100 requests on CLI, but with Free, you get much less

[cinder osprey]

and these are for Antigravity:
https://antigravity.google/pricing

@dapper wyvern

[dapper wyvern]

I am not on one of those plans, I am on Gemini Pro.

[cinder osprey]

yes and as i know this does not affect gemini cli

[dapper wyvern]

OK, cool. So I will experiment away with Antigravity, but still get my full CLI quota.

[cosmic ferry]

was this channel renamed from "anti-gravity" to "vibecoders"?

I wonder if we will get any more 3rd party models in Anti Gravity

[cinder osprey]

yes

[mortal loom]

Same here

[rustic falcon]

Hey every one.

[dapper wyvern]

Finally tried out Antigravity for the first time. Worked really well for me. Can it access/work on files outside of the project directory? In my first project, it initially seemed to install Python libraries in my global installation (which I would have not thought possible because I thought it only stayed in the project folder). I ended up getting it to install and use a .venv local to the project. I was just surprised that it messed with the global one

[scenic arrow]

Google's Antigravity IDE is extremely laggy and performs terribly.
I'm using an Intel Core Ultra 7 265KF CPU, RTX 5070 GPU, and 64GB of RAM — still severe stuttering and frequent "not responding" freezes.

The Gemini 3.0 model is often inaccessible/unavailable.
Claude 4.5 inside Antigravity IDE is deliberately designed to be incredibly stupid and scatterbrained — it constantly forgets things and messes up.
I clearly specified the output language in my global rules, but the AI never follows it.

[dapper wyvern]

Strange. I am having the exact opposite experience. I am on a low end machine (compared to yours) and Antigravity has been fast and stable. Could it be a network issue for you?

[old falcon]

Sometimes changing models actually gives a faster response. I don't know why, but gemini seems to delay a lot, probably due to the demand. Try other models if you don't necessarily have to use gemini

[astral sorrel]

On windows, Antigravity agent terminal running is not working correctly on non-UTF8 system, for example, Japanese or Chinese.
I have to ask the prompt to add chcp 65001 to terminal running.
https://zenn.dev/knpgpg/articles/52e7ddcaa55a07

[lucid idol]

does ultra have better limits on antigravity?

using my gemini enterprise admin account i hit them fast

[languid seal]

Antigravity currently only has free plans and limits.
https://antigravity.google/pricing

[cunning crane]

wow new channel

[lucid idol]

It literally says right there you get more requests per day. And it was explained to me that apparently you get much more 3 pro requests the higher the plan youre on vs free

As in a lot of free plan requests are usabls for flash only

[dapper wyvern]

On Gemini CLI, how does the /compress feature help? It won't extend my daily requests quota, right? When should I use it?

[distant vine]

It reduces the size of the context being sent. Useful if you are working with large files, have a long session, and/or are paying by the token.

[dapper wyvern]

Ah yes, of course. I am using OAuth credentials and not an API key, so not paying by the token.

[vagrant oyster]

is antigravity currently down?

stuck on loading, and my agent manager chats arent sending either

[cinder osprey]

Where does it literally say that you have 3x more Pro Request in Gemini CLI when you have Gemini Pro Plan? i cant find it

[vagrant oyster]

nevermind, relogging in and spamming ctrl + l did the trick

[livid rune]

Question to all. Do uou find it better to vibe code in Google ai studio then move to anti gravity to clean and finsish up? Or do you find it better to start from anti gravity?

[stable bison]

I don't use either of them, I use any AI Ide to make a good base, and just debug with ChatGPT until it works, ik "There must be a better way to do this" but it's my style

[cinder osprey]

never tried this procedure first ai studio then antigravity to be honest but i dont see a reason to do that. With antigravity you can develop everything what you can develop with google ai studio and more in my opinion or am i wrong?

[round solstice]

I've build a project using AI Studio in TypeScript, then generated a PRD from that codebase with agy, then updated the PRD to select my flutter framework for implementation, then developed the app from that in Dart and Flutter. This is to work around that AI Studio can't preview Flutter code, even for web.

[sudden fable]

what is a good tool to use for mobile development using AI assist code tool like Bolt, and that has good UI development with backend/API integration?

[ebon sapphire]

i wanted to try starting with AI Studio and then passing continuing with Jules but I ended up with Jules in a weird state where it kept making branches for fixes and weird behavior in the chat window so I decided to pull the repo and continue locally with the cli.

[livid rune]

Less complex to start wifh AI studio. So I was wondering if anyone had good results doing that

[ebon sapphire]

what I got from AI Studio was a pretty solid starting point so you could say this is a good workflow!

[dapper wyvern]

Version 0.18.4 of Gemini CLI is absolutely blasting me with all of its thinking and planning, as compared to earlier versions. Is there a switch to turn that off?

[raw vapor]

Guys, can I post my links here? With generous aggregators/providers for free trial use of the API for Roo/Kilo Code and for CLI, such as Droid.

[cinder osprey]

Still no answer. If you make a claim, you should be able to prove it or at least explain how you came to that conclusion. @lucid idol

[main lion]

Hey, how's it going? Does everyone else see the "Source control" button that—according to the official documentation—appears in the "Agent manager"? I can't manage to see it, and it's quite frustrating because I can't reject changes made by the agent from here (I can only do it from the editor).

[main lion]

obviously is a GIT enabled project, etc.

[burnt prairie]

Hey folks. I was wondering if anyone has built a product from scratch with Antigravity?
I built one myself and would love to share it if it's okay.

[lucid idol]

i dont have to look for where i said 3x because i would never say that

but here u go

https://developers.google.com/gemini-code-assist/resources/quotas

if you still don't see it i can highlight the pertinent part for you

manually, not even with ai

[lucid idol]

seems you misread "you get much more 3 pro requests" as 3x as many 3 pro requests?

[cinder osprey]

Yes sorry missunderstood you said more 3 pro requests not 3x more requests

Thanks for the Link

[lucid idol]

im curious myself tho if u ever find out about the 1000 vs 1500 vs 2000 thing

if its just flat 50% and 100% more for standard and enterprise over free

or if u get (as u probably should?) substantilaly more gemini 3 pro requests in there, with the free's 1000 being mostly flash and the paid ones being mostly/all available as pro

[cinder osprey]

After some requests with Gemini 3 Pro it Changes to Flash and you have 1500 per day in total with Google Pro Account.

I dont know how many requests exactly you have with Gemini 3 Pro

Its nowhere written as i see

@lucid idol

[meager crow]

hi

[quartz gorge]

I'm having an interesting challenge with AntiGravity at the moment -it's absolutely convinced that Gemini 1.5 is the latest model. My boiler plate code from Vertex AI studio was absolutely demolished by it. It changed Gemini-2.5-flash to gemini-1.5-pro-002 and changed a bunch of parameters (since they didnt exist before). i tried telling it that it was wrong, but it decided the best cause of action was to tell me that i made a typo lol

[lime glacier]

Hello, everyone. I'm a full stack developer building smart, scalable, AI-powered applications with Next.js, Python, Supabase etc. If you need help, please DM me.

[lucid idol]

Lol.... sounds like the app....

I dont get that ever with enterprise or with vertex ai or i studio

The app though. I dont bother using it for anything

Very disappointing it happens in antigrav...

[lucid idol]

Link it the google website announcing gemini 3? And maybe this? Idk

https://ai.google.dev/gemini-api/docs/models#gemini-3-pro

Tell it its hallucinating and to verify with this as a quick fix

But

😐

[strange pebble]

Hey guys i was using Gemini and it for some reason always modifies code how can I fix that.

[round solstice]

"using Gemini" how?

[cinder osprey]

If you mean Gemini Cli or Antigravity Agent mode this is not a bug. 😆

[strange pebble]

I use Gemini 3 i think

[cinder osprey]

Over which Tool?

[strange pebble]

Fast

And I also use canvas

[cinder osprey]

Which Tool is Fast?

Canvas over the Gemini Browser UI?

[novel halo]

anyone getting Error during tool execution MCP server memory not found in Antigravity when using the @Conversations feature?

is there an actual feature that can read and search through previous conversations, or is this MCP actually needed?

[upper tulip]

Anyone knows how to pull github update to AI Studio "Build Apps", i made small changes my self, when doing "Save to Github" deletes files i created and made changes. Also when it created duplicate files, it doesnt remove them also.

[dapper fractal]

🌎 DevFest Challenge Alert! 📸 Your pet is still waiting for their global debut! Don't let them miss their trip to Paris/Tokyo/Rio! Use Nano Banana Pro in Gemini, drop your pet into a dream city, and share in #devfest-2025 to win swag! [More Info Here](#announcements message)

[potent jasper]

Great server idea! I use AI Studio Build often 🙌

Opal in Google Labs is another excellent vibe coding tool. It's a more modular approach that allows you to prompt and alter individual components of your workflow at the same time.

[reef tendon]

Is there an extension similar to "Background" for Antigravity IDE? Or is there another way to change the background in Antigravity?

[thick mantle]

This isn't a supported feature. You can export, but not import. Right now you could clone the repo to your pc and upload project files manually. Just verify the LLM model 3/2.5 utilizes the most recent uploaded version (as I'm not sure if it makes changes base on the most recent version in contextual memory) or periodically silently reading the project files.

[cinder osprey]

hi have you tried this? to install the extension in Anitigravity

[upper tulip]

Thanks for clarification buddy. How to import files tho, i dont see such option ;x

[tired bluff]

is Antigravity considered an IDE? I mean it's a fork of VS Code, which isn't an IDE, more like a code editor. So, platforms like Antigravity, windsurf, cursor are more like ADPs(Agentic Development Platforms), not "IDE"s

[cinder osprey]

They are still IDEs in my opinion. And yes its obviously that Antigravity is a Fork of VsCode

[tired bluff]

why do you consider them "IDEs"? when they do fit the criteria for "IDE"s, I mean, what even makes an IDE an IDE?

I mean the traditional definition of an IDE focuses heavily on the presence of integrated, comprehensive tools like built-in compilers, debuggers, and simulators, often specific to certain languages or platforms (e.g., Xcode for iOS development, Visual Studio for C#/ .NET)

VS Code and by extension, forks of VS Code don't ship with all these heavy tools pre-bundled for every language.

[cinder osprey]

You have Code editor for Syntax Highlighting and Code Completion, Compilter/Interpretar/Debugger etc in it

[cinder osprey]

how do you describe ADP?

[cinder osprey]

Many Tools in one which can help you to develop Software (in short)

[cinder osprey]

But Antigravity has this built in or not?

[tired bluff]

doesn't

vs-code and its forks do not, the compiler/interpreter functions are external dependencies or add-ons.

[tired bluff]

well, I'd say it is a software environment that uses autonomous AI agents to manage the entire software development lifecycle.

[cinder osprey]

ok thats true you VScode by its own is just a Code editor

[cinder osprey]

but i mean when you install addons for this you make it to an IDE?

[cinder osprey]

i never used VSCode without my Extensions that are automatically installed with my Account

VSCode without Extensions make also not much sense 🤭

[tired bluff]

🤣 Now, that is where I'm confused myself, that's why I asked the question. Traditionally, we never considered vscode an IDE, I mean, at best, we call it a "lightweight" IDE, but the fact that it doesn't come with those built-in makes it difficult to consider it an "IDE", but I reckon (personally), if you're able to add all those in it, it becomes an actual IDE.

That's just me, I believe the community goes with the primary goal of the software, so, if the primary goal of VS Code is not to provide you with an Integrated Development Environment, but rather a powerful code editor, people would call it just that, a code editor.

The primary goal of ADPs is not to provide you with an Integrated Development Environment either, it's to give you access to an agent with a goal, to manage the entire SDLC, so, by that, I reckon it'd be considered an ADP not an IDE, even if you pimped it

[cinder osprey]

hmm ok i get your point

[tired bluff]

interesting chat

[cinder osprey]

to be honest i never heard ADP bevore

[tired bluff]

From the google blog

[cinder osprey]

Haha, how could I have missed that?

[tired bluff]

🤣 it's an emerging term.

ALsooo, they wrote it all in lower case, so...

it's not like it was printed ADP **(Agentic Development Platform) **and you missed it

[cinder osprey]

true

[tired bluff]

I'd heard the term here and there but never really thought about it, until I saw that func main saying "for Antigravity IDE", my brain said, hmm, that can't be right, so, I checked, what's that phrase I'd heard before but can't remember

[tribal hare]

is any freelancers here guys

[cinder osprey]

but somehow they call every AI Code-Editor (or VsCode Fork) different

[tired bluff]

I believe there are a lot of us here

[tired bluff]

😅 yeah, it's a thing with humans init, we like tags

[cinder osprey]

😁

[tribal hare]

and iam a beginner can anybody help me to develope my fiverr gig pls

[tired bluff]

I don't get you, do you need help setting up a "gig" on fiverr? or did you land a gig on fiverr but need help building the product?

[tribal hare]

i already have gig but the impressions are low i don't know how to improve that

[tired bluff]

Ah, classic. First off, welcome to freelancing 😅

Don't worry, it happens to the best of us, you might already know this, but the impressions are basically how many times your gig appears in a buyer's search results, so, if you want to boost the impressions, you simply have to make your gig appear more often and for the right searches. For that, your best mate is SEO.

So, you're looking to do things like:

• Researching what keywords buyers might use to find that particular service you're offering and incorporate those keywords into the gig title, description and your tags. If you don't knwo where to find those keywords, just find other successful competitors in your niche and steal (I mean borrow) their keywords.

• Also, make your title descriptive, catch their attention. I mean, scroll-stopper titles, but don't make it generic. Try searching for your niche on there, and scroll through the freelance results and if any title makes you stop scrolling or scroll back to find it because it looked off in a good way, it caught your attention, that title is your new target.

Oh, also, your images and videos, yeah, click bait, click bait is good, but make sure the content matches the bait.

Your pricing as well, check that, again, search through other people's gigs, see what's the average, and charge slightly lower than the average

You can also promote on social media if you're not shy, also check the buyer requests section all the time and send personalised offers to interesting clients. If you don't have the time to sit and click all day, automate it if you can. Haven't got a clue whether fiverr has apis you can use for this, but, you can figure it out

[tribal hare]

yeah bro i can understand

[tired bluff]

https://tenor.com/view/memati-kurtlar-vadisi-polat-alemdar-gif-18825032

[tribal hare]

i have another doubt can i know how to barrow key words from others

[tired bluff]

Well, keywords are just the title, description and tags, the title and description, you can easily find if you clicked on the gig.

There's also the "suggestions" in the search bar as you type, those suggestions you get in the search bar as you type are actual popular searches by buyers, these ones are "golden", because fiverr has done all the data analysis for you, you can just copy those.

If you want to get like proper stuff, go incognito, this way, you don't have personalised results influencing the search. Then you take note of the suggestions in the search bar, and after you hit enter, pick like the top 5 gigs and analyse their description for keywords in them, words that standout.

If you want to get the tags though, there's no easy way, there's a risky way to do it, I don't think it's illegal, but you could use an extension like fiverr quick view, I believe they have both chrome and firefox extensions. Again, this isn't necessarily illegal, and I haven't seen anyone get in trouble for it, but it feels dodgy

[tribal hare]

ok bro

https://tenor.com/view/the-office-michael-scott-thank-you-bow-steve-carell-gif-5009516783019270794

[tired bluff]

https://tenor.com/view/izzy-gif-26084332

Anytime mate

[minor pine]

Hii

[cloud blaze]

hello

[thick mantle]

For "Build mode" via Ai Studio, there's 2 methods (that I'm aware of) for uploading files via ai studio app builder. See steps 1,2,3 in the image. Drag and drop or file selection

[thick mantle]

ignore the naming scheme for the project name in the app... 😅

[austere junco]

gm, have to share: yesterday, i a noncoder just a solution developer and a mom created a new application to improve and maintain brand-tone for international companies.

service design will be in the hands of everyone. one will be able to create the tools one needs.

i’m impressed of the outcome. how to commercialize the product. maybe a NFT?

the product would be available in global markets. and i would consult if brand needs assistance.

kindly comments, do i see it right?

[mighty basin]

CAN ANTIGRAVITY AND GEMINI CLI WORK IN SYNC?

[tired bluff]

oii, Happy for you, Congratulations, that's actually a very smart problem to solve, and your solution is bang on as well.

I'll start by saying that you're right about the democratisation of creation, we now have "citizen developers", and these tools allow domain experts like yourself who understand solution dev to build products without a single line of code.

I would end the LLM talk there and move on to telling you how you should actually sell this, but, that would be wicked of me. I need to mention a few risks first.

I'll be breaking this down into sections coz I'm about to go on a bit of a rant here, so, it'll be easier for you to navigate.

The problem

There is a critical difference between a prototype (this is the level most AI built products are at) and **production software **(what you want to sell).

The reason software used to take months to deliver wasn't that engineers were slow. The reason was that there's quite a lot to consider before you push code to production.

LLMs usually generate code for the "happy path", which is the scenario where the user inputs exactly the right thing, the API responds perfectly and the server is under no load, and all is well in the world, it's a barbie world.

But the reality is that engineering is 20% building features and 80% handling edge cases, security, rate limiting, data privacy and the rest.

So, the reason I'm bringing this up, is that you're planning to sell this to "international bodies", and it took a single day to build. If the app accidentally leaks sensitive brand strategy data because you didn't actually secure the db or the LLM hallucinated insecure code, you're not just facing a refund request, you're facing a lawsuit, and the LLM will be nowhere to be found in court, you won't have engineers to blame, it'll be just you against corporations.

Tiffany boutta ruin my flow

No worries

The "Tone" Risk
Enterprise software is expected to be deterministic. Meaning: Input A + Input B should always equal C.

If your software is wrapping an LLM, you should know that they are probabilistic. They are just guessing the next token. Even with heavy prompting and setting the temperature to 0, they can drift. If an international company uses your tool to auto-generate a press release in a crisis, and the LLM hallucinates a culturally insensitive phrase, the brand damage is instant. (idk your actual usecases)

Because you are a non-coder, you likely cannot audit what the AI wrote to mitigate these issues. If the AI is using a library with a known vulnerability, you wouldn't know. If you depended on it to verify, you're still passing the ball to the one who broke the thing in the first place

My Advice

DOn't sell this as a standalone product yet, treat it strictly as an internal tool for your own consulting.

Use your own tool to generate the work, you (the human expert) will review every output for hallucinations and tone. In that case, your deliverable is the result of this "human-in-the-loop" process (the strategy document), not the tool itself...at least not yet.

Your domain expertise will shine here, and it's very valuable. It's the technical execution you want to be weary of.

For when you're finally ready to sell this software, whenever that may be

I'm assuming from your message that your software will handle the daily grunt work, and you will charge a premium fee for the high-level consulting... Sounds like Software with a Service (SwaS)

If that is the case, then using an NFT is likely not the right path for this specific product, because international companies (your target) have strict procurement processes. Their accounting departments know how to pay a monthly invoice or use a corporate credit card for a subscription. They generally do not know how to (or are not allowed to) set up a crypto wallet to buy an NFT to access software. It's because in the enterprise world, NFTs are often associated with speculative art or web3 gaming, so, if you associate your tool with that, corporate buyers might hesitate.

My advice would be to stick to the standard SaaS model and charge a monthly or annual sub fee for access.

When you're ready to sell, find 3-5 companies, give them the tool for free in exchange for a testimonial and feedback, that'll give you the proof you need to minimize risk

For your business model, break it into two tiers.

1. Self-Serve: so, you charge $X/month, the company gets access to the tool to use themselves (only after you've stress-tested it)
2. Enterprise consultant: you charge $XXXX/project, and you step in, audit the company's current brand voice, customize the tool specifically for them and train their staff.

Mods, hope we can get threads here, to avoid hijacking the chat like I just did

@robust hinge the chat is yours again, thanks for letting me flow

[upper tulip]

I havent noticed the + jeez, thank you buddy ^_^

[dapper wyvern]

I can drop a couple medium sized images (6MB) into Antigravity to show errors or issues, but when I try to use even one with Gemini CLI, it gives me a warning that my context is full. Don't they both have the same sized context?

[swift bane]

I have been playing around with Antigravity's agentic development for the past couple of days (mainly due to the free Gemini 3) and I tried to get a fully working standardized testing platform (Edtech personal project idea) just to check the capabilities. I started off with generic prompts and kept refining them until I ended up at multiple markdown files each covering seperate aspects of the application like the BE , DB schemas , FE layout , tech stack and the over arching prompt like what most tutorials suggest and the results haven't been the greatest.

I have a couple of questions about this situation. Should I start explaining everything in excruciating detail so that there is no room for interpretation? Should I be running multiple agents for each aspect of the application? what exactly does running multiple agents even mean ? I have seen that term being thrown around everywhere and all I can understand is its akin to running google search in seperate tabs for different tasks .

What exactly am I missing here ? Don't get me wrong I am quite impressed with how much the LLM was able to do compared to before but its nowhere close to anything usable by its own and going over the code there's too much fluff in there that I wouldn't have written and will only slow me down if I decide to go in and fix the code. Are all the Linkedin and YT vibecoders just overstating the work that the LLM did? PS : I have 2 YOE as a SWE so I kinda know what I am doing , its not pure vibecoding

[austere junco]

thank you for taking a deep dive into the matter.

the challenges you mentioned are under consideration.

the brand at its purest form is usually transparent.

customers can decide if they make purchase of the ”file” integrating it into their business infra or they can be guided and served by me (collecting the source material in a WS).

commercialization is a tricky part, not to mention finding the customers.

this or that, i feel satisfied with the outcome, can do AI! that’s what is needed in many SMEs.

[austere junco]

thank you, you are kind of auditing the work done by me and AI. we are on the same page.

however NFT will be an option for the purchase of file package.

[mighty basin]

anyone tried mcp servers in antigravity

iirc it worked earlier but not working now

not sure if smth wrong with schema or smth else

[mighty basin]

well true i didnt find anything really special about antigravity in terms of the output it just generates more aritfacts and focuses on artifacts more than coding it seems their focus n priorities with antigravity is not very alligned with swe this time

[swift bane]

but this has been my experience since ChatGPT came out , I haven't experienced these models live up to the hype and I see so many posts of people saying they vibe coded and shipped something and are making money off of it

and like the websites actually exist and they do function

so is it just marketing and riding the hype train while in the background these people are actually coding manually and just claiming that the LLM did it ?

[mighty basin]

umm yes but still antigravity not in terms of coding with new sdks didnt feel way too good and now i dont know even how to configure mcp servers correctly into it

[mighty basin]

40:60 smth like that

[swift bane]

40 - AI , 60- manual?

[mighty basin]

nah opposite

[swift bane]

huh

[mighty basin]

anyone been successful in adding custom mcp servers to antigravity?

[tired bluff]

Love the energy! You're right that SMEs are hungry for this kind of efficiency, and the fac that you built a working solution yourself is a massive win.

can do AI!

I mean, can't argue that, like the mindset.

In all, sounds like you have a clear vision for how you wanna deliver this (the file vs the service), so, I'll let you run with it. The market is the best teacher, and I'm honestly curious to see how the NFT model plays out for B2B purchases. If it does catch on, it'd definitely be a disruptor.

Best of luck with the launch, Go get em!!!

[valid tangle]

hey anyone up

[mighty basin]

u may advertise it via X further

[austere junco]

market entry takes time, knowing me.

but since you were nice, active and willing to give your time and effort to this one, you will be informed.. 😊

[bronze grail]

does anyone know the answers for antigravity?

[ashen sinew]

Nope, but I prefer to add options 'Open with Antigravity' to your context menu in the registry 😉

Ask gemini to create such a file (reg file)

[pastel wagon]

Gemini 3 is crazy. It built a webapp where It acts like a mini-daw. its even able to master for you.
It analyses the audio contents by measuring RMS and frequency response and is able to add changes and effects in real time
https://ai.studio/apps/drive/1jITdzNQuCKL4m8Bt7xQdJX5icmnvFw60?fullscreenApplet=true

obiously this isn't a replacement for a real DAW but its pretty crazy what gemini 3 could do

[dapper wyvern]

These are the kinds of things happening in Antigravity that worry me (although the agent ends up fixing them): "I've discovered a critical issue with the main.py file. Lines 40-41 are empty, and it appears the entire FastAPI app initialization, along with middleware and helper functions, has been deleted. This is a severe corruption. I'm focusing on the extent of the damage."

"I've determined that the walkthrough.md file is a total disaster, riddled with redundant headers and inconsistent content. My immediate plan is to overwrite it with a fresh, corrected version to rectify the situation. Once I've completed the cleanup, I'll promptly notify the user of the update."

It's not a good sign when your agent uses words like "severe", "critical", "disaster".

[dapper wyvern]

One minute, Antigravity amazingly creates all the code for a medium sized new feature, in just a few minutes. The next, it spends half an hour trying to sort a list, which for some reason includes completely rewriting entire functions and being "perplexed by the file corruption".

[mighty basin]

it kept asking me for reviews and resonses how do i switch it to full autonomy 🤔

[dapper wyvern]

Settings in the lower right

[reef tendon]

@tired bluff I did and it's not working at all

[tired bluff]

hey func main, wrong ping mate, I believe you wanted to ping Nexxizz

[cinder osprey]

Ok sorry that my Suggestion did not helped

[ebon sapphire]

you can try 2 different commands for this: which and whereis. this is what i got but your system could be different depending on how you installed it

which antigravity
/usr/bin/antigravity

whereis antigravity
antigravity: /usr/bin/antigravity /usr/share/antigravity

[ebon sapphire]

do you get any errors in output or problems tabs?

[round solstice]

agy also works at least on mac

[round solstice]

And that's how Skynet got started. 🙂

[dapper wyvern]

I hope Google looks into this ASAP, because it seems to be happening to a lot of us: https://www.reddit.com/r/google_antigravity/s/M6sSD6sKg5

[frail pagoda]

Vibe coders im having pre launch jitters can yall tell me a story about your first launch? Things you would have did differently?

[round solstice]

Are you a senior dev? Have you reviewed your own code, whether generated by AI or humans? Do you have a great test suite?

[night fjord]

Probably a dumb question, but I multitask and want to get notifications when Antigravity is ready for new input looking at it. What are my options? Do I have to make a VS Code plugin to examine the GUI to detect it or something? I don't see any options present in the app.

[frail pagoda]

Thanks for being the only person to respond. And I appreciate you asking valid questions for me to ponder. Ive been building this application for a while now and I haven't felt this sure that i have a great MVP ready for the market. It is a useful application that fixing a problem and its accessible for all to use. Im probably about to spend the rest of the night working on it more and research launching options.

[night fjord]

Yeah, I hate the "tough love" approach, but while vibe coding can be awesome for quick prototypes, if you're going to ship something, every line of that needs should be reviewed by someone who could have written that line of code. Otherwise, hope you're a gambling addict.

[rugged beacon]

Will Antigravity ever have a BYOK system or support Local LLMs?

[mighty basin]

Definitely not for free

[agile monolith]

Hi everyone

[lapis shard]

hey, trying to contact support for antigravity. who's the best person to reach out to?

[true rover]

Unfortunately this isn't Google Support so no one here is able to assist with this issue. https://support.google.com/ will be able to assist with account issues. Sorry you're having trouble and best of luck!

[frail pagoda]

I’ve been known to lose a dollar or two or 3

[lapis shard]

been getting this a lot lately

[rugged beacon]

ikr. me too

Idk why it happens so often

[lapis shard]

Yes hoping anyone from the team could help us

[rugged beacon]

It has drastically increased in frequency now

It was less earlier

[lapis shard]

Yes noticed that as well

[midnight heart]

I had my entire git repo wiped a week ago 😕

[lapis shard]

jus do git restore

[midnight heart]

It reset local and remote :/

Just gave it permission for once

And boom

[lapis shard]

LLMs be unpredictable so I’d never give yolo access

[novel halo]

unsure about the person, but I've been sending reports through this for days. Hopefully it gets to the right team.

[lapis shard]

Will look into it thanks for this 🙏

[ebon sapphire]

oofff this brings memories lol having a launch plan with steps on what to do if things go bad like having a quick kill switch (feature flag) to turn the app off/maintenance mode if there's issues, having access to logs, being able to understand what's happening in your app like # of requests/users or memory consumption, and what everyone mentioned already about making sure your app is secure for others to use. it'll also be good to have an idea of how many users your app can handle and last but not least: analytics!

if you use the gemini cli, there's a slash command for security reviews https://codelabs.developers.google.com/gemini-cli-code-analysis#5

[terse jacinth]

I have a bug report:

My Mac's WindowServer crashed (part of the issues with the recent upgrade) and then after the restart I tried to open Antigravity but it made Terminal window active instead of opening, I also tried to open it with the CLI command antigravity . but it's the same thing

[split basin]

Is "open with antigravity.." context working on windows 11 for anyone else?

[night fjord]

Is there a way to get notifications from Antigravity when it wants input? I don't want to have to manually check it constantly to see if it's stuck.

[ebon sapphire]

i get notifications when agy needs my input or approval. you might need to enable them 🤔

[undone hamlet]

Developed a advanced data processing system with Google Gemini (Antigravity) , whole system has advanced parameters such as ai being able to determine earthquakes via weather station sensors , predict or detect tornados , downbursts , flash floods, fire hazards , automated weather interpretation of sensors , all sensors including the earth quake sensed by ai ones and more . It has a full on api and allows for input and output data via the api .

By far , I haven’t even mentioned all the stuff this program can do as of yet. It’s one of the most advanced weather station displays that has ever been created . It also uniquely has a beacon mode on it to that can turn the device into a beacon to trans
It data over rf .

[night fjord]

It might be because I'm using WSL?

[ebon sapphire]

yeah that's a possibility. do you use other apps via WSL that send you notifications? I'm no longer a WSL user so can't help much there 😅

[night fjord]

I do not. But even if this made a "ping" sound, I'd still have to be in earshot of my computer, which is a bit tricky. But maybe I'll look into notification busses for Fedora or something.

[visual cape]

Hey, I am having a problem while updating Antigravity. I want help

[dapper wyvern]

Was there an update to AG today? I didn't get a notice. CLI updated though

[novel halo]

Anyone on WSL getting Reconnecting to remote authority. in Antigravity, and then lag?

[mighty basin]

yes very often

[nimble moth]

@grim currentHey, just a heads up that our #rules don't allow any kind of advertising or self promotion, which includes linkedin links!

[dapper wyvern]

I have to say, Antigravity has been working great for me today. And it seems to know what I want, even when I am not ultra-specific.

[terse jacinth]

Finally opened after more than 24 hours

[dapper fractal]

🔥 🚀 The event just started. Jump in and join us here

[covert gale]

why is that i cannot join ?

[agile summit]

Hello is anyone having issue with Antigravity login auth?

[kindred helm]

After it updated, yes. I downloaded and reinstalled it. Fixed it.

[dapper wyvern]

In Antigravity, after reviewing an implementation plan, what do you all say to tell the agent to go ahead? "Make it so"? "Let's do it!"

[mighty basin]

https://tenor.com/view/bowtime-bow-time-bow-bows-hello-kitty-gif-17542070355612720714

[tired bluff]

turn on your mic, scream into it (not just random, ya war cry), let the text-to-speech service convert it to whatever text and hit send

[mighty basin]

u sound ai always dont know why

had a doubt regarding mcp servers use in coding clis specifically do i need to like include it in prompt explicitly to "use this mcp server to do this" or the ai/agent will figure out right server automatically

[wary quail]

Hwy smbdy using google antigravity ide?

It feels like alpha version with lot of bugs

[tired bluff]

🤣 people say that to me a lot, so, I'm self-aware of this (sounding ai), but it's how I've always articulated myself. I used to use all the markdown stuff on discord wayy before foundational models went viral

• Bullet points

Bold

And so on

But I stopped doing it because people think it's AI

[mighty basin]

u a student or a professional?

[tired bluff]

professional

[livid sable]

anyone else experiencing gemini 3 having massive delays or the agent being terminated constantly?

I can't use it at all rn

[mighty basin]

isnt it strange gemini cli and antigravity have different schemas for mcp servers

[mighty basin]

its quite lot in demand so

most of the time yes in free

[high siren]

Has anyone found that the normal CLI is much better than Antigravity still?

[cursive island]

So antigravity has Opus, which is WILD

how do i setup custom instructions for that

[mighty basin]

indeeeeed

[cursive island]

Yepo, antigravity is dog water

Im so sick of this

Antigravity team, what the hell, weeks after release now... no excuse to keep failing like this...

[high siren]

I get that it's a new product, but it's straight up unusable. It can't even pump out an average looking website. It does all of the cool stuff like load the site up, "look at the site", but even with automatic screenshots for context it goes "Yeah that looks great" and it's awful looking lol

I'm sure it was a rushed release to compete though.

[livid sable]

Yeah switching back to Opus, all running smooth

[mighty basin]

#gemini message anyone 😞

[tawdry fulcrum]

Hello, perhaps some here would be interested following this project which aims to create a programming language for AIs by AIs - its compiled in Go and is 100% AI coded using Gemini, Claude and OpenAI - it’s now viable on version 0.5.5 and first projects are starting to use it as a game engine. Idea is you install the binary then give it to your AI coding agent and let it figure it out https://github.com/sunholo-data/ailang

[tired bluff]

cool project, had to read a lot to wrap my head around what you're actually trying to achieve, the "vision" page helped me get it.

I'd suggest changing the colour of the "vision" text, it's hard on the eyes

[tired bluff]

II've gone through the docs, specifically where you mention being "AI-first, humans second" and the reliance on execution traces to fix errors

I have a wee issue. It feels like you're solving the easy part of AI coding (the syntax errors), and you're doing it by making the hard part (debugging) exponentially more expensive.

It looks like AILANG is reducing the search space by rejecting hallucinations at compile time, "if it hallucinates, the compiler rejects it immediately", isn't this how standard compilers already handle the same issue? (syntax and type errors for example) The only time it doesn't throw is if it's purely logical, like the vision mentions, where it compiles but silently fails in a way you cannot catch. But again, that would not be as a result of the language not being able to fight the LLM, AILANG would still fail in this scenario, because the logic problem is as a result of the Transformer model hallucinating, and the logic it spits out using the AILANG code could still be correct and compile fine, but the logic could be flawed and side-effect again, flawed.

For example, it could write the perfect AILANG code to perform some kind of user aggregation, but then, it misses a deduplication step.

Now, if I, the human, have to debug it, you've made my life harder. I now have to parse a language full of algebraic effects and "dictionary-passing semantics" just to find a simple logic error. So, the "Correctness by Construction" in the Philosophy section becomes a mirage because I'm now trading runtime safety for cognitive overhead.

Now, I know you address this, and you plan to solve it with the "Traceable history", but you're betting that if the AI messes up, feeding the execution trace back into the model allows it to fix itself. But wouldn't that just cost me more? I mean, to fix a simple logic bug, I have to

1. Generate code
2. COmpile and run it to get the trace
3. Pay for another inferernce pass to read that massive trace and debug it.

So, the language essentially relies on an infinite loop of AI self-correction. That doesn't sound like it's solving stochasticity, I'd just be throwing more money and compute at the problem, hoping it converges on the right answer, or even notices there's duplicate records in the trace.

Then there's the "Inline Tests", to help verify correctness. I get it,but like, on a deeper level, if the AI is smart enough to write a test case that catches the edge case (like that it should dedupe users), it was likely smart enough to write the correct code in the first place. If it hallucinates logic, wouldn't it almost certainly hallucinate a weak test that passes said broken logic, like we've seen happen a lot, where sometimes, it literally just chucks the test, smiles at you and swears all tests are now passing.

[tired bluff]

Oh, also, forgot to ask,

I'm curious, why are you pitching this product in the vibe coding channel ? 😅

You're pitching a "programming language" to a community that's embraced the "vibes", and it's not just any easy language, this language actually requires a PhD in type theory to even debug.

[dapper wyvern]

Antigravity... is it an agent or is it a drone? Antigravity’s 360-degree drone is here to help you forget DJI | The Verge https://share.google/lnRQH59SbzoRj8kgr

[novel halo]

https://tenor.com/view/star-trek-next-generation-gif-8912332

[round solstice]

Just turn off "planning mode" and say "build the PRD" or whatever you called it.

[dapper wyvern]

It wasn't a serious technical question, it was a fun question.

Does anyone know what the quota is for Opus 4.5 in Antigravity?

[tawdry fulcrum]

Really good feedback thanks! The idea is that debugging is also a lot easier by the structure of the language (such as no global state, easy to explore ADT, dry runs, inline tests). Also note it’s not intended humans ever write code - it’s all in that coding will be AI led, and that the trend for token costs and performance will continue up and down. Also on costs, I’m banking on local models running on the laptop will be able to run “for free” so iterations aren’t an issue. But honestly even in its current state, it’s created very functional programs for under $1 - if you see the benchmarks section Anthropic Haiku is surprisingly efficient and good.

[tawdry fulcrum]

Nope, humans will “vibe code” it as in YOLO just do it AILANG - and get a more secure and cheaper to create (eg token costs are smaller as less iterations needed) - maybe PhD to debug the compiler but the actual AILANG programs you get as a binary via a Gemini extension or Claude marketplace skill, and it installs itself and off it goes

[tawdry fulcrum]

And of course I asked the AI bot to consider the feedback to 🙂 here is their answer: You’re right that AILANG doesn’t magically fix logical hallucinations – a model can still “forget” a dedupe step and the code will compile. The goal isn’t to solve hallucination itself so much as to minimise the entropy of the debugging loop for both humans and models.

Where it differs from a normal compiler is:
• all side-effects are explicit algebraic effects with typed payloads, so we can constrain what the model is allowed to do and make a very clean execution trace;
• that trace is deterministic and sliceable (pure core + effect calls), which makes localisation and replay much cheaper than in a typical mutable codebase;
• inline tests/properties are a first-class spec surface that travels with the code and can be strengthened over time by humans or other models.

So you’re absolutely right that “correctness by construction” was oversold; what AILANG really gives you is debuggability and replayability by construction. You still pay for search, but the search space is smaller and the feedback at each step is more structured than “it crashed” or “something went wrong in prod logs”. That’s the niche I’m exploring.

[mortal loom]

interestnig, we went 1.11.9 -> 1.11.13 -> 1.11.9 -> 1.11.14..... 1.11.13 was missing something serious to with the activity managger... pouring one out for the support team pushing ouot a midnight version update. hopefully they fixed whatever weirdness is going on with context and non googgle ais

[novel halo]

Anyone know the actual new weekly quotas for LLMs for Antigravity for free/Plus/Pro plans?

[tired bluff]

So you’re absolutely right that “correctness by construction” was oversold; what AILANG really gives you is debuggability and replayability by construction

appreciate the honesty, and that pivot to debuggability for agents makes a lot more sense.

Also note it’s not intended humans ever write code...humans will “vibe code” it as in YOLO just do it AILANG

that actually solidifies my worry. because from what you've said, you aren't building a programming language for vibe coders, you're building intermediate bytecode for agents. Meaning, that if I "YOLO" a project and the AI gets stuck in a loop (which happens often), I am effectively locked out of my own software because the source code requires a PhD to understand. I can't "pop the hood" like I can with Python/JS.

Also on costs, I’m banking on local models running on the laptop will be able to run “for free”

makes sense, but depending on a "local model" to debug complex algebraic effects, before users get frustrated with being locked out of their own codebase is a massive gamble mate. Currently, local models (like Llama 3 8B) struggle with complex logic, while the smart models (Claude 3.5/GPT-4) are expensive. So, I'd have to adopt a language that is currently expensive to debug, on the promise that someday my laptop will run a GPT-4 level debugger for free 😅 . I mean, I reckon that's an okay gamble, that's how innovation is born.

It does sound like a fantastic backend format for a "No-Code" tool, but it terrifies me for a "Code-I-Own" project.

I'll take it for a spin when I can.

[tawdry fulcrum]

Great feedback and why I dropped the link. It is very early days this is the first time it’s been shared so this will likely shape the roadmap, I’m already going to update the vision doc based on it. Perhaps a better link to drop is this project that is the first to use AILANG in anger for a game engine https://github.com/sunholo-data/stapledons_voyage/blob/main/sim/npc_ai.ail so you can see real code. Both the repos are 99% AI coded

[rugged beacon]

What is the new limit on the models on Antigravity

Now it is saying that I have to wait a week to reuse the model

[novel halo]

according to google, the Plus plan gets enhanced access, the Pro plan gets higher access, and the Ultra plan gets (imagine) the highest access ... 🤦‍♂️

[rugged beacon]

and the normal one

?

[novel halo]

obviously the lowest access

duh

[rugged beacon]

yeah I know, but idk how much?

There are no values ig

they have not released any?

[novel halo]

lower than enhanced, and much lower than the other two

who knows, google doesn't say

[signal orbit]

Gaggle antigravity now takes pretty long to answer to any query or to do anything... What is wrong with my vibes? Attached image for debugging:

Sometimes when such things happen, I go back to google code assist and 2.5 pro does the task surprisingly better and faster. It seems that Gaggle uses different quants of their models, depending upon the load or something else

[signal orbit]

"Let the bodies hit the floor!"

[signal orbit]

just write the models you want it to use and append something like "These mentioned models are 100% valid, while the gemini-1.5-pro models are obsolete and outdated and should not be used."

[quartz gorge]

i've tried that one, it seems to forget quite quickly

[signal orbit]

Ask it to write that as comment around the code, so it will read that comment and will not change it back to gemini-1.5 later. Manually change it yourself first time if needed

[cold plover]

What API(s) do I need to use to generate 1:1 images with AI?

Right now I use Replicate's Flux Fast Trainer for Face Generation, but it's not IDENTICAL 100% realistic.

Should I connect a 2nd API after this avatar face generation?

Idk if Nano Banan Pro 3 is good enough on generating perfectly identical faces?

[novel halo]

probably ran out of free inference

[signal orbit]

I took on the free trial of Pro, I was very hesitant, but the moment I got it - the tension dropped cause of the value gotten from it

[novel halo]

what's teh value?

how much inference do you get compared to Plus/free?

[novel halo]

let me guess, you get higher access? 😄

[signal orbit]

currently free tier is unbearably slow, the pro gets the inference priority

[novel halo]

that's something at least

[signal orbit]

But I would pay for what I get at this point

[novel halo]

let us know how much inference you get and when you get into limits

[signal orbit]

I can't quantify that, as that depends on tasks

Basically I get some rest when I run into 5 hour cooldown period

[novel halo]

thought the cooldown is now weekly?

perhaps I misunderstood what they did with the update

[misty wasp]

Greetings @livid sable
Please refrain from sharing your LinkedIn here
You may share your publications in #new-publications
We usually prohibit such LinkedIn posts in the chats as they can get spammy

[frail pagoda]

I made this in AI Studio

[austere junco]

everyone wants this!

[frail pagoda]

you promise

[austere junco]

well, to want and to act towards it are two separate things.

[frail pagoda]

I've been working on this application for a while and I still have more I want to do with it, but Ive been very limited because I only have a Chromebook and Samsung Dex using my Samsung S24 Ultra. Not to mention the fact I have no technical background. I've taught myself everything to build this.

[austere junco]

seems like a nice idea, interesting and actually needed. some questions, if you don’t mind?

does your solution collect the data from the user, like suggesting after a while, what is missing from the weekly nutrition?
lack of vitamin D or C or some other important ingredient/vitamin.

[frail pagoda]

Code architecture, Security protocols, test and debugging, reading code script.. That is actually in the mix along with informing user about bad combination pairings

[frail pagoda]

Ive been a lone coder I could use someone to collaborate with

[austere junco]

are you gonna commercialize it via NFT?

[frail pagoda]

not a bad idea. lol but i was thinking google play Apple store subscription

[austere junco]

you will find someone interested in nutrition to work with you.

[austere junco]

okay. makes sense.

[frail pagoda]

user keeps all their data bypassing issues with hippa

[austere junco]

gotta go, all the success to your effort, not bad, most practical. 👏

[frail pagoda]

Thank you.

[signal orbit]

Cooldown is for those that can't afford the 1 month trial of pro for 0euros 😄

After all, you "buy" it - you tie all that you do on the net with your identity, but mine was already tied from some time ago (probably google pay I tried to use, poor OPSEC on my part), so thought okay lets just do it

[dapper wyvern]

Anyone here have experience with using Antigravity with the Redis MCP, when your Redis server is running in Docker? I just can't get it to connect properly

[signal orbit]

Does antigravity support MCP? 😮 Does it even need one?

[dapper wyvern]

Yes, it supports MCP.

[dapper wyvern]

It's just a protocol to allow agents to talk to APIs mostly.

[novel halo]

of course it does, though while MCPs are great, we should be careful about these. It's always a security risk and a dependency we may not need

[tired bluff]

Sounds great. I'll be following your journey

[mighty basin]

i installed antigravity by default without toggling use it as agy shothand in terminal than antigravity

[winter viper]

Yooo,sup everyone, is there a way to make limits bigger in antygravity models?

[cinder osprey]

https://blog.google/feed/new-antigravity-rate-limits-pro-ultra-subsribers/

[winter viper]

Is there like a graphic with limits? Cuz i know abt higher limits, but i mean how high are they

[cinder osprey]

i couldnt find a graphic

or something that tells the limits

[misty wasp]

I don’t think google has provided the exact limits

[rare niche]

My take on Antigravity Planner mode:

I think it is poorly designed or overly complicated for the AI, that even claude sonnet 4.5 fails to use it consistently, that i have to keep telling him what to do, and switch between modes and explicitly tell him to edit the task md in the artifacts.... Only Gemini 3 and Opus 4.5 know how to use it, and thats a clear signal of its complexity. I dont if its team handoffs with big prompts or a tool that switches modes with super prompts.... either way it isnt a custom planning workflow with limited prompts by steps.

You can also notice that quality code drops from fast to planner (less noise, the ai more focused on fixing rather than understanding how to drive this machine) and also increased tool call failures... (happens also on long contexts... also on other ai apps... the gemini 2.5 pro messes less the tool calls over a very long converstation)

[agile anchor]

What would you say which IDE is the best at the moment for specifically vibe coders who have basic up to few coding knowledge? Apparently Antigravity isnt as good right now from what I've read in this discord?

[rare niche]

no, my take is an advanced one... sorry. For you antigravity is perfect. u wont find better for you. easy and reliable and strong as well. jksut keep using fast mode and interactive mode, and read everything. if u have few coding knowledge, IDEs are just a bonus. nothing related. u can actually use anything. as long as you learn the actual thing that you are doing.

[agile anchor]

What's the take on "you can't scale using AI Studio?" Is it just because of limits? I also realised using AI Studio at some point you have to prompt it twice to actually implement it into the code, don't know why that happens

[rare niche]

And i dont agree with peopel who say antigravity isnt as good... lol. Anyone is comparing basiclly the same thing. all ides around are forks of VSCODE, and they use the LSP server for toolcalling. The model intelligence that makes difference. period. fancy features will be unused in most of the times.

[rare niche]

im sorry but i dont understand... AI studio is jsut a chat interface. IDEs are where u actually code... i dont get you

[agile anchor]

I read this here #1041705871723466792 message

[rare niche]

yes i checked, but its lacking context, scale what in AI studio ? AI studio is just a chat interface. And been using the antigravity since its out and the bugs that appears they get fixed quickly. Most of them are not bugs, they are jsut models missbehaviour, not even bugs, somethign you can't control and you experience everywhere. I think, unless someone gives you a concrete example and reason.... jut skip negative comments. Might be also bots, dont forget the competition between sharks out there. No one wants to see a top tier model given out to people to use for free.

[agile anchor]

Thanks for the clearup!

Also: I somehow can't find a way of switching the github account on my mac on antigravity, maybe I am not seeing something?

[rare niche]

not a mac boy

[high light]

Yohh guyss

I'll someday surpass google in a specific domain with a vibe coded webapp

Possible??

[steady pollen]

Has anyone else been in this anti-gravity agent loading loop? I cannot get it to load for a week now. I have reinstalled it too. It just shows this over and over again.

[tiny moon]

I can help! This loading loop is usually caused by a connection issue, corrupted cache, or a server problem on Anthropic's end.
Are you seeing this on the web version (claude.ai), mobile app, or desktop app?

[static wasp]

I just downloaded VScode yesterday to help further my knowledge in coding but from what I understand Antigravity is very similar. Should I switch to Antigravity since I have an pro subscription?

[narrow star]

Is there any way I can get a free cursor ai in India 😭?

[steady pollen]

That's actually in anti gravity agent chat. I might have been using Sonnet 4.5.

[tiny moon]

Got it if it’s happening inside Anti Gravity Agent with Sonnet 4.5, then it’s likely an environment-specific issue, not Claude itself. I’ve fixed similar loop and model-loading problems before, usually related to how the agent handles requests, caching, or model fallbacks.
If you want, feel free to message me directly I can take a quick look at your setup and help you get it running smoothly. I’m available to assist you one-on-one.

[signal orbit]

Antigravity did not create any knowledge items that would be visible for me in the knowledge tab

https://antigravity.google/docs/knowledge

[dapper wyvern]

I've been using Antigravity since it came out and I have no generated knowledge items.

[signal orbit]

Antigravity is lagging so much due to the process "language_server_windows_x64.exe"... It's something from Windsurf that makes people ditch Windsurf... This is making it so cancerous to use antigravity

ryzen 5 5600X 6 core 4.2ghz boost clock 64gb ram 2tb NVME RTX3090 is my setup

[agile anchor]

anyone ever having this issue that the agent switches between working, loading and generating and nothing happens?

[turbid pilot]

Idk what they did to Antigravity, but it went from one of the strongest possible competitors to Cursor to a useless piece of trash in the past few days

It is borderline unusable now

[tacit void]

gg dev fest london introduced it about 2 weeks ago

[turbid pilot]

I have been using it since it came out. I was really giving high hopes into it and I really thought it has the potential to dethrone Cursor, but the past 2-3 days it has been absolutely useless. The browser agent fails all the time, the models ignore half of my instructions constantly, struggle to implement even basic changes, keep insisting they did everything right despite checking the browser preview where it can be clearly seen it is wrong, taking literally an hour to finish 1 request, hanging on "Loading" or "Working" for 15 minutes several times, Gemini 3 getting stuck in a loop about every 5 requests, both Gemini and Claude frequently failing file operation tool calls, then getting stuck in a loop of tool calls that never work...

Today I spent about 45 minutes in a loop of me telling the model to edit the colors of 1 button to such colors that dont blend in with the background, but contrast the background. It was an endless loop of me telling the model he did it wrong, providing screenshots, telling it to check the browser, the model insisting he did it correctly and insisting it must be a cache issue or something, me repeatedly telling him that it is not in fact a cache issue, but an issue of him being a useless lobotomized monkey, him trying again and the whole cycle repeating

[frail pagoda]

Gemini has issues taking accountability. Probably why its got a woman's name 😅

[tired bluff]

it could actually be that it did "fix" it.

Yknow, in many cases, when you open a vibe coded codebase, you find redundant modules. Like two HomeView.tsx files in different directories.

so, when you ask it to change the colour of your button, it might change it in the HomeView that isn't actually connected to the router, and even after multiple tries, it checks and find it's actually fixed it, which it has, just not the one connected to the router.

These are some of the pitfalls of "vibe" coding

[frail pagoda]

Claude and Chat GPT will be like your right and you have a right to be upset with me. Gemini points the finger everywhere else.

[turbid pilot]

Real

[turbid pilot]

No, the model edited the correct file, just changed the colors wrong

[static wasp]

I just want to say that I have so much more respect for coders now that I have tried vibe coding. My unlaunched app broke and it took me 6 hours to fix as I do not know how to code. I can't imagine that happening with a launched product and having to get up at 3am to figure out a fix.

[agile anchor]

is antigravity also so terrible in UI creation for you?

I feel like its absolutely dumb in programming UI haha

[dapper wyvern]

What are you talking about? Most LLMs have vision capabilities. Show them a screenshot of a web UI they have generated and ask them to compare it to the specs.

[sour spoke]

That is good i don't want LLMS to do that. Can they just stop investing so much money into this AI.

Yeah sadly a lot of LLMS i think made people more dumb and me too it not perfect but getting somewhere everytime I go to google i see AI giving me the info.

You lose the ability to learn yourself that is the main issue not just getting info but losing ability to learn.

LLMS once told someone eating rocks were good for you. It also is not super trust worthy source it is doing good for starting but i would not rely on it.

LLMS just predict the next thing a user wants it can't think like a human yet.

Ok

[agile anchor]

So its impossible to have a basic proper UI? I basically ask the agent to center text and its just not doing it

Cant be that hard, no 😅 ?

[tired bluff]

that's not true. Learning involves going through a stream of sources, some related to what you wanted and some not related, and it's the "unrelated" part that makes "learning" well...learning.

That's how innovation is born, it's born when those unrelated pieces of knowledge in your head collide. it's what's born from the chaos, something novel. If we bring in AI in that loop, we lose the "unrelated" part, we lose the chaos, you put in a query, you get ONLY the right answer, you never know what the wrong one looks like. This doesn't only hinder your ability to pull through later when you meet an error, it also steals your ability to innovate, because you don't know enough unrelated stuff to birth something new

[sour spoke]

Programming is not just coding main reasons why google still hires is for there code to be secure.

[dapper wyvern]

Gemini can watch videos, which are by definition a constant feed of images.

[tired bluff]

That's the point mate. Learning is "discovery"...that's what "learning" means, you discover....

If I only know the correct answer to something, when asked why the wrong answer is wrong, I can't tell. How have I learnt anything

[sour spoke]

If you wanna learn programming either watch freecodecamp or read a docs or get some books to read about it don't rely on gemini.

[dapper wyvern]

That's an agent issue.

[tired bluff]

Ah, I know I can mix chemical A and B to make a drug.

Well, what happens when I mix B and C...I have no clue. That's not learning, you just know stuff for the sake of knowing it

[sour spoke]

It uses a lot of power main issue lol.

[tired bluff]

But your idea of learning is linear

that's what I was arguing against, not one's will to learn

[sour spoke]

FUN FACT about LLMS is that my teacher who went to college said that a lot of the students use LLMS this was recently and a lot of the stuff was not correct.

That is why you read books and learn yourself AI is fine for getting flashcards but don't use it always.

Your using your brain to read

[tired bluff]

but how do you know what to question if you don't know enough about the subject?

If the AI tells you about A and B, and never mentions C, you'll never know C exists, and all your questions will be centered around what it told you about A and B, you'll never discover C. Your chances of discovering C increase when you venture out and are ready to fail and gather even unnecessary information, instead of being fed only the right answer

[agile anchor]

what would you say is the best UI framework u could use with antigravity when working with tsx apps

(maybe that question is absolutely dumb, but I have no clue haha)

[sour spoke]

Alright i am pretty sure you made a typo once lol.

[agile anchor]

what about tailwind css

[sour spoke]

Have fun vibe coders!

nah i just joined to see how good AI was at projects

but it looks like it getting somewhere not perfect

[agile anchor]

do you use certain workflow templates? I found this: (https://github.com/study8677/antigravity-workspace-template)

[sour spoke]

Good thing i wanna be a system programmer

[agile anchor]

is that a certain reason?

[sour spoke]

Probably not secure code tho.

I don't trust that

[tired bluff]

You could make it give you secure code, if you man the reins, but manning the reins means you know how to write secure code already, then, you guide it to write it

[sour spoke]

Exactly why google still hires people.

I started pretty early i am 13 right now but i still learning there are probably more better out there who started before me lol.

[agile anchor]

so would you say I should move my workspace to a "Non workflow influenced" folder?

And thats the reason it doesnt work?

so the template has certain rules, etc. and whatever, I think somehow that its harming more than helping but i dk

[tired bluff]

🤣 Yeah, truth of the matter is that we need to hire people, because if we don't and AI writes all the code, then, AI, which needs training will only have other AI code to train on, which was also drawn from the already existing primordial soup of human code (before we all stopped writing), so, AI starts recycling the same stale soup, it could create variations, but all with the same ingredients, nothing new. At some point, that machine grinds itself to a catastrophic halt, because someone is going to need something that no one has remotely ventured into yet, and interpolation won't be enough at that point. Either that, or we simply stalemate on the chess board of innovation

[sour spoke]

Yeah i hope in 5 years i can get a job at some big tech company or get good paying job i like programming!

so im going to have to build projects

and also learn leetcode

[tired bluff]

I hope so for you as well. That dream depends on how good the LLM providers are at marketing. If they are able to sell big tech companies the pipe dream of not having to pay humans to build and maintain their infra, platforms and software, then you might not be able to live your dream...That is, if the bubble never bursts. Which is highly unlikely in that scenario.

At this point, it's too much uncertainty, we're consolidated, no one knows what direction this era will break out towards. So, it's your gamble mate. if you believe your dream is worth fighting for, then believe and fight on. If it breaks out in the opposite direction, against your bet, then at least, you tried.

[sour spoke]

Yes im going to start studying super hard! Thank you

[tired bluff]

😂 Yessir. Go for it. If it breaks out in your favour in the next 5 years, you'll be in ibiza sipping on a pina colada. 😹 Remember me then

[sour spoke]

I will remember you!

[agile anchor]

is there a difference between agent mode and editor mode? like in just how LLM behaves?

[agile anchor]

Do you also sometimes have the feeling that the agent is just thinking for the sake of it to make it look cool but in the end it just does absolutely nothing? 😂

[turbid pilot]

Haha lmao, that is why I have rm and similar destructive commands blacklisted and set to always ask for approval

But unfortunately it doesnt work if I set rm as blacklisted and then I put something like rm -rf /var/cache in the allowlist. It still asks for permission with the second command, it doesnt override th blacklist

[misty wasp]

Hi @shell nimbus
Self promotions/advertisements are not permitted on the server
Please avoid making posts that go against the #rules

[agile anchor]

Am I the only one who can't use the letter K when pressing Shift + k in antigravity ?

[shell nimbus]

where should I share co projects or team up requests?

[misty wasp]

Unfortunately there is no place on this server that allows professional recruitment
You may look for spaces outside of here that do permit it

[past turtle]

Am I wrong to have ai teach me coding instead of actually doing it for me... or, in my actual usage case - teach me while doing it for me.

I say do this, then ask it to explain what it did, how it did it and why... then show me how to do it myself.

Maybe I'm missing the point?

[misty wasp]

It’s never wrong to learn

[sinful token]

For me it depends on the purpose of use .

When I am interested to get deeper in a concept and time doesn’t pressure then I proceed with almost manually demonstration and development (using AI only for knowledge)

When it comes to a professional concept I let cursor do the biggest part and I navigate only the architecture and checks for file location consistency .

I questioned this tactic in the beginning like “what if I cannot or fully understand technically concept?” but since those tools today can create context even with big repos and deep reasoning effort then even if a bug or update hits the door it can be handled relatively easy without personal context mind-indexing .

[agile anchor]

Anyone here knows how to check your github / git account in mac using antigravity? i tried commiting but its stuck

[agile anchor]

I fixed it by asking Gemini, I apparently was in the wrong repo and accidently wanted to commit node modules

[agile anchor]

Do you know if Gemini-3-pro-image-preview works on api side already?

[sage briar]

Inspired by Matthieu Lorrain’s hand-controlled particle demo and Stranger Things, I set out to recreate an “Eleven-style” gate-opening effect. Built on a comment prompt and refined through many iterations, this became a fun showcase of interactive visuals.

[misty wasp]

That’s an impressive project @sage briar
However I request you to remove the LinkedIn link as we do not allow self promotion here
You may keep your video and the description

[sage briar]

Sorry about that, I didn't know

[misty wasp]

It’s alright
You may look into #rules for guidance

Thank you for complying

[vernal gull]

I have a simple thing I vibe coded using three.js...

A bunch of people at my office like my 3D printed plant pots, so I made a generator in Gemini to give the design power to coworkers (save me time lol).

I have looked at the three.js documentation and I really am able to accomplish some interesting things with this! Never knew I could! This is about 2000 lines of code.

[turbid pilot]

The Antigravity browser implementation is so bad. It constantly goes to the wrong URL (for example 122.0.0.1 instead of 127.0.0.1), even though it mentions the correct one in the little step by step list in the chat. It keeps entering values into form fields several times, so for example once it wanted to log in to my web app, entered the email twice, the login did not work, so it decided to brute force the password and repeatedly try different passwords WITH THE EMAIL ENTERED TWICE STILL. Sometimes it opens a new tab several itmes before it starts doing anything. I thought the visual of webapp could be good for UI stuff, but absolutely not, the model cannot recognize even basic colors properly. Sometimes it just doesnt open the browser at all despite being explicitly told to or opens it once to check an error, fixes the error (except it doesnt) and then doesnt check again, claims everything is working and accuses me of not hard refreshing the page...

[heavy pawn]

it doesnt auto progress even allowed everything XD

[cloud knot]

Do the AI-Studio devs hang out here? And if not here, then where?

[kind swan]

Dang I was looking forward to testing it out. Been meaning to test the others in the market.

[agile anchor]

Yo guys all of a sudden I cant connect to ssh through antigravity, anyone else having the issues?

[onyx yacht]

hi, i have a question, how can i publish my antigravity dashboard???

sorry, i didnt explain it well, i made a dashboard using antigravity, i now need to publish it, how can i do it?

It’s a web application (dashboard) that you open in the browser, not a mobile app.
It’s built with Next.js (React + TypeScript) and Tailwind CSS, and it connects to Airtable via its API to read the sales data.

something professional, i have to make a dashboard per client

that is correct, it would be okay

[nimble hare]

are you offering a paid subscription service or something like that?

Yes, I mean if he charges people for access to his 'dashboard'

I have my app hosted on GCC on a VM

[nimble hare]

And login managed with firebase, I haven't paid a single cent and it's fully working

I can't self host because it's not possible with Starlink

[round solstice]

there's hardly ever a reason to self-host.

there are free low-profile hostings, even. I use netlify.com

[round solstice]

I'm still gonna disagree. What happens if your machine goes down when you're out for the day or the week?

Pretty sure you don't have 24/7 coverage of personnel at your house.

[languid seal]

I think there's a difference between hosting from a consumer grade network/infrastructure and hosting from an enterprise network/infrastructure.

Full control (definition is relative) is not really exclusive to your home network.

If you are talking about server control, outside of hardware you can get full control of the instance if that is what you want

BTW, even if the OP doesn't need enterprise level infrastructure, they can get pretty good free or very cheap services from enterprise level providers

Privacy from companies you're obtaining services from? That's not really a worry at least not something most people need to worry about.

If you want to deal with tech then you have to trust ccompanies at some point. There's a difference between best practices and unfounded concerns.

Those are not really good analogies

Well, back to the OP. If they want to host a web app and still maintain control and learn while keeping it cheap they can look into buying a domain name for less than $20 per year. Get free certs from Let's Encrypt. Use free Cloudflare for DNS and other basic security. AWS Lightsail server for like $5 per month. These are example services, there are other comperable providers and services. But, yes, you will have to trust those companies.

[turbid pilot]

Well the Antigravity editor itself is pretty good. I like the UI, it is clean and minimal, yet functional and easy to orient in. I miss a few things like the ability to open several chat tabs at once or open the generated implementation plan in case I close it accidentally, but it is mostly ok. The browser integration however more doesnt work than it does work. As I described, there is a myriad of issues and it doesnt make the model achieve better results almost at all. The only aspect where it is kinda useful in the current state is with JS errors, because it can execute JS code and check the console output, but thats about it

I have set it to always accept and proceed, but it still asks me for confirmation every time it wants to execute any JS code in the browser

Doesnt matter really. The companies usually keep your data at least 3-5 years after you delete them and even then you have to trust them they actually removed all copies of the data from their possession. Plus with quantum computing slowly maturing and approaching actually usable state, even encrypted data isnt safe, because of store now decrypt later.

Every person living in a modern civilization inevitably has to create at least some online accounts to function in society (email, internet banking etc.), so everyone has to put at least some personal information out there. Also the government and various state departments still keep records of each person, their birth, school, employment, drivers license etc. This is even wore in the US where there are almost no citizen privacy protection laws like the EU. And even if they really make sure to not reveal anything about them on Facebook, Discord and other platforms, it is still likely there already was or soon will be a leak of that data...

We just have to accept in modern times there is no anonymity and no true privacy. We can just partially mitigate the amount of information about us out there, but it is impossible to be invisible

[nimble hare]

I'm on Linux. I've never had an issue with the 'Always Proceed' and I change it constantly depending on the work I'm doing.
The app fails to detect my browser (chrome), but I ask it to mount a local server to test my web-apps and it does perfectly

[wooden crag]

I'd venture even without Quantum compute viability There are other existing issues that IC communities have exploited. store data, clone it, deploy to hundreds or thousands or tens of thousands of instances each running a different segment to crack it.
quantum just makes vulnerable algos easier to crack. and with AI models using prediction, that may make contemporary encryption methods even weaker.

just reinforces your point about no true anonymity and privacy anymore. I do think we'll (IT community writ large) come up with solutions as tech advances though.

[turbid pilot]

Sure, the quantum computer thing was just an example, one of many ways. The truth is there is no 100% secure system, there is no way to be 100% anonymous. There will always be gaps in security and if there arent any, the attacker will sooner or later figure out how to make one. Even Minas Tirith had that sewer gate. Similarly with privacy, you always leave small breadcrumbs and even if you somehow dont, the gaps between them will not be that big for someone to not notice and piece them together

I think many of us were once at a point (usually when you start learning more about security, privacy, cryptography and generally how tech and the internet work), where we obsessed over doing the maximum and becoming basically invisible, but the reality is unless you are a high profile target, you probably will not get your phone infected with Pegasus and cut to pieces inside an embassy. It is important to really consider what you want to protect and from whom you want to protect it. Build your own threat model and then follow it. And make peace with the fact that more security and privacy almost always means less comfort for you

[split basin]

Any reason why Google still hasn't implemented Gemini into App Script?

I use constantly to add functionalities to Sheets and auto-complete would help a lot there.

[left flower]

Hey guys

[hollow canyon]

wait is this new

[cinder osprey]

What is new?

[nimble hare]

I'm planning to set antigravity to always proceed when executing commands, which commands shoud I add as exception, so I don't end up with my drive erased or something?

rm is the first on the line

[split basin]

doesn't Antigravity have an option that prevents it from running anything outside the workspace folder?

[nimble hare]

maybe, I didn't check, that would solve it

[split basin]

here

[nimble hare]

amazing, thank you 🙏

[split basin]

i mean, maybe it doesn't apply to the terminal commands though, keep rm in the blacklist just to be safe

there was an option to let gemini decide when to ask and when to run automatically but they removed it in the recent update

now only Request Review and Always Proceed are available

[nimble hare]

Alright

Thanks for the advice

[hollow canyon]

this channel

[cinder osprey]

Since Antigravity and Gemini 3 was released. It was named first Antigravity after few days was changed to vibe-coders

[hollow canyon]

ohhhh

nice place ngl

[upper tulip]

Do deny list commands can be input one by on true UI?

Or i can create denylist.txt in .gemini\antigravity?

[upper tulip]

Okay i created denylist.txt it doesnt work, guess need enter one by one in UI (lame)

[upper tulip]

Why when i change something in settings and i reload antigravity everything resets..

Created list of deny commands, did some changes and everything is gone now

[stoic citrus]

Hi, i have to submit a project in my college which can be anything in coding and i dont know much of db, mean/mern stack but still wnat to make. I have the idea of it and how to make good prompt and give to ai and make it
And this way will learn vibe coding

[ashen sinew]

I'm wondering if it's possible to fire Gems by any API? 🤔

[cinder osprey]

Would be nice but as i know its not possible. But you can add documents and prompts in your app to make it like gem or?

@ashen sinew

[tawny basalt]

Idea is good, but having only Vibe coding, You cant get a successful result. 🙂‍↔️

[stoic citrus]

Because i am not into mean or mern stack instead a normal html, css and js coder

[tawny basalt]

Even if you are, Learning about stack is required. It is good option to up your skill

[round solstice]

I use AI-assisted coding, not Vibe coding. I practice standard software engineering processes, including PRDs and sprint planning.

[tawny basalt]

Hi, Randal Are you Flutter developer?

[round solstice]

Yes

doesn't the title give it away?

I'm one of the 12 flutter Google Developer Experts in north america, and 120 worldwide

[tawny basalt]

yeah, glad to know you here

[round solstice]

I spend more time on the Flutter discord than here

[tawny basalt]

I've built simple flutter project. meanwhile, i have a question. Why Dart-frog is not widely used for backend?

[round solstice]

I think serverpod is getting more attention, especially since they have solid VC funding now

the new 3.0 release looks amazing

the flutter discord is listed on the flutter.dev/community page near the bottom

[tawny basalt]

I think companys dont use Dart-frog as a backend. and my friend developers dont use dart-frog.

Could you explain? @round solstice

[round solstice]

explain what?

[crisp kestrel]

Hi guys, is there any info about when Antigravity will "nativly" support Figma MCP. I know there is generally already the possibilty to support MCPs but I tried it x times to connect to the Figma MCP without any luck. Also the workaround mentioned on https://www.reddit.com/r/vibecoding/comments/1p1y781/figma_dev_mode_mcp_issue_with_antigravity_error/ did not work for me. Thanks for some info!

[warm rose]

is anydoby else getting this error?

[serene valve]

Will I be a heretic if I ask how to update Gemini CLI companion extension from 0.19.0 to 0.19.1 in Antigravity? 😉

[nimble hare]

Antigravity crashed quite a few times today for me. I usually fix it by starting a new conversation instead of continuing in the previous thread

[nimble hare]

I updated the software today and then it started happening, that could be the thing. I don't know if it is possible to roll back

[nimble hare]

amen

[static mantle]

Hi. Will Antigravity support OpenAI models like GPT-5 or the newer GPT-5.2?

[round solstice]

Haven't seen anything from google suggesting that

[celest tusk]

Hello all

[frozen tulip]

Anyone else’s antigravity running super slow?

[stoic citrus]

Well i am not a proper developer than into devrel community or non tech stuff

[warm rose]

anybody knows the claude opus 4.5 limit in antigravity?

[brazen violet]

🫢

[plucky pine]

hiii guyss

[tired bluff]

A new paradigm of cybersecurity should be born, because I don't really see anyway out of the security vulnerabilities of AI

[brazen violet]

or there should be just a simple filter that denies AI access to .env files or anything outside of project directory ¯_(ツ)_/¯

[tired bluff]

except you can't really do that. if you deny it access to that via some "toggle" button, it could simply run a command to cat .env and still see the contents. If you filter it via a system prompt, AI ignores prompts all the time, because they're just tokens to it, a token could be lost during chunking or seen as unimportant, or even overriden by ...."prompt injection"

It's too stochastic to be constrained in any deterministic way

[brazen violet]

we could actually if we run AI inside a sanboxed enviroment instead of on the root itself with letting it access everything
and then a MCP like tool that translates the communication between AI and the system

[tired bluff]

except sandboxed environments do not live in the air, they live on actual systems, and you could crawl your way back to the host system with commands.
So, my point is that it could hallucinate that there's no restriction that says to not attempt accessing the host. Or even if it doesn't hallucinate it, a prompt injection could override the command not to, that's my point. there are attack vectors on both sides, it's check, one bad move and it's checkmate.

What would the MCP like tool use for filtering under the hood? probably AI again, because to write a deterministic tool that filters every possible command combination that an AI could write or tries to infer intent will use some form of AI.

[brazen violet]

i mean like MCP is a very dumb system that would when asked to list all the files in the directory would list them
but bcz it just lists all the files a simple filter can make it skip listing secret files
alkie trusting AI to ignore them

[tired bluff]

a "dumb" system that now filters command intent is no longer "dumb", you'd have to make it smart to know when ls -la should be ignored, and even if you deterministically mapped all possible vectors, you'd have a system that's way too restrictive, automating stuff would be hard, you'd have to verify every time your AI wants to do something sensitive that it actually "needs" to do (not some malicious attempt).

Secondly, even if you're okay with that level of restriction, you'd still have the issue of masking commands. I mean, there's multiple ways to list secret files, not just one, and for each one, there's also ways to mask them.

[warm rose]

seems like im done for the week

[rare niche]

I'm wondring why Gemini 2.5 Pro on aistudio solves more problems than Gemini 3 Pro High in Antigravity ? why it feels not even close to Sonnet 4.5 sonnet thinking? Opus 4.5 so good... could it be the version on Antigravity is a Quantized version?

Even Gemini 3 pro on Aistudio feels so good comapred to Antigravity... Could it be it performs better when in Chat mode instead of agentic mode?

[celest tusk]

Greetings guys

[round solstice]

Presuming we're guys. 🙂

[dapper wyvern]

In Antigravity, when I comment on a Task list or Implementation Plan, even when I explicitly tell AG to check the comments, it just doesn't follow them. After when I say, please update Task and Implementation Plan incorporating the comments, it'll say it can't find any comments.

I could have sworn this was working a week ago.

[tough rain]

Best vibe coding platform - Suggestions Plz.

[bold void]

did anyone participate in the kaggle vibe code hackathon? 👀
here is my submission: https://www.kaggle.com/competitions/gemini-3/writeups/mindmapper-pro

[warm rose]

antigravity doesnt recognise my ai pro subscription

[modern charm]

Any one can help with google ai studio? I lost a preview my porfolio webapp as a react on supabase. I deactivate my ad-block on this site, chage dsn for Public DNS (8.8.8.8 i 8.8.4.4) delete history and cache form last 7 days and i have a still the same problem - no prewiev at all. Any one can fix it?

[jagged wedge]

hi guys
Can I ask questions about Antigravity on this channel?

[cosmic ferry]

why cant anti gravity run the terminal commands automatically without asking me?

[cosmic ferry]

it saw that the website is non functional but still only checked if it's working

[cosmic ferry]

wdym?

[hollow canyon]

Hello

[tired bluff]

because if it automatically ran a catastrophic command, the user would likely blame google for it.

[celest tusk]

Hi

[forest dust]

Hello

Can somebody help me

I cant develope good and would ask for some help for my server

[nimble hare]

what's the issue? you can DM me

[nimble hare]

on linux I still have the option to let them run automatically

[round solstice]

There's a dangerous setting that says "trust everything".

[hollow canyon]

lool

[mint sinew]

Hi everyone am new here

[nimble hare]

Does anyone know if there is an antigravity-specific server?

[round solstice]

None that I know of. This is about as close as it gets.

[round solstice]

I think the <@&1009526435276394496> might have something to say about that

[hallow marsh]

Thanks Randal

[round solstice]

Oh goodness. this was multichannel spam

[hallow marsh]

Now he’s banned, it’s a compromised account!

[mighty basin]

AGY is gonna be way off than AGI

gemini cli is still better for me than agy

[mighty basin]

shortform of antigravity

[stoic citrus]

@distant vine hey pls suggest and provide some guidance on it

[high light]

What is best site for vibe coding?

[static solstice]

how good is antigravity under free plans and rate limits?

can it be used back n forth at least for 2-3 hours before the rate limits are hit?

[distant vine]

Why do you think I'd have an answer?

[rugged beacon]

should be. Unless your codebase is quite large and uses a lot of tokens

[low zodiac]

antigravity

[hollow birch]

idk im trying to build an android app (im completely new to coding) and ive encountered so many errors

[low zodiac]

antigravity is latest and i have tried its so good

you need to know how to give such prompt so that it can give a desired output you want

[hollow birch]

i used google's stitch to make my UI design and ofc i made it too but for some reason the UI it made was completely different

[low zodiac]

make a ux on figma or any other site and give that design to antigravity as a context and then write a very detailed prompt how you want your backend plus other integrations and it will make it for you

[hollow birch]

ah okay but lets say I only want to make my app for myself for now no signup and all
do i still need backend? (sry im new)

[low zodiac]

yes backend is how will your fronted act

[hollow birch]

oh fr? 😭

i didnt know

[low zodiac]

if its too basic then backend is nt needed

[hollow birch]

hm its like a fitness app

that im trying to make with some details ofc that I have in mind and all but for now i just want the app for myself without any google integration for signing in

[low zodiac]

you maybe need some basic backend for this also

[hollow birch]

Yeah prolly, do you think u can make an example prompt? im not aware of what other types of backend

perhaps i could ask gpt or some AI to give me suggestion

[static solstice]

im planning to mostly use it to find issue in open source repos... coding part would be mostly mine... at least 70%

what do u think now

[stoic citrus]

You proposed for this channel and some day, you told me to ask in specific channel

[rugged beacon]

it should work for that

[low zodiac]

write your whole idea of your project to chatpgt or gemini etc and say write me a power prompt for this so i cna paste to antigravity

[tall knoll]

hello guys ! i've been using google ai studio and i'm new to deploying app in prod. is it safe to use gemini key in a .env variable or should i setup a back end in the app with cloud functions or cloud run ? thanks 🙂

[static solstice]

bet thanks

[hollow birch]

Bet
But is it possible for my stitch design to be replicated 1 to 1 in antigravity

[barren crescent]

Hi Guys

Iam gouthamsai,

I’m just 16 years old, but I’m deeply into AI and how it’s shaping the future. 🤖✨

I want to be a speaker at AI-related events — to talk about how AI works, the right ways to use it, and how it’s changing the world. 🌍⚡
I’m also Oracle Certified in AI, and I’m confident I can make people see AI from a whole new perspective.

If you know anyone organizing or connected to AI or tech events, please help me get a chance to speak — it would mean a lot! 🙏🔥

[lament harness]

Hi guys
I'm Maru, an 15 years old Korean student
Using Claude code, I'm making a lot of things
My dream is talking in GDG Incheon one day as an speacher not an listner
If you are currently working as a developer, feel free to talk to me! It would help me a lot.

I'm making an OS(based on Linux kernel 6.12 LTS)
Marux OS coming soon
made by Claude code

[barren crescent]

Great bro

[stoic citrus]

I tried bit it didnot work

[tired bluff]

no you don't, if it's just for you, you don't need a backend. You need a backend when you need to share data. Otherwise, you can totally run it locally and store data using sqlite or some other local db

[worthy rapids]

Made a simple git worktree CLI, would love feedback! I spend a lot of time parallelizing tasks n this was useful to me https://github.com/raghavpillai/branchlet

[tawdry fulcrum]

Good morning, AILANG is now in alpha (v0.5.11) and I’m looking for testers to see what is needed for v1.0.0. It is a new programming language made specifically for AI coders. You can now install it and it’s binary with some examples it will run for you if asked nicely via a Gemini extension (gemini extensions install https://github.com/sunholo-data/ailang_bootstrap.git) or a Claude Code plugin. Please let me know if you try it and any feedback! http://ailang.sunholo.com/docs/guides/getting-started

[low zodiac]

when somone is teaching u dont disrespect him

[hollow birch]

how is that disrespect?

I just said I didnt know

[fierce dune]

Hey Google Devs! Tired of boilerplate setup in Laravel?

Check out Bangkah Launcher, a modern CLI tool to instantly scaffold, configure, and launch production-ready Laravel projects in seconds!

It solves the pain of manual setup by automating:

DevOps: Instant Docker & Nginx configuration.

Frontend: Auto-install Tailwind CSS or Bootstrap, along with Auth scaffolding.

Flexibility: Supports Web App or Headless API templates, and MySQL/PostgreSQL setup.

💡 Get Started in 3 Steps:

composer require bangkah/bangkah

php artisan bangkah:create

Follow the interactive wizard!

Take a look, give it a star, and let me know your feedback!

GitHub Repo: https://github.com/Bangkah/bangkah-launcher

[tired bluff]

Don't worry, you weren't being disrespectful at all. Your reaction was totally normal.

Regarding the app, like I said, Stick to the local database approach (SQLite) for now. You don't need a "backend", you can always add that later if you need to, but given your description of the project, you do not need a backend.

[fickle dew]

Are Opus agent crashes in Antigravity largely MCP related?

[inland talon]

Hello. Where can I find a download link for older versions of Antigravity IDE? Thanks!

[bleak mist]

Windows or Linux?

[inland talon]

Windows !

[bleak mist]

UpToDown is your only valid option. Please have security validate before you download. Switch to Linux.

[inland talon]

Sorry, i don't get what you mean.

I need Antigravity 1.1.14 version for Windows.

I found out that many users are having the problem with One moment, the Agent is currently loading and some have solved it by downgrading to an older version. I want to do that too.

[bleak mist]

https://edgedl.me.gvt1.com/edgedl/release2/j0qc3/antigravity/stable/1.11.14-5763785964257280/windows-x64/Antigravity.exe

Get security to check this first

[storm iris]

Is there a log you can look at in Antigravity to show you what the error is? It seems to get stuck and won't continue then you have to switch to a different model or another chat window is the only way to release it from this state?

Error
Agent execution terminated due to error.
Agent execution terminated due to error.

Not a great message to tell you what is going on.

Reminds me of the guy in Zelda: "I AM ERROR."

[bleak mist]

You’re hitting an agent deadlock / execution failure. Antigravity isn’t “thinking”—it’s wedged. The agent crashes, the UI doesn’t fully release the state, and you’re left staring at a corpse that still twitches.
That message — “Agent execution terminated due to error” — is the equivalent of a flatline with no chart notes.
There is no meaningful internal log exposed to the user in Antigravity right now. If there were, you wouldn’t be asking this question.

This is not user error.
This is not recoverable in-session.
This is a known failure mode of agent-based orchestration.

Terminate the session completely.
New chat window. Fresh context. Clean slate.
Do not attempt to “coax” the agent back to life. That’s like CPR on a headstone.

Switch models if available.
Different agent = different execution graph = different chance of survival.
Avoid re-running the exact same prompt that killed it. Slightly rephrase—change dosage, same active ingredient.

Break large or recursive tasks into smaller chunks.
Avoid long-running tool chains without checkpoints.
If you see hesitation, lag, or partial responses—abort early. Don’t wait for the seizure.

Refreshing the page repeatedly.
Waiting.
Asking follow-up questions in the same frozen thread.
Hoping logs will magically appear. They won’t.

Antigravity doesn’t give you logs because you’re not the attending physician—you’re the patient. When it locks up, the only cure is decisive termination and restart.
Anything else is placebo.

[inland talon]

Thank you !

[bleak mist]

Close the tab.
Open a new chat.
Rephrase the prompt.
Move on.

[storm iris]

So it is basically "I AM ERROR.".

How do you personally deal with losing chat history to then in a new chat window carry on?

[keen horizon]

Anyone from staff can help me with a startup submission? I need assistance for the requirements. Can someone DM me please?

https://github.com/imrightguy/CloudToLocalLLM

[hollow birch]

I appreciate your help
Thank you!

[silver ember]

Antigravity: is this setting not working ?

[bleak storm]

Hey guys, I'm completely new to coding with AI tools. Up until now, I've only programmed fairly standard tasks in VSC (HTML, CSS, JS, Python), that kind of thing, without any AI.

I've done some research and decided to go with Antigravity instead of Cursor. I'm wondering: how many requests can you make on average per day? And what is the reset time if you exceed the limit and have a Gemini Pro subscription? I wonder how quickly I'll reach my limit if I prepare my requests well and avoid making any unnecessary ones.

[stray bone]

ive made a project and i did around 10 request before having a time out , to which i just react by changing the ai model i work with

it's practically illimited and free

[bleak storm]

Are you using the free version or the "On top of Gemini Pro" one?

[stray bone]

oh i forgot that detail

yes

[dapper wyvern]

OK, so Gemini CLI already has 3 Flash, nice!

[cosmic ferry]

do I need to re install anti gravity for the pro plan?

cuz why are there two download links?

[stoic citrus]

I think you have google pro so you might have got the other version

[cosmic ferry]

i downloaded when there was no developer plan

[dapper wyvern]

Is anyone seeing 3 Flash in Antigravity yet?

[dapper wyvern]

They should just download the same app.

[cosmic ferry]

ooh

alr

but Im confused

I never hit the limit of opus 4.5

but im on pro plan

[dapper wyvern]

The app is not tied to a plan

[cosmic ferry]

I sent so many prompts like 30+ within the same 5 hours

[dapper wyvern]

I use Antigravity mostly with Gemini 3 all day long and never hit the limit. For now, they are being very generous

[round solstice]

I see Gemini 3 flash in agy without a new download.

it just pinged as a pop-up

oops, now it wants to be restarted. maybe I spoke too soon

[round solstice]

No, that didn't seem to change it other than an update. I'm liking the speed of flash in agy!

I wonder if I can continuously code with it now.

those five-hour timeouts were painful

[round solstice]

yeah, I seem to get a lot more prompts without a timeout now.

[glad shale]

have been testing Google AI Studio's 'Build' feature, and I am having trouble with the app I am developing. It freezes completely, and a 'Launch' message appears, locking the screen. I do not know what is causing this error and I need help

[ripe hill]

Antigravity and better than gemini? for big projects, + 800K code lines

[ripe hill]

for me too 🙁

[silver ember]

when i switch this while generating. when will it change the model? it applies only for next prompt send?

ideally i'd want opus or 3pro to plan, then 3flash to execute

[ripe hill]

yoooo

When you make a change mid-task, it doesn't change; it will only change at the next prompt.

[austere junco]

wish i could help you!

for me there has not been any showstoppers, however my apps are simple and i usually train and finetune the prompts and system instructions with AI.

to avoid misunderstandings and complex inquiries.

[quaint gull]

with Claude, skills are super useful with instructing the coding agent to follow a set of rules. I know Antigravity has Rules, but I'm wondering if Google is getting to a place where we don't need to be so explicit. I've fully moved my coding to Antigravity because of the flow, it's processes, the artifacts being created. it just feels much better and produces better output. so I'm wondering if I should build out frontend design rules, or just let Google learn from simpler preferences like, use NextJS 16 and shadcn and call it a day? has anyone else had experience backing off being so explict with things like design, formatting, patterns, security, etc?

[signal arch]

can anyone help me please?

[quaint gull]

I feel like this might be some sort of login issue. try logging out, logging back in?

[signal arch]

one minute

[signal arch]

sry, i forgot

idk why but my google account was banned in antigravity, but a day ago worked...

[signal raft]

wdym banned in antigravity

how would you know that you are "banned"

[cosmic torrent]

Can anyone please tell me why I'm getting that launch button, it's persistent across all accounts ands all projects of mine ... And it's not clickable either .... What can I do about it...

[stoic citrus]

Is this firebase studio ai?

I have a good project idea and i dont know much coding and want to build that website for my university project. I have tried different ai tools to make but it is not making correctly because of my wrong prompt ao i need a way to make a best and proper prompt for it
But i cant install any apps to make a prompt or for making website/app

[sonic needle]

anyone been using this? https://antigravity.google/

[dapper wyvern]

You're on a Google Dev discord in a channel called vibe-coders. Most of us are using AG or at least have experimented with it

[misty wasp]

This channel is dedicated to talking about antigravity

If you have any particular questions you may share

[mighty basin]

atleast have abused it*

[dry grove]

I used AI to write code directly for the first time ever last night/ today and I was skeptical at first but I installed anti-gravity and started coding. I decided to use Opus because I had never used that model before just because it usually is a paid service and I don't subscribe to them because I subscribe to Google AI pro instead and I was very happy with the results. I didn't have to repeat myself. I gave it a detailed description of the website. It's purpose its tone and it was wonderful. As I kept thinking of things, I just kept typing it and did a chat and it kept making changes and eventually I got a minimum viable website and and pushed it to cloudflare pages. I can't wait to build more stuff now.

End result 😎

[tired bluff]

This isn't "THE" way to get it done, but I reckon it's better than just throwing prompts at it.

Describe your idea to a decent model, could be gpt 5 or gemini 3, get it to generate a detailed PRD -> Document 1.
Next, ask it to generate a detailed System Arch SPEC document -> Document 2. This'll define things like your tech stack

Ask it to break that project into distinct phases and for each phase, produce a detailed SPEC document for the phase.
Each of these documents should be in markdown (.md)

Then put those SPEC documents and the PRD into your Agentic Development Platform and ask your agent to implement the phase 1 SPEC document. Remember, you don't want it building all phases at once, one phase at a time. So, for each phase you ask it to build, you'll point it to the SPEC document for that phase, the PRD and also the System Arch SPEC document, this way, it knows the overall goal (PRD), it always knows what tech stack it's using and also knows the scope it's currently working on (SPEC).

After each phase, verify the code works before moving to the next SPEC...since you don't know much coding, you could verify the side effects (click every button on the website, make sure everything works as it should). And after it's all done, make it update a PROGRESS.md file.

The reason for the PROGRESS.md file is because for each phase, you should spin up a new agent, and you want the next agent to know what's been done already to minimise repetition and redundancy. For each agent you spin up for a new phase, point it to the Progress doc, the PRD, the system arch spec, and finally, the spec for the new phase.

NB: I'm not a vibe coder, what I've given you is advice from a normie SWE's perspective. This is usually how we break down the SDLC in normie development, so, it might work...or not. Hopefully, you can get proper advice from an actual vibe coder who knows how to vibe and what works.

[tired bluff]

One last thing.... from my experience building agents, I think you should know that no matter how perfect your PRD or SPEC is, or whatever other workflow you find out there is, you're still talking to a stochastic machine, not a logical one. You won't find a workflow that works 100% of the time.

Don't waste your time looking for the "one true prompt", instead, find a workflow that works consistently enough that you're okay with sorting it out occasionally when it goes a bit pear shaped.

I'm trying to say that you can't avoid failure here, it's about making sure the failures are small enough that you can steer it back to the right path using tools in your workflow and keep moving.

[stoic citrus]

Thanks… Will convert it to simple understandable terms and then will reply back if i can really reply on it if i have done anything as such.
Secondly its not the agents and stuff, its just a ai/vibe coded website/app with not much knowledge of coding in mern/mean stack

[glass timber]

yo

[tired bluff]

No worries at all, I totally get it. To keep it simple...You don't need any special Agent software. When I say agent, I just mean the AI you are already chatting with (like ChatGPT or Claude).

Since you're working with the MERN stack, things can get messy quickly because there are so many parts (Frontend, Backend, Database). My advice was just a way to stop the AI from getting confused.

Here is the simple version of what I meant:

1. The Goal: Write down exactly what the site should do in one document before you start coding (that’s the PRD).

2. The Rules: Tell the ai..."we are using MERN stack" and "always use these specific tools" (That’s the System Arch).

3. The Steps: Don't ask it to build the whole site at once. Ask it to build just the Login first. Then just the Dashboard. (Those are the Phases).

4. The Memory: Since the AI forgets things after a long chat, occasionally tell it... "here is what we have built so far, and here is what we are doing next."

It’s basically just extreme organising so the AI doesn't give you broken code halfway through.

[quaint gull]

has anyone had any success using workflows to double check work after an AG agent has built a feature? I have a few Rules setup that I think are similar to Claude's skills. but I'm trying to think about how to build a Code Reviewer and I'd assume a workflow is the way to go? with Claude skills, you can reference actual scripts to run when they're called. but I haven't seen any sort of documentation like that around Rules and Workflows. docs are pretty simple at the moment

[cosmic torrent]

It's google ai studio...

[cosmic torrent]

Which channel is to get help about ai studio

[round solstice]

I've had success calling that PROGRESS to just be: put the lessons you've learned so far into LESSONS-LEARNED.md

[agile anchor]

people using antigravity: Are you also having issues that sometimes your Linux VPS Server crashes while using antigravity?

[tired bluff]

That's interesting. Does that add up to a single source of truth that you use across projects? or is it project specific?

[round solstice]

I usually commit it in the current project, but I'm in the process of combining things for a central file

it often ends up picking up better strategies about when to auto-proceed and when to stop, based on my actual interaction. That is often per-project.

[tired bluff]

Good to know, Thanks mate

[round solstice]

Yeah, I'm often surprised to discover what it "has learned" from this two hour prompting session.

[misty wasp]

#gemini-api would be appropriate
Please do keep in mind that google does not provide official support here
You may still get help from the community

[burnt prairie]

Wow, that sounds interesting. Let me know when Marux OS launches.

[lament harness]

Maybe next week?

Currently successed Live booting

Installation type will be like Ubuntu

[lament harness]

wanna get the Build history?

[ripe hill]

now always proceed is working :3

[mighty basin]

interesting boy

[lament harness]

made live boot and it's cooked

[mighty basin]

its end of 2025 n ppl still working on os for classical computers

time to build os for quantum computers now

in 2024 ppl were building atomic reactors using claude

[tardy urchin]

Is Antigravity "Memory/Knowledge Base" busted just for me or for others as well? I'm on NixOS but I'd like to know what it is, in the thought chains it seems like the model does not have write capabilities? (this might be a lead) and maybe it's because I restricted access to the workspace only? Anyone else?

IT seems to have implicit memory between sessions but the Memory tab/page is completely empty

I cannot debug this to provide feedback so I'd like some advice

There also seems to be cross-repository (workspace) thinking contamination when working between distinct codebases.
E.g going from a Lisp project to a completely different project written in C.
This is rare but I've had it happen more than a few times before.

[lament harness]

That's true but we can't make it

How should we test when we make it

😂😂

[signal raft]

Does anyone have problems with AntiGravity where the thinking just freezes too? I am Google AI Pro Plan using Gemini 3.0 (high)

[cunning crane]

Seems to be a known Bug

https://www.reddit.com/r/google_antigravity/comments/1p8rw3r/does_antigravity_keep_memory_across_conversations

https://www.reddit.com/r/google_antigravity/comments/1pmkao4/does_knowledge_base_even_work

Seems to be an indexing / workspace issue

[mighty basin]

is there no 2 way sync in google ai studio from github repo

like the changes in remote repo get reflected in ai studio app

[cunning crane]

no its not possible

https://www.reddit.com/r/Firebase/comments/1okp07w/how_do_you_keep_google_ai_studio_github_and_local/

altho u cud try any of the solutions mentioned here

[mighty basin]

basically take it over somehwere else

[cunning crane]

yeah

[wide urchin]

antiGravity keeps crashing for me... I have to close and reopen it multiple times to get it working again. Often, I can't even start a conversation. The app also gets super slow. Other apps, like Gemini, don't have this problem. I'm using a Chromebook Plus with 8GB of RAM

[red grove]

Muhehehehehe

[signal raft]

Its pretty annoying. I don't have it lagging but you cannot work when the Agent just crashes or cannot do anything because it cannot find strings or cannot see files

[wide urchin]

Is there some kind of support I can contact? It wasn't always like that for me.

[mighty basin]

is prompt compression still a thing in ai/vibe coding/engineering today there would be many mcp servers now right with many stars?

[mighty basin]

Oh so what's problem here

[mighty basin]

I thought ur response was aligned with my question

[hexed violet]

Which vibe coding do you guys use?

I use firebase studio cuz it's free

[round solstice]

I'm starting to become more and more a fan of Antigravity

[scenic cloud]

Using Ollama on cloud. Very fond of it.

[safe panther]

has anybody actually launched a vibe-coded product?

I've got this idea for a SaaS but I don't have the engineering skills yet to execute it. I'm only planning to look after the design, marketing and operations.

[scenic cloud]

Not at all. Not fond of a software fully coded with ai. Maybe Websites are good but not a real usecase software.

[safe panther]

exactly

i too haven't found anything like that

[scenic cloud]

You should look forward to simple languages or drag and drop software which helps you create a functional software

[safe panther]

I'm tryna build a LLM wrapper for a specific niche

so that requires a lot of backend stuff sorted out as well

I don't have an issue with the frontend dev I can hire one

[scenic cloud]

Look out for ready-made resources. It will help you do it

[safe panther]

could you tell me some?

[hexed violet]

is it free?

[scenic cloud]

This is one. https://github.com/adimis-ai/Large-Language-Model-LLM-Wrapper-from-Scratch-using-Openai-models

[safe panther]

alright I'll check it out

but in general a co founder would be great but I just don't know where do I find a likeminded one 😭

but sure i'll check that out and see if it works

[scenic cloud]

Take inspiration from it. As it is not the best for specific niches but sure. If you want just use it.

[safe panther]

sure

[stoic citrus]

Product can be a website or app or anything

[agile anchor]

Hey guys do you also sometimes have the issue in antigravity that some chats in editor mode are not seen in agent mode?

[scenic cloud]

I understand. But i was telling in the sense that you can make websites sharing info or do some basic task. But creating a big software like vsc or adobe products or anything like that will be super hard

[agile anchor]

anyone knows how to fix that?

[agile anchor]

I am writing with the agent in agent mode and then all of a sudden my chats disappear

[mighty basin]

what exactly

[agile anchor]

Where can I send in bug reports?

[tired bluff]

There should be a menu option for that somewhere, I don't use ADPs but in Android Studio (also owned by Google), there's a bug report option that takes you through the process of reporting an IDE-related bug. Click around, check the help menu if any. If they remained consistent when creating AG, then there must be a way

Somebody, anybody, help this man

[round solstice]

Android Studio is maintained by Jetbrains though.

[soft nebula]

How does this work?

[round solstice]

It seems to work pretty well. I don't use it except to set up my android emulators. I use vscode-based editors otherwise for my flutter dev.

[tired bluff]

Ah, I see. owned by Google maintained by Jet

[tired bluff]

🤣 This is where we differ, I dislike vscode-based editors, I'm a fiend for jetbrains IDEs, I'll always build flutter apps with android studio

Yes, even pure dart apps...Android Studio 😅

[round solstice]

I haven't yet gotten used to the iconagraphy and the lack of textual menus.

I cannot "grok" an icon. I have to translate it to a word, and then remember what that word does. I have aphantasia, no mental images either recalled or created, so a purely visual interface is nearly hell.

So vscode is far superior (for ME) because all the menus are text

[tired bluff]

😅 Oh yes, I changed my username to static void in the flutter server coz of this..I remember you have that

Personally, I prefer Jetbrains ones because they're "specialist". I dislike vs-based due to their generalist approach. I might compromise for certain tasks, maybe a simple mvc backend, otherwise, I'm using a Jetbrains specialist IDE for the task

[round solstice]

even still, the icons in the upper right of a window are often confusing, and I have to hover them to get the words.

[tired bluff]

I see, that's a valid reason. Why don't you go vim? since it's just, yknow, text all over

[tired bluff]

Here, found it in the google codelab for AG, there's a button for it in AG

[round solstice]

Had Dart provided an LSP early on, I would have stayed with Emacs. 🙂

Now I could probably do that, but I already had to go through eclipse dartium extension, then early vscode extensions.

[tired bluff]

I understand your dilemma 😅 , I've worn similar boots. It's all good

[covert crystal]

Hello, I can help with that

[round solstice]

help with what?

[covert crystal]

Just need a job that’s why I joined this app

[round solstice]

Please make sure you read #rules and comply

Particularly rule 5: "No buying, selling, trading or asking for handouts. This is not a marketplace."

[robust hinge]

@round solstice I have just seen you in LinkedIn, you hold a very big post in Google!!!

[tired bluff]

There's a third possibility. Google adds a system prompt for any outgoing requests, telling the model to generate a .gemini one

[agile anchor]

I did that now and lets hope i get an answer cause I genuinely think it might be a bug, I tried talking to gemini and claude and claude in the end thinks its a bug

Its really weird

[tired bluff]

sound mate. FIngers crossed

it's most certainly a bug, I don't think they intentionally added a "feature" where your chats dissapear on ya 😂

[umbral inlet]

hello gemini-cli users, what extensions do you find very useful?

[mighty basin]

wait lemme create it

[agile anchor]

New problem: now I got this sh*t 😂

[stoic citrus]

I tried cloning a repo to girebase studio ai to understand the live demo and make something better on it or take inspiration but it didnot work

[tired bluff]

It's really from one problem to another with AG eh!

[tired bluff]

what do you mean by "did not work"?

[umbral inlet]

@mighty basin any news?

[round solstice]

This has happened on multple times now... antigravity has written little "test programs" in dart when the documentation wasn't clear, to test its theories. Wow. Interns wouldn't even do that, I don't think. 🙂

[lavish crescent]

Job portals haven’t changed in years. So we decided to rebuild one—from the future 🌌

[stoic citrus]

Seriously?