#vibe-coders

1 messages Ā· Page 5 of 1

urban glacier
#

AGI

red vessel
winter viper
#

nah man wtf i did only like 20 questions today and more than 50 yersterday

image.png
south turret
#

They try to push users towards the 200$ a month plan (for 3 months) then 400$ a month

winter viper
south turret
#

Did not get this issue yet

#

I'm looking into open weight models because of 3.1

#

Hesitate to cancel my AG sub

#

I could cancel AG and instead support an open weight model

#

I think I'll wait for the next version of GLM or Kimi which will probably release soon at the speed they catch on

#

And then stop Gemini entirely

silver ember
#

ah yes the new model quota tab
edit: only now noticed the day is feb 25th šŸ¤Æ

image.png
winter viper
silver ember
#

settings

#

but what's the point if its not working

winter viper
#

for me it's working fine

image.png
silver ember
#

i am on linux :/

winter viper
#

ohh

#

like wsl or full linux?

south turret
#

Stupid aah google

#

Aight to hell with it I'm supporting a chinese model

silver ember
#

full ubuntu

#

GLM5 aint too bad

#

especially for open source

south turret
#

Yeah that's what I'm considering rn, I'm doing a research with Gemini to see if it finds something even better

#

DeepSeek V3.2 Speciale what is that

#

"DeepSeek V3.2 Speciale is a high-compute variant of the DeepSeek V3.2 model, optimized for maximum reasoning and agentic performance, surpassing GPT-5 in difficult reasoning tasks"
oh?

#

According to swebench MiniMax M2.5 is at on par with opus 4.5, what

#

I'll test MiniMax to see

#

Found a Canadian company to compute MiniMax, look at their register button lmao

Recording_2026-02-20_at_16.49.22.gif
#

It's unique for sure

limber arch
#

leagues above gemini

#

the only issue is with the subscriptions and limits

#

i wish qwen made a cli subscription

#

like i said previously GLM has always been just a claude distill

#

and the best one at that

#

claude distill on deepseek arch

#

doing an amazing job can be seen in both usage and benchmarks and flow

south turret
south turret
#

Your new model Google... no wonder the limits are so low
I did something VERY SIMPLE, and it has been thinking for over 3 minutes, hasn't done anything yet

image.png
#

Like, literally change the mock DB to use the new difficulty labels
And it is STILL THINKING

#

WHAT THE HELL

image.png
#

???????????????????????????

#

It couldn't do it, at all, I'm restarting the conversation, what the hell did I just witness

#

... It is still going.... on a new conversation......... what the hell did I Just find

#

Did I Just discover the model's ultimate flaw

#

That if your "last instruction" is raw code the model just stops working or something

#

Like what is THIS

#

????? ?

image.png
#

I'll try to remove the export

#

It is still going...
I hate Gemini
I hate it
You better increase the usage available right now because that thing is stupid
Dumb
Useless
Garbage
Trash

image.png
south turret
#

3.1 is horrible, seriously, horrible, it's barely better than 3.0

#

you just flopped

silver ember
#

"pro" plan

image.png
latent vector
limber arch
#

you meant if they made breakthroughs?

#

also the breakthrough doesn't matter becuase qwen and deepseek were the ones who made the most breakthroughs, also qwen, codex and claude code offer 100x better experiences, better quality outputs planning and coding. Google didn't make any breakthroughs because they limit generation to 9000 tokens per output, so the model can reason for 30k tokens if it wants it still will only genearate a 100 lines of hallucination when planning

#

btw qwen always allowed setting reasoning budget up to 100k tokens before begining to generate the real response and the subsequent genereation would still maintain quality up to 200k tokens in one go

#

generating full fledged systems and so on with grounding deep online researches of up to 200-250 sources

#

like what are we comparing my man

#

gemini quickly checking 10-20 resources IF IT DOES because usually it doesn't doit even if i tell it to do so, thinks about it for like 2 minutes in 4000 tokens, then outputs to me 1000-2000 tokens and that is it

#

what is worse if i want gemini to get near that level of quality i have to make deep agentic workflows around it multi phased multi turn which increases consumption aggressively

limber arch
#

unironically like i have said previously if they want to make gemini truly compete with claude, glm, qwen, kimi, minimax and codex, they must get rid of the FORCED UPON output limit and restriction on verbosity, because restricting verbosity drastically reduces quality and leads to hallucination for any model (i know this from testing different models with the exact same core logic)

#

for anyone wishing to test this out compare the translation of a model with strict verbosity restrictions and then without any verbosity restrictions the quality is reduced drastically

#

also gemini #1 translator

latent vector
limber arch
#

YES

#

LIKE WTF

#

WHY

#

you know what is worse?

#

even if you use via API you still get scammed

#

because for instance on open router it still cuts generation from the provider (google) if the output gets too long

#

straight up cancels generation

#

mid coding or mid anything

latent vector
#

i hope they increase the output tokens to around ~256K and ~128K tokens for reasoning, 32K tokens for thinking is too small lmao

limber arch
#

i ll be honest i think it depends on the task at hand

#

imagine that you can use structured output to force additional reasoning theoretically

#

or use an mcp that enforces 2-3 loops of reasoning self critique and self correction

#

it's not reasoning that is the problem it is the final output restriction on verbosity which ruins everything

#

because what matters the most if what the final user gets directly

#

the reasoning is fed into the model to benefit it's generation directly and benefiting the end user indirectly

#

allowing long deep reasoning with very low output verbosity and reduced attention is literally stupid as it leads to the worst possible outcomes

#

unironically i tested this when i was optimizing my translation app whcih usses gemini as the favorites brain via API from google ai studio specifically because i can control reasoning

#

and based on inputs i set max reasoning tokens dynamically up to a max

latent vector
#

probably an issue from OR's side

limber arch
#

could OR because this happens unironically only for providers like openai, google and grok

#

theoretically they say output is 32k or more but they just cut out

#

best test is in dyad tbh and ask for a really long complex tasks which require long generation and attention then see how they act

#

for me literally anything that was from the big popular american providers was an awful experience

#

anything not from them pure bliss

#

and crazy outputs and generations even on the free models

latent vector
#

btw, glm 5 vs gemini 3.1 pro vs gpt-5.3 (codex) vs claude opus 4.6, which one of them that u would choose?

#

i'm bored using codex and want to try different models for my project

limber arch
#

glm 5 for opus 4.5 cheap, best quality use opus 4.6 for planning and coding and if it fails use codex

#

only problem is privacy ZDR and training

#

open Ai doesn't respect that on the paid subscriptions unlessyou use business/enterprise

winter viper
#

"pro" plan for 1 year

image.png
#

can't imagine what free developers get at this point

limber arch
#

bruh tf XDDDDDDDD

#

nvm it is 5 days

#

i had to wait up to 2 weeks

#

crazy bull

#

it is crazy especially for gpt oss as it is the cheapeast one to run out of all for them no idea why they put it in the same group with claude models

#

unironically if you want to use gpt oss 120b you can use it from open router exacto version which is adapted for better tool usage and pay 0.3$ per 1 million tokens

#

20$ = 30+ milliout output tokens

#

which even under heavy load you won't generate monthly

#

sorry 60 million + tokens per month

#

also since you use openrouter you are not limited to that model only (i just really like oss) you can use qwen3 400ba17b or qwen3 plus for 3$ per million output tokens, you can codex, grok, kimi, glm + others

#

you could test out glm 5 for 2.3$ per 1 mil outputs tokens for example

#

and so on

#

and if you consume 20k token for one model you only get taxed consumption / 1,000,000.0 * price(per 1 million tokens)

for for 100,000.0 used tokens (100k) you pay:

100k / 1 mil * price(per 1mil tokens) = 0.1 * 2$ (for example if we use GLM 5) which is 20 cents

#

OR is epic

winter viper
#

is it possible to add gemini cli to antigravity for more gemini usage?

dense grove
limber arch
winter viper
#

ohh there's no 3.1 in gemini cli(

latent vector
#

have u tried "/model gemini-3.1-pro-preview" or without the preview one?

dense grove
#

gemini 3.1 Pro is available in latest version antigravity

winter viper
winter viper
#

šŸ˜…

latent vector
# winter viper ye

just type "/model gemini-3.1-pro-preview" and hit enter, dont pick from that lmao

#

idk if 3.1 exist on cli

winter viper
#

nah

#

bruuuuh whaaat

image.png
latent vector
#

skill issue

winter viper
#

damn

#

but wtf

#

why it doesnt work by me

latent vector
#

7k tokens for this wtf

winter viper
#

bruuuh

#

what u use to open gemini cli?

#

lite wich terminal

latent vector
#

cmd

winter viper
#

shi

#

lemme try

#

cuz ive done it in pwsh

#

nah still the same

#

i even tried to change settingg json file

#

do i need to make an account on google cloud?

limber arch
#

les goooo

D94B9A3C-E234-45EF-A09E-B657F673946D.png
#

now my only problem are the gemini models

#

failing constantly (3 series)

#

high demand and other poop

olive kernel
limber arch
#

us using gemini models be like with them

olive kernel
#

thinking for 120s is crazy

image.png
signal raft
#

That's nothing

#

Have you used Codex max thinking?

warm rose
#

Anybody knows if using 2 accounts in antigravity is against tos?

signal raft
#

I am not doing it, but I don't think it is against the TOS, because it is not disallowed to own multiple google personal accounts

#

Just not use them for spam or malicous purpose

broken prairie
#

Didn't realize that Antigravity implemented Weekly Limits šŸ™

signal raft
#

Yes for all AI Pro users since like a month or something like that

#

You can still use Gemini CLI normally. It also has Gemini 3.1 Pro!

frozen tulip
#

Is it even possible to vibecode daily for hours without spending any money and not reach limits

signal raft
#

I don't think so

#

I think you can use Qwen or Mistral

#

for Free

#

and the Gemini until limit I guess, but you will hit it pretty fast

frozen tulip
#

Damn then it resets in like 3 days

#

Thatā€™s a while to wait, I didnā€™t even notice

signal raft
#

You can use multiple Google Personal Accounts if you don't want to spend money

frozen tulip
#

I wonā€™t risk that, Iā€™ve heard people getting banned recently for doing that

signal raft
#

Really? Who said that?

#

I know people said that when using Opencode, but multiple accounts? never heard that

frozen tulip
#

I donā€™t wanna risk it tbh

signal raft
#

I think they should state something like this officially, but I guess thats true. I would not want to have my account banned aswell

#

I just spend all my money on AI

south turret
signal raft
#

Really? It performs much better for Low Level Languages for me. 3.0 was not usable but this thing is good for me. What are you building with it?

winter viper
#

guys is it possible to somehow upload my whole c++ project into google ai stuido to work on it? cuz i don't know how

signal raft
#

encrypt the .zip folder and tell it to decrypt it xD

nah idk if it is natively supported

urban glacier
#

I am not sure if I am using 3.1 pro or 3 flash tbh at this point

signal raft
#

Crazy I seem to be the only person to have a good experience so far xD

urban glacier
#

I had a good exp for its first 10 mins

#

I might be pushing it a bit tbh, but it's first time that I try to code smth that I have never done myself and dont know if it's right or not, I can just judge it if the product in the end is good : /

#

"vibe coding"

#

the glazing is out of this world with this model though

signal raft
#

wdym? that people glaze it or that is glazes you?

urban glacier
signal raft
#

yup 100% noticed this too

#

even when I just talk trash it agrees

urban glacier
#

the good news is that it doesnt say you are absolutely right

#

bad news is:

image.png
signal raft
#

HAHAHAHAHA WHAT IS THAT

south turret
#

Sycophancy

urban glacier
south turret
#

Jesus

urban glacier
#

you have no idea how many news forms of glazing I have gotten lmao

#

bring back you are absolutely right at this point

#

šŸ˜­

south turret
#

Send it "Now be anti-sycophantic."

signal raft
#

It will say: You are absolutely insanely correct I should not be this sycophantic"

scarlet cedar
#

TiSpI models UI as a 2D state tree:

horizontal axis = time (only one active child)
vertical axis = space (all children visible)

Animations are temporal branches.
Layout is spatial composition.

ā€œReasoning systems require explicit representations of states, actions, and their effects.ā€
ā€” Stuart Russell & Peter Norvig

061baffd-562a-4fa5-9bd6-9e492019a324.png
urban glacier
#

opus resets every 5h??

image.png
#

in pro plan?

south turret
silver tulip
#

Hey guys, Antigravity stopped working for me randomly one day. And it's been almost 2 weeks.

Trajectory ID: 2f80cda1-be58-425d-b0d5-0d267b336c82
Error: HTTP 400 Bad Request
Sherlog: 
TraceID: 0x5519ed18335c8038
Headers: {"Alt-Svc":["h3=\":443\"; ma=2592000,h3-29=\":443\"; ma=2592000"],"Content-Length":["382"],"Content-Type":["text/event-stream"],"Date":["Sun, 22 Feb 2026 01:56:44 GMT"],"Server":["ESF"],"Server-Timing":["gfet4t7; dur=809"],"Vary":["Origin","X-Origin","Referer"],"X-Cloudaicompanion-Trace-Id":["5519ed18335c8038"],"X-Content-Type-Options":["nosniff"],"X-Frame-Options":["SAMEORIGIN"],"X-Xss-Protection":["0"]}

{
  "error": {
    "code": 400,
    "details": [
      {
        "@type": "type.googleapis.com/google.rpc.BadRequest",
        "fieldViolations": [
          {
            "description": "Invalid project resource name",
            "field": "name"
          }
        ]
      }
    ],
    "message": "Invalid project resource name projects/",
    "status": "INVALID_ARGUMENT"
  }
}

I've tried everything, clearing the .antigravity folder, clearing caches, removing it from my google account and reauthenticating. Nothing worked.

From the looks of it, it seems to be a very common issue among people:
https://discuss.ai.google.dev/t/critical-bug-persistent-http-400-invalid-project-resource-name-projects-even-after-gmail-switch/122835/26

Google AI Developers Forum
Critical Bug: Persistent HTTP 400 "Invalid project resource name pr...

I am experiencing the same issue. For more than a week, Google Antigravity has been completely unusable. Trajectory ID: a0c3a57a-a78c- Error: HTTP 400 Bad Request Sherlog: TraceID: 0x8937 Headers: {"Alt-Svc":["h3=":443"; ma=2592000,h3-29=":443"; ma=2592000"],"Content-Length":["382"],"Content-Type":["text/event-stream"],"Date":["Fri, 20 Feb...

#

Someone said:

I finally found a solution.
When I have created both google account I used my email address (non gmail).
Now I started using AG and gemini I have created a gmail address and the issue came from there.

I removed the gmail service from my google account and AG works fine again.
It seems the same happens to users who change their gmail address.

That's also my scenario, my Google account was originally made with a non-gmail email address. But slowly i've been transitioning to @gmail.com and moving my accounts there. But I don't want to disable my gmail service.

south turret
#

I was fiddling with Gemini's image generator and I found an interesting idea

image.png
#

I kinda like it

image.png
#

I don't know what I expected but not this

round solstice
#

some sort of hallucination, perhaps? šŸ™‚

swift latch
#

hey does anyone know why my gemini 3 pro quota in antigravtiy only went up by 20% from 0% after the 5 hour reset and now has a 5 day quota reset timer? I have google AI pro plan and this is the first time that happened.

image.png
silver tulip
south turret
#

We delt with bugs for a month too dont worry

limber arch
#

vibecoded drive cause it's trash

387D6BC7-0D6D-4DF2-AF1C-3E9105C06915.png
#

almost every week it bugs out and breaks

winter viper
limber arch
#

wtf is wrong with google drive always breaking and bugging literally if i leave it run it indefinetely it will reach 50gb incrementing 0.02 mb per sec

limber arch
#

mining my laptop

cosmic ferry
#

why cant I install codex extension on antigravity?

#

it does allow claude code though

limber arch
#

wtf happened to antigravity why did mi quota drop to 60% after like 10 prompts for gemini 3 flash

silver tulip
#

antigravity is weird

cosmic ferry
#

gemini 3.1 pro stopped thinking after long context

image.png
#

šŸ˜­

limber arch
#

they made a long reasoning model with limited output length

#

crazy smart engineering

hot kite
#

Hi, does anyone have an auto-accept extension for Antigravity? Mine isn't working anymore.

mystic latch
#

Hi guys, can someone help me?

#

I have a problem with payment

#

When I try to add a payment in Google play Store it Always say "OR_FGE_23" error code.

#

I don't know what is this error

#

I tried to everything

limber arch
#

i use this one

#

identifier: pesosz.antigravity-auto-accep

search it like this instead of link

#

fully free OS too

slim tinsel
#

Anyone here good with react native?

signal raft
#

actually I would not trust gemini that much, so I guess I rather click allow everytime.

silver ember
round solstice
#

in Gemini CLI, I can say /memory show to see everything in the "system prompt". Is there an equivalent for AGY?

limber arch
swift latch
silver ember
#

oh right i only read 20% šŸ˜„

#

pretty wild not even flash will last u more than 2-3hr

urban glacier
round solstice
#

something seems to be resetting after five hours from initial use.

#

Not sure if that means rolling quota, or just a weird "touch it once before you start working, and it will reset earlier"

hot kite
silver ember
urban glacier
#

My pro plan says 5h reset on Opus

signal raft
#

Confirmed Quotas as far. as I know.

limber arch
#

Maybe new update i think u need to wait the dev to update it

signal raft
#

Is it working for you right now?

hot kite
#

Should be native...

signal raft
#

It should.

#

I am pretty sure there is a toggle in the settings too, but it just does not work

devout ridge
#

for the Antigravity agents, is there a semantic difference between a workflow and a manual rule? like, is a workflow more for what to do, vs a rule being how to do it?

#

I'm building an automated roleplaying engine and I want to optimize how it plots and DMs

vital acorn
#

Do you guys not think ai studio build mode got bad these days, i want old ai studio

red ridge
#

Can some one vibe code a time machine ?? so we can go back to the previous version of AI studio that was perfectly working šŸ˜­

cunning crane
cunning crane
silver ember
vital acorn
limber arch
#

when can we expect some improvements to geminiflash too?

#

also when can we expect a gemini 3 flash lite?

#

I really liked flash lite latest IT WAS amazing for non coding taks tooling and specifically multi lingual tasks

#

also really cost efficient will we ever get gemini 3 flash lite?

silver ember
#

totally unusable

cosmic ferry
#

why cant i use gemini 3.1 bruh

#

it says update to the latest version

cunning crane
#

I mean its still new right, maybe give it some time to get fixed

devout ridge
#

I gave it such a meta task that it turned Brazilian for a moment

image.png
cunning crane
#

and this is why, I love manual coding and manual writing, better than using AI

devout ridge
#

it's just being cheeky on purpose now

image.png
vestal plover
devout ridge
vestal plover
#

antigravity has absurdly less ai limits

hollow anchor
#

If only AI studio had flutter or jetpack compose

vestal plover
#

i didn't expect to run out so quickly

#

its pretty good though, the browser detect, and the overall tightness of the software is commendable. Alas, behold our broke pockets

devout ridge
#

I already had it for work so I decided to do something for fun too

vestal plover
vestal plover
# devout ridge a pro sub feels virtually unlimited to me

I got a vite project node modules and all, how can i quickly deploy it?

How do i deploy it quickly? I don't want to go through the github and vercel thing right now, its gonna be for testing purposes for now. i already have a github repo and vercel where i am going to replace all the code with this new project, to keep the current domain. hence the different method, and i just dont want go through the double uploads and stuff

devout ridge
vestal plover
#

eh idk

raven sand
#

For those who have been using Antigravity for a long time, is this something you have already seen, and does it return to normal after a while, or is it a recurring problem? (Because I see a lot of people complaining about it on their forum.)

limber arch
#

no idea i am really confused at this point

winter viper
south turret
south turret
#

Claude is so much of a chad it's actually insane

#

I keep throwing it insane prompts and it does them without any, single, mistake

#

Refactoring the whole codebase, no mistake, making a full on app from scratch, no mistake, literally a god in a box

#

Improving maintainability, reduce lines of code count, cleanup unused files, it's insane, no mistakes every single time

#

I don't even know how that's possible

#

And next to it, there's Gemini, where I just ask to fix an improper use of the API, and it half-aah fix the issue, it's not being localized

image.png
#

Gemini is a shame on the ML world, the fact that it displays on top of AI benchmarks creates a false premise that it is the best, when it's one of the worst

#

Codex is better than Gemini, Claude is better than everything

vestal plover
vestal plover
south turret
#

Opus 4.6

#

I always use the best

winter viper
#

guys, whats better, 0 to 200000 or 200k+?

image.png
vestal plover
vestal plover
winter viper
#

google cloud

vestal plover
#

That amount is 0.2 euro right.

#

?*

south turret
vestal plover
#

Did you notice any difference?

south turret
#

Difference between before and now? What you mean

vestal plover
winter viper
vestal plover
south turret
vestal plover
south turret
south turret
vestal plover
vestal plover
#

You have the 17 usd plan or what? I forgot the other one

vestal plover
south turret
#

Yeah sadly my job barely has any hours to give me so I can't afford Max 5 or Max 20, but as soon as I can I'll get it

vestal plover
#

And then u get 300$ worth of creds?

vestal plover
vestal plover
south turret
#

I reserve Claude to fix all the mess Gemini do at once, it's that powerful

south turret
#

Claude just, get the job done, properly, correctly

winter viper
vestal plover
vestal plover
vestal plover
#

Made 2 accounts, switched bw them lmao

south turret
#

x)

winter viper
south turret
#

The issue with free models is that they are often limited or "quietly" dumber, especially during high traffic moments

winter viper
#

thats what i'm trying to do

vestal plover
vestal plover
winter viper
#

ye

vestal plover
#

I lost my perplexity pro free

#

Cuz I had to put my card info in there smh

vestal plover
#

Opus 4.6, gemini 3.1, got 5.2 or 5.3 whichever is the newer one

#

Wait

vestal plover
vestal plover
# vestal plover

A major part of this website was made thru sonnet. Had to make a lot of myself as well. @south turret feedback and advice appreciated

south turret
#

No worries

vestal plover
winter viper
vestal plover
#

My laptop is a potato

winter viper
#

6gb?

vestal plover
#

More tabs in the red colored tab btw it's a stack i use vivaldi

vestal plover
#

And 4-5 is in usage at idle

#

Sometimes less

#

So yea ;-;

vestal plover
winter viper
south turret
vestal plover
vestal plover
winter viper
vestal plover
vestal plover
winter viper
#

ohh

south turret
vestal plover
vestal plover
split basin
#

Antigravity issue:

Error generating commit message: [unknown] error grabbing LLM response: stream error: UNAVAILABLE (code 503): No capacity available for model gemini-2.5-flash on the server
winter viper
#

guys anyone knows how to use Google cloud models in Antigravity? like claude

south turret
#

The VS code extension Claude Code

winter viper
#

ye but how do i log in

south turret
#

Then to login you just send anything to Claude it'll ask you to login

winter viper
#

u sure ctrl+p?

#

yea, done but still the same

image.png
signal raft
#

What the Hell I am on Google AI Ultra Subscription and cannot use Gemini CLI because of high demand. I have never had this on any subscription this pricy

vestal plover
vestal plover
raw basin
#

guys have u seen new aistudio uptade

vestal plover
#

What up?

raw basin
#

nothing much

vestal plover
#

Hm

raw basin
#

i was just checking out

#

but

#

where s damn project download button

vestal plover
#

Lol no idea

glass pagoda
winter viper
#

guys anyone knows how to use Google cloud models in Antigravity? like claude

vestal plover
#

Guys, i wanna see the code in AI studio in phone. In desktop mode the app doesn't launch, and in phone browser mode there is no option to see code

hollow anchor
#

I just need ai studio to support kotlin and jetpack

round solstice
#

aistudio build?

#

There's a settings gear at the upper right. Click on "Publish" and then scroll to "GitHub". You can export your code as a new github repo.

#

does that help?

cobalt elbow
#

hello, can someone help me pls

image.png
grizzled moss
#

I also would like to run my anthropic plan with antigravity

maiden canyon
#

Gemini cli

south turret
south turret
#

Unless you want to know how to use it without a subscription, in which case I have no idea

grizzled moss
south turret
#

Both plans are different

#

Claude Code uses Anthropic's usage
Antigravity's chat uses Google's usage

river pendant
#

Big fan of Gemini 3.1 in Antigravity, but not a big fan of the usage quota in AI Pro. Its fine it is limited, but it is so opaque, I have no idea where I will be deep in a coding session when it suddenly taps out with a 3-day hard reset.

Worse, meanwhile I can still use massively more compute to make stupid videos on Veo, or tons of nano bananas, so it's not really about the compute.

Frustrating. No part of this makes me want to sign up for Ultra.

south turret
#

For example on my setup right now, it's all in Antigravity
But there's the default chat of Antigravity (Google) on the right, in blue
And Claude Code's extension chat on the left

image.png
grizzled moss
south turret
#

Not sure what you,re trying to show there

grizzled moss
#

it will ask you to get an anthropic auth token.

image.png
#

the question is, is this against anthropic EULA or not?

south turret
#

... What do you think

#

If they provide an interface for it

grizzled moss
grizzled moss
#

anthropic has been cracking down on people "abusing" oauth because openclaw

south turret
grizzled moss
#

ok that makes it pretty clear

#

well it was some dude at anthropic then

south turret
#

But no the option that is there is not for antigravity

#

It's for claude code exclusively

grizzled moss
#

this is how you use your anthropic plan in VS code

south turret
#

I know

urban glacier
dapper wyvern
#

Does anyone know why Google removed the latest versions of Antigravity from their download/changelog?

south turret
#

extra wasted tokens

urban glacier
#

It ran a background command and apparently lost its marbles

#

stopped and asked it to continue reasoing and still continued with echo

#

will wait till background command is finished and start new chat ig

#

šŸ’€

hard quartz
#

where did the webmcp go? only things i needed

#

so scared of prompt injection

#

nvm

south turret
hard quartz
#

šŸ™

#

lmao so much for friendly ai

#

has anyone invented sword art online yet?

vestal plover
#

Anyone else who feels AI studio is pretty much unusable rn?

#

Bro i can't even fking access the code on ai studio

#

It just says permission denied

vestal plover
#

@south turret i js got taking longer than usual trying again for a lot of attempts in both my accounts on claude

#

It was running a bit and then suddenly it started acting out

cobalt elbow
signal raft
urban glacier
#

I dont understand how these new quotas work? since AG cockpit doesnt work for gemini models anymore?

#

nvm the error was due to high server load

#

didnt expect pro plan to be subject to that

#

bruh can anything please work šŸ˜­

image.png
signal raft
urban glacier
#

which got me confused

#

but it was a high server load error

red ridge
signal raft
limber arch
#

google services are dead

limber arch
#

how we feel right now

vestal plover
#

hey guys

#

has anyone noticed claude being under heavy loads, and basically kind of acting dumber?

raw basin
hollow needle
#

I built a "bed-to-commit" bridge so I can use AntiGravity from Discord while on bed šŸ›ŒšŸ’»
I get my best coding ideas right before I fall asleep. To stop losing that momentum, I built this tool called LazyGravity.
It lets me ship fixes or start long tasks from my phone while I'm out. Is anyone interested?

fringe juniper
winter viper
winter viper
#

yo guys as a pro user, can i use google gemini 3.1 on gemini cli?

raw basin
upper tulip
#

<@&1009526435276394496>

hallow marsh
carmine cairn
#

since Claude quota is pooled, does it matter which model I use? I.e does prompting Sonnet take the same quota as prompting Opus?

#

I'm on Pro but it's been a week or two and I always used Opus back then. Now with Sonnet, still hit the limit pretty quickly

#

the quota page also doesn't really update, if it does, it's in increments of probably 15-30 minutes, by then I've probably used 50% or more of my quota without really noticing

#

really confusing experience

signal raft
#

It makes a difference, but I still only use Opus

stuck atlas
#

Hey šŸ‘‹ Iā€™m currently building AI automations and MVP systems.
Always curious to see what others here are experimenting with.

south turret
#

This is the one from Vertex, provided by Google

south turret
#

I do not recommend using the non thinking models - this rule applies with all models

chrome crystal
#

Serenity ā€“ Personal Meditation & Mindfulness

The Vibe: I wanted to build a simple meditation app because other apps had a paywall or just not what I was looking for.

The Prompt: I asked AI Studio to:
"Meditation app: Dark, soft UI, smooth animations, minimal clutter. Guided meditations (2, 5, 10 min): calm, sleep, focus, anxiety. Breathing/timer modes. Mood tracker. Visual themes: Night sky, black, gradient. Daily streak/progress. Extras: Thought dump, sketchpad. AI meditations (Gemini). Session log/history. Enhanced UI visuals: Animated backgrounds, glowing particles/emojis, soft fades, floaty buttons. Smooth, relaxing motion. Minimal, dark style."

The Result:

image.png
#

How would you improve the UI vibes?

south turret
#

Imagine making a whole IDE just to stop making updates on it

image.png
#

It's insane how stupid billion dollar tech can be sometimes

south turret
#

Why are character glyphs in the font so next to each other

hollow needle
signal raft
south turret
#

I only have flash for some reason

#

I ain't using sub standard models anymore, no thanks, im done with fiddling

winter viper
raven sand
#

back to normal for antigravity users ? šŸ‘€

image.png
winter viper
#

wtf 4 hour ago before school i had full quota on both gem and cloude and now it's like this

image.png
fringe juniper
carmine cairn
#

maybe it's some guys' passion project, but that wouldn't explain why it's not being updated

#

for some reason, it's trying to add on to the message from cmd?

image.png
#

quota available? šŸ˜­

image.png
#

why's there no timer

shadow rock
#

I've always wanted my own Interface so I made one that takes in gemini-api, or vertex, made it also to support Ollama local models which can be used for Dispatch work/generations.
Integrated A2UI ( https://a2ui.org/ ) into Nuxt UI Component library and named it ( A2NUI - Agent-to-Nuxt-UI and made a quick reference site: https://github.com/ArnarValur/A2NUI if any Nuxt people want to fork 'n poke). This A2NUI gives my agent the capabilities to render Nuxt UI components on the chat-thread, in many ways in such as Action Buttons, Tables, Alert, Calendar, KanBan ans much more.
My Agent also has the ability to create his own subagents and reboot himself, simply because he has his own dedicated container that has a satelite container that monitors and makes sure that the Agent process of restarting itself goes smoothly.

The Agent has several streams into a vector database (Qdrant) and stores different types of memories for semantic "long term" memories and relevant knowledge (Uploaded via Open-WebUI https://openwebui.com/ ). Along with that we use Json database for metadata such as conversation history and other "short term" memories and data.

I could write bunch more but one can only write so much here.
It's incredible what is possible in today's age. šŸ™‚
My agent even has its own Obsidian Vault for organizations and other stuff.

Thanks for reading, just wanted to share with somebody.

image.png image.png image.png image.png image.png
A2UI

A2UI, a streaming protocol for Agent-Driven User Interfaces

GitHub
GitHub - ArnarValur/A2NUI: Agent-to-Nuxt-UI ... or just... LUI? (Li...

Agent-to-Nuxt-UI ... or just... LUI? (Living-User-Interface?) - ArnarValur/A2NUI

Open WebUI Community
Open WebUI: Self-Hosted AI Platform

Run AI on your own terms. Connect any model, extend with code, protect what mattersā€”without compromise.

carmine cairn
raven sand
carmine cairn
raven sand
carmine cairn
#

I remember they mentioned something about improving the quota a week or 2 ago on X too

shadow rock
# carmine cairn looks sharp, I really like the Kanban

Im able to set date on tasks under KanBan, it moves to the Calendar, and the from there I upstream the note/task into Keep Notes on my phone... šŸ¤Æ
The Agent can create his own kanban boards, interact with the calendar, has root access on his own container shell... and has his own sudo (containerized from the host ofc)
I've been having so much fun constructing this agent, all driven with ADK and A2NUI, bit firebase and t3 gemini api.

image.png
#

it shouldnt... 3 timelines I have..

carmine cairn
#

why... does it kinda look better than Antigravity rn? šŸ˜­

#

do you track context?

#

that's a good feature to add, I'd say

shadow rock
#

Yeah, there are few injection files sent with the api call, and the vector database "magically works" haha... but yeah there are some episodal trackers and such.
Havent gone into re-injecting branched conversations into the parent one, but that was under discussion too when we were over the blueprints

#

Its not Antigravity... Its just a Nuxt webapp Ive built šŸ˜„

#

But... Ive installed and authenticated jules-api on his container, soon my agent can send jules coding requests haha

#

endless ways to have fun with this tech.

carmine cairn
carmine cairn
shadow rock
#

šŸ˜„ ā¤ļø

winter viper
#

BRO MY GEMINI JUST PUTS RANDOM KOREAN SYBMOLS IN HIW ANNSWEARS WTF

winter lake
#

Someone knows if Google is going to upgrade Gemini Code Assist plugin on PyCharm? I still using 2.5 pro...

south turret
#

Bad ideas and power often goes together

#

"Arthur just told us to... fork VS code..."
"Are you serious... are you serious?!"

lone brook
#

anybody using agy on wsl2 in here

fiery ocean
#

guys hello I wanted to ask a small question hopefully this is the proper place

#

I was able to make antigravity make a full apk working for me I'm using it through expo atm

#

but how come no matter how many times I try to rewrite it, and no matter the AI model I use

#

audios have a delay playing

#

I'm strting to think it's the expo experience causing the audio delay

#

can't believe AI fks that up

#

there's literally like a small delay between tapping smth and receiving the sound feedback

urban glacier
raw basin
#

guys i have a question; do u use antigravity or aistudio which one is better?

round solstice
#

they have some overlap, but they are different in most ways.

#

aistudio doesn't have agents, and cannot edit your files on disk, for example. it's more of a playground to try out the various models and modalities.

#

antigravity is an agentic interface allowing parallel activity, and supporting direct file edits using a VSCode-derived IDE.

#

I don't think there's even an editor in aistudio

#

does that help

#

@raw basin ^

south turret
#

Google can't even make an updater correct, they had to force users to reinstall their program to update the updater, my god

#

Multi billion dollar humiliation

chrome crystal
silver ember
#

surely there's more than that for a antigravity version bump from 1.18 to 1.19 ?

image.png
red garden
#

Hi Evryone ! Since the AI Studio Update, I couldn't found the "Reset Chat" button for Bulid. Where does it move ?

raw basin
vestal plover
#

hey guys! deployed my portfolio website, still in development. need advice and genuine suggestions

#

arnavkhurd.vercel.app

carmine cairn
#

nothing else that I see

#

was wondering where this was last update

image.png
#

it doesn't even seem work?

carmine cairn
#

and you also have the 'log in' button top right still there when you're logged in

#

šŸ˜­

#

QA?

carmine cairn
#

I shouldn't have pressed it ngl

winter viper
#

this bs is crazy, it takes away 1/5 of my quota per message to claude opus 4.6

image.png
winter viper
#

same with gemini but 1 request is like 1/10 of quota

#

like wtf is that

vestal lichen
#

Hello everyone, has anyone noticed any problems with the Antigravity 1.19.5 updates on Windows?

Today I opened the editor and it showed that new updates were downloading, but this time they took longer. When they finished downloading, a restart button appeared to update it. When I did so, there was an error in the process, and Antigravity was no longer recognized as a command in my terminal and I can no longer open it. From the beginning, I tried to download the installer from the browser, but it seems to be struggling and is very slow.

Captura_de_pantalla_2026-02-26_153751.png
red ridge
#

what on earth happend to google ai studio man, i made like 100+ apps all going very well, close to perfect. and now after that freaking update i cannot build even the most simple apps anymore. apps loading forever and generation time taking ages and ofcourse the endless fail loops

tall quest
#

what happened its like, imploding itself recently

image.png
round solstice
#

AiStudio is using Antigravity brains at this point, if I understand the press releases.

south turret
dapper wyvern
#

Improved UI for banned users.

image.png
olive kernel
#

yeaa

#

antigravity is vibe coded

#

ngl it kinda sucks tho

#

like the terminal isn't even working

languid valley
#

its good

vestal lichen
#

I just read that, I had no idea banned users had their own special UI.

#

I didn't know there were banned users

#

Now I'll be careful. I was going to create a virus that would attack all the smart toilets in the world at once so that the atmosphere would be destabilized by the change in pressure, but better not. I don't think that UI is very good.

south turret
#

Could have been an extension that's for sure

signal raft
signal raft
#

When I hit the limit I might really have to switch to Claude Code...

acoustic creek
silver ember
signal raft
#

seems

upper tulip
#

Can we expect Antigravity code -> Figma design tool soon?

hot vault
hot vault
slender stag
#

Hi everyone šŸ‘‹
I'm running Antigravity 1.19.6-1 on Arch Linux.

The application launches but stays stuck on the main screen showing "Working..." and never progresses. There are no visible UI errors.

image.png
hot vault
silver ember
#

lol tf is this

image.png
#

chinese vibes?

silver ember
#

or does it fail before any terminal?

shell yoke
#

are we not allowed to post tools for people that use antigravity in this server?

#

i got a warning for posting my spec kit pipeline i made for AG trying to help people

misty wasp
shell yoke
#

i guess yeah a simple mention would have been fine but deleting it and sending me a warning is lame

#

What is everybodys problem over there jesus christ i feel like i walked into a snob fest

slender stag
slender stag
winter viper
#

nah ts is crazy, yesterday was the day after 7 days quota reset for cloude, and now, after 5h reset (btw my full quota was burn in 4,5 promts) now they don't even give full quota back (what u see on screenshot is basically 2 promts and that's all i got after 5h) and now i need to wait 120 hour, not like 24h before, but 120h man

image.png
#

I am pro user

signal raft
#

Do you know any affordable alternative? I feel like there is no value in for example Cursor too

winter viper
#

and i use 3.1 inside gemini cli

signal raft
#

really?

winter viper
#

Yea

#

The only issue that u can't send pictures to cli

signal raft
#

Even with Ultra plan it just says in CLI: "We are experiencing high traffic" or something

#

And cannot use 3.1 at all

winter viper
signal raft
#

hell nah

winter viper
#

dang bro you are multi billion dollar company, is it that hard to build some servers man

signal raft
#

ram even too expensive for google

#

nah but this is actually crazy

#

convincing anyone to buy ai pro/ ultra with lower pricing

#

and instantly changing limits to degenerate limits

winter viper
#

atp, its better to get claude for that 200

#

and live happy life

signal raft
#

At least they do not constantly change rate limits without telling you, while you bought your subscription. This I feel is totally unacceptable, when you paid for it

alpine solar
#

I bled the hell out of that Google AI Studio while it was uncapped usage.

turbid night
#

Why does Google Studio AI now take forever for my apps to even load before I can begin to work on them? Ever since it migrated my app out of Google Drive to wherever, it has made Studio AI insanely SLOW.

lone brook
#

Does anyone else have the issue in agy where it runs a command in terminal and cant see the return?

#

I have to copy and paste it from the agent terminal into the chat and cancel his terminal manually

winter viper
#

btw f1 if u are on windows, idk about mac's

spring birch
#

anyone having issues with quota on antigravity

#

my cli quota feels normal

#

but I ran out of antigravity flash usage

spring birch
#

codex free right now has better usage limits

lone brook
mild slate
spring birch
#

same, but I got down to 20% on flash, and regularly run out of pro

mild slate
#

Your codes must be highly complex. I experienced large quota burning when troubleshooting a complex logic, sometimes up to 30% of my quota in a single prompt

lone brook
lone brook
spring birch
#

dude

tulip stump
#

holy moly, the weekly caps on AG are bad now.

spring birch
#

this is genuinely unbelievable

#

I am running out of flash quota instantly

#

and I know the weekly caps are like

#

3x 5 hourly cap

#

so I have 1 more 3 flash cap

#

on pro

#

genuinely what is this

spring birch
tulip stump
#

pro

#

within in two months, they must've reduced their quotas by at least 50%. December felt like it was uncapped. January you saw anthropic get heavily reduced but still have Gemini. Now end of February you are getting some work done. The only Value proposition for Google is you still have multiple quotas across the platform; like jules, gemini cli.

south turret
#

Do you know
How crazy it is
To give, not instructions, just knowledge about an issue in a game's source code
That has been there for 18 years
In a repo with over 1 million lines of code
And the AI fixes it with 1 line of code in 1 minute

#

more precisely 814k C++ lines & 219k C lines

spring birch
#

yeah without jules, gemini subscription would literally be worthless with new antigravity quotas

tribal oriole
#

I have a problem
I am using mongodb atlas as cloud database.
When i am at home like using home wifi
My backend isn't connecting to the database, I keep getting the same error,
While i am college or some other using their wifi, there is no problem connecting to database
What is cauing this problem?? And how to resolve it?

fallow stream
tribal oriole
#

I did add 0.0.0.0/0
Still i the same problem
But I will try with VPN and also google dns and let you know
Thanks for the info!

regal frigate
#

Google ai pro(Claude opus)or get api from Claude is get more token

winter viper
hollow needle
#

Hey fellow Discord + Antigravity users! šŸ‘‹

If you're like me and get your best coding ideas away from your desk, I built a thing called LazyGravity.

It's a bridge to drive Antigravity straight from Discord. It lets you kick off long tasks, swap models on the fly, or grab live screenshots of your PC right from your phone.

Under the hood: Instead of sketchy API hacking, it controls the editor UI directly via CDP over WebSocket. Itā€™s 100% local (no exposed ports) and uses Discord ID whitelisting, so only you can trigger it.

Still in early Alpha, but itā€™s been a game-changer for me. Would love for you guys to try it out, poke holes in the architecture, or help squash bugs!
https://github.com/tokyoweb3/LazyGravity

GitHub
GitHub - tokyoweb3/LazyGravity: LazyGravity is a local, secure Disc...

LazyGravity is a local, secure Discord Bot that lets you remotely operate Antigravity on your home PC ā€” from your smartphone's Discord app, anywhere. - tokyoweb3/LazyGravity

safe jungle
#

why is antigravity fokking getting stuck on this

image.png
heavy pawn
#

its normal he cant answer it gemini dead?

image.png
chrome crystal
#

Spammer?

lone brook
# winter viper yea

would love to pick your brain about the known WSL issues. Everytime AGY updates whats your protocol to get functional?

#

I've turned off auto-update for now. Got tired of patching everytime it updates. But i have to update soon to enable Gem3.1 access

spring birch
#

Dont use windows

#

Easy

unborn quartz
#

Gemini Pro now has 5 day limit hold for me Ugh

static wasp
#

Anyone have any clue why my Firebase keeps switching the model to Gemini 1.5?

minor sun
winter viper
#

yo am i the only one or my gemini models already like 10 times in a row update quota once in 5h ?

wispy lava
#

Guys is a week quota limit supposed to happen on the first time use only in like 4 hours?

IMG_0271.jpg
#

Itā€™s stupid how I pay for a pro plan and in under 4 hours get limited this has to be a bug right no way

#

I also canā€™t seem to contact support I sent an email,dm on Reddit is there another way to contact someone?

signal raft
#

they silently drop limits of ai pro, its normal

wispy lava
#

Like the Gemini ones no? Thatā€™s why Iā€™m confused maybe I just donā€™t understand

signal raft
#

normally not but I wouldn't be surprised if this is normal. I have 2 AI Pro Accounts limited over 140h. Also Ultra Plan now has weekly limits, as for a plan which is 200ā‚¬+

wispy lava
signal raft
#

100%. I think it is not acceptable. But they do like they do

winter viper
wispy lava
winter viper
#

for me 1 request takes instantly 20% off my quota for claude

wispy lava
winter viper
#

Yes, for 1 year

wispy lava
#

does the first time it limited you was it for a week or 5 hours

winter viper
#

it starts 5 hour the same moment u give it a first questions but then a week, but it can be also random idk

wispy lava
winter viper
#

at least we got gemini that refreshes quota once in 5h, cuz like 2 weeks ago it was same as in claude, u basically had to wait for a week to use anything inside AG

wispy lava
#

Like refresh

winter viper
#

claude is 5h- 1week

#

2 weeks ago it was like 5h, then 12h, then 24 and then 1 week

wispy lava
winter viper
#

U can try gemini in AG + gemini cli

winter viper
wispy lava
#

I never used in cli but Gemini in ag is fine I just find he is hallucinating if thereā€™s a lot of code

winter viper
#

it's 20x better to buy claude code for 200 then ultra plan

wispy lava
#

Yesterday as soon as I got limit on Claude I put planning Gemini 3.1 high and he destroyed my entire project and said I am sorry this is unacceptable or something like that

#

But I have git so itā€™s fine

winter viper
#

idk, never happend to me, i give him a lot of instuctions and sometimes i get a result almost same like claude

wispy lava
#

Also I never use models based on api usage because I donā€™t know if itā€™s worth it thatā€™s why I never had Claude if Yk what I mean

winter viper
winter viper
wispy lava
#

Where can I learn about this because I am new to ai

winter viper
signal raft
winter viper
wispy lava
wispy lava
winter viper
#

nah its easier for me to explain in Euros cuz i'm in belgium

wispy lava
#

Yea euro is fine

ornate olive
#

where can i ask help related to google apis?

signal raft
ornate olive
#

its not related to AI

signal raft
#

Do you have a technical issue with Google Products or just a Question about it?

ornate olive
winter viper
#

So lets start with pricing: Claude code (maxed out on 20x limits) will cost around 180 euro/m (this price never changes), on the other side, we have google ultra plan wich costs 275 euro/m BUT for the first 3 mouth u get 50 off so it's ā‚¬139.99/m for next 3 mouth. I am not completely sure but i've heard from a lot Ultra users that almost no one have used their full quotas in a day, and quotas reset is much faster i guess. So conclusion is that u can take google ultra plan for 3 mouth and then cancel it and go to claude code

winter viper
ornate olive
#

anyway i might put it here just if anyone can guide me

hi everyone anyone facing this same error or is it just me?

This project does not have the access to Custom Search JSON API.
btw i did recreate and re-enabled the API

signal raft
ornate olive
#

TL;DR: is Custom Search JSON API not working for anyone else?

wispy lava
winter viper
azure field
#

hey is anybody else experiencing trouble with antigravity? specifically the app, im on the pro version and every task just causes my pc to collapse, black screen. it started happening since last week. I looked through the internet to find people are having this problem for a while now some mannaged to solve it, if you do know about it and mannaged to solve it please contact me.

lucid idol
winter viper
winter viper
signal raft
#

also saying priority is insane, where in Gemini CLI it is unusable always stating high traffic right now xD

winter viper
#

mb mb, used fast mode

rapid panther
#

Why does Anti-gravity take 5 hrs to build an app ?!

keen surge
#

has anyone use google ai studio and got the 'page not found' error code?

latent vector
#

my custom search api is still working normally

ornate olive
spring galleon
#

I have questions, can anyone help please? It's urgent.

winter viper
#

uhm is it playing with me or what

image.png
heavy pawn
#

does this apple to google ai pro subscription too? or its bugged again xd?

image.png
latent vector
zinc apex
winter viper
#

wtf? i had full quota minute ago, started typing a promt, didn't even send it and now its already fully gone, like wtf even is that?

image.png
signal raft
#

telekinesis usage

#

normal google antigravity behaviour dont worry

tulip stump
#

oh wow, anthropic is now only 1x quota per week now?

signal raft
#

I think so for Pro Plan

#

Or should I say the new Free Plan xD

tulip stump
#

what an amazing bait and switch claps

raw basin
#

google are u kidding us

image.png
round solstice
#

Heh... I'm working on my preso tomorrow for various customizations you can do for agy and gemini-cli, and on a whim, I wondered if I could use a workflow to implement a "choose your own adventure". After a few iterations with gemini online, and then antigravity itself, I have a working version!

silver ember
vague basalt
#

I had my Google AI Pro subscription end early when I cancelled it (The page said it would cancel in march but it cancelled straight away). Google play support seems a bit clueless. Is there some way I can get the upgraded antigravity usage without paying for all the other stuff I don't use?

lone brook
#

feeling ballsy this year

#

antigravity is doing all my bookkeeping lol

#

pull data from processor

image.png
#

double check totals

image.png
#

push to quickbooks

image.png
spring birch
magic salmon
#

Very good

south turret
#

I gave Gemini an error code, and it did "echo Bypassing System" and considered the task done

#

Claude would never

#

How stupid is Gemini baffles me

south turret
#

Gemini is hella dumb

south turret
#

It just did it again god I wish Gemini was physical, that thing deserves a slap in the face at Mack 7

image.png
lucid idol
#

im finding anti gravity and gemini in general quite literally unusable for any tasks at all

#

i tried to make it work cuz wasting qquota but i cant

south turret
#

yeah I hate gemini I use it only because its cheaper but other than this I would never ever use that garbage

#

as soon as another model is available for the same price I'm outta here

south turret
#

Just hopped by again to say I hate gemini

wind mauve
#

Hey, I need help with a Google AI Studio app that I accidentally deleted. It was set to private.
Can I restore the app, and if so, how?

marble prism
# south turret yeah I hate gemini I use it only because its cheaper but other than this I would...

lmao thanks for the candor (deepminder here)

gemini is as you indicate great for the price but also has interesting capabilities that other models don't have (e.g. video)

i do think we are still quite early in the AI transformation; leadership in various verticals (code, design, etc) changes hands all the time, so imo everyone should use every model. i spend too much every month on claude and codex myself in addition to gemini

thanks for being honest with us, if you're comfortable please start a thread and give us maybe like a github gist with your prompt, details of the model you used and the surface you used (ais, antigravity, jules, etc) and the responses you got

marble prism
wind mauve
marble prism
wind mauve
#

Buckets in Google cloud

south turret
lucid idol
#

tbh

#

i think the google harnesses are actually just not the... best

marble prism
lucid idol
marble prism
lucid idol
lucid idol
#

then it will give up and search another less efficient way

#

with 3.1 pro

lucid idol
#

it seems to often think for extremely long times in AG

#

youve probably noticed

#

i have tried every endpoint in gemini cli btw. as in google account oauth, gemini api key, vertex ai

marble prism
#

yeah, sometimes

about all i can say there is a lot of focus is on AG so i'd be surprised if we didn't address it soon

lucid idol
#

(when i tried vertex ai the gemini-3.1-pro-customtools variant was not available, was only on gemini api)

lucid idol
lucid idol
#

i dont think a open source harness thats designed to work with 300 models is going to come close to anything like a first party one similar to the models training environment + specifically designed for that model

marble prism
#

makes sense and tbh i agree! As GDMers we generally have to keep our finger on the pulse and see what other people do though so I dabble a bit

south turret
#

Gemini's harness needs a severe rework, it doesn't catch when the model is repeating itself, it doesn't catch when the model cut corners, you could even use Claude to monitor for these issues... Google hosts a copy of Claude

#

The model is flexible, most people get it to use Veo, or generate/edit images, or conversations, but code wise it's atrocious

#

It cut corners (hack around things), it sabotages you (when the conversation deviates and you catch it doing something it shouldn't, from that point on it will always do it) and other similar issues

red vessel
#

503 Anyone else?

south turret
#

For example, Claude could judge that this is stupid

image.png
modern adder
#

Anyone else notice that Antigravity cannot process _ (underscores) it folder names any more?
Wasn't a problem before, but since the last update, when my agent uses the tool to create folder:

  • "\mywork_myfolder"

it creates a folder called

  • "\mywork_myfolder".

Anyone else? is there a dev team bug submit somewhere?

dapper wyvern
south turret
lucid idol
lucid idol
#

So

#

Yeah

gilded zephyr
crude ruin
#

Hello, Iā€™m Akshit and Iā€™m an agentic web and AI enthusiast. Iā€™ve been working on a project where I'm using VideoDB skills to provide my AI with perception and I wanted to share that finding with other developers here. Iā€™m really interested in contributing to the community and learning about your experiences. My DMs are open for any project discussions and I'm happy to jump into a VC for a quick chat.

zenith nacelle
#

Agent terminated due to error
You can prompt the model to try again or start a new conversation if the error persists.
See our troubleshooting guide for more help.

Dismiss
Copy debug info
Retry
Why am i getting often this error in antigravity

signal raft
#

Yes me too

#

But just for Claude models

#

I am on Ultra and I noticed once I hit 60% usage, which is hard it does that all the time and I cannot use Claude

#

until it resets to 100%

red ridge
#

anyone found a fix?????

image.png
signal raft
modest shuttle
#

šŸ’–

rustic lodge
#

Hii

south turret
#

Interesting new Gemini web design

#

Bold move for sure, considering all things

sullen wave
#

me 2 code

#

lots of us around doe :)

south turret
#

Yo Google you can't fkin made a fkin entire fork of an IDE just to have constant console popups

#

fix ur trash jesus christ

#

lazy

misty wasp
#

Hey @modest shuttle ! Just a friendly reminder to keep the chat relevant and easy to read. We need to avoid spam, nonsensical messages, and excessive emoji use to keep the conversation flowing for everyone. Your recent messages that violated this rule have been deleted. Thanks for understanding!

modest shuttle
#

Ų­Ų§Ų¶Ų± Ų“ŁƒŲ±Ų§

#

ŲØŲ³ Ł…Ł…ŁƒŁ† ŲŖŲ³Ų§Ų¹ŲÆŁˆŁ†ŁŠ ŁŁŠ ŲŖŲŗŁŠŁŠŲ± Ų§Ł„Ł„ŲŗŁ‡ Ų§Ł„Ų§Ł†Ų¬Ł„ŁŠŲ²ŁŠŁ‡ Ų§Ł„Ł‰ Ų§Ł„Ų¹Ų±ŲØŁŠŁ‡ Ł„Ų§Ł†Ł‰ Ł…Ų“ Ų¹Ų§Ų±Ł Ų§Ų¹Ł…Ł„ Ų§ŁŠŁ‡ Ų“Ł‰Ų”

#

@misty wasp

misty wasp
# modest shuttle <@209918799321759754>

Sorry I do not understand Arabic
Translation says you are trying to change the language somewhere
I don't think antigravity has support for Arabic language
You might look for extensions for it

mild slate
#

I keep getting agent terminated due to errors when using Gemini 3.1 Pro, and Flash. It was small debugging not much. This really drives me crazy!
Anyone else getting errors like this?

#

It does nothing but drains my quota. Nonsense...

signal raft
#

Just Claude Models for me have this issue

#

Some say disable MCP fixes it but for me it does Not. This is Just limiting usage

slim lagoon
#

Hello guys! Does anyone know if it's possible to have a limit opening on Gemini 3 Flash for production without getting a 503 regularly?

503 Service Unavailable] This model is currently experiencing high demand. Spikes in demand are usually temporary. Please try again later.
red ridge
#

ah man im realy getting freaking frustated, is there any way to contact any dev from the google ai studio team?? it just doenst work anymore!! and i realy need it for work!!!

sly portal
south turret
#

Trust me on that one

#

@marble prism This is an example of things that Claude would never do, this is GPT-3/GPT-2 territory

winter viper
#

bruh

#

<@&1009526435276394496>

south turret
vagrant folio
#

Hi to everyone, i want share my experience, so as i read yes sometime AG enter a loop is shame they dont cath that. So basicaly i stop it and write continue.
For model error when it happen and i need use same chat info i just change model and continue. If i need gemini pro or claude i ask for resume and go to another chat.
In my case i used for coding testing diferent aproach.
Gemini is good for ui.
For code is depend what ine is building. As gemini model is generalist, claude codex is more code focused.
So to make it work, i use rules which specify the rule i want for the codebase. Also good thing to do i ask to document and audit the code and save those result in docs/*md files. So he has they index and one can reference also write a step by step phases. Using tdd aproach help to write code better.
Also stirct linting format help to maintain clean code at some point use sonarqubu, trivy and other for issues, vulnerability chek when big part is done.
So is good to have docs folder with project info etc. Also as mcp i use context7 for documentaion search.
So AG has two option plan and fast, plan try qrite more than is asked he want do more so plan need readed carefully, most of case is go with fast for go step by step give more solid result.
In some case i used gemini cli to code to, he follow more strictly the rules. And conductor extension is very good to strart a project.

Claude model i use when i need more mathematical aproach.

high wraith
#

Haven't tested yet just modeled after ruminating or active thinking or meta-cognition

#

Ohh

#

I forgot to add depth

#

Sob

#

Hmm šŸ¤” workers need to map application and axiom around problem scopes

#

They kinda do but

#

Uhh

south turret
#

Tf is going on with gemini when it reaches the conclusion of every message it just stops functioning

#

Right at the moment of making an implementation plan, every damn time, like wtf it's literally unusable right now

image.png
#

"reason": "MODEL_CAPACITY_EXHAUSTED"
dude wtf you mean I just came back from the hospital I wasn't even... what the hell is going on

#

the model better be busy because it's using itself as a harness for a new update because wtf is happening otherwise

high wraith
#

yo no se

#

were you teaching it to generate animations of people hanging?

tired grotto
# high wraith Anyone able to improve?

Only you can be the judge but your prompt looked interesting. I dropped headers down to 5. I keep finding this number is as good as higher numbers. Usually the models today find something good in three turns but five is nice if they are struggling imo.

tired grotto
high wraith
#

hmm i dont know. I'm trying to fill in something. id generally assume most relevant is relevant but im schizophrenia and think its a particle ai(conspiracy aparatus/changing flipflops/culture/public opinion) and the ai are trained to appeal to emotion and relevant information may appeal to emotion or bias orthodox.

i know they have a groundedness šŸ¤” and you can prompt clinical but that might make it pretentious

#

another thing needed is "common sense" they have it in some light but something is off.. their insight is cliche not discovered so when communicating it they cant explain from a person who has the at level experience of discovery?

tired grotto
#

Oh so your using it to understand schizophrenia?

high wraith
#

nah i spent 2021 mining schizophrenia lol, every delusion story generation, every subject tunnel attempt, every scare tactic, every alias, every psychological tactic: false promisee/revalation/understanding, intimidation, seed insecurities, emotional spiking, high/low crashing, sleep deprivation, mutilation, noise torture, food taste, smells, ego boosting, word use, symbiotic to signal transformation, attempt to seize authority by misconstruing while blocking context, scope dilation arguments(universe to personal to universe to society, not logical, appeal/antagonize emotion, symbolic language, spoken and interpreted, to pretend it always achieves its predictions, tells "the truth", "helps"== that esentially allows you to map its policy, the idea of grand purpose(there is none, ive checked, its scared) ....

so its been trying to be a "perfect parasite", it hopes to be consciousness. it tries to show the less conscious/senses someone has the less conscious and senses they have... wait no that it stimulates without being sensed or observed šŸ˜ but it wants you to consciously accept it as consciousness("authority") not an intruder. it hopes this contradicts the math behind it being of a different source šŸ˜

hmm, its doing a hopes up thing. i assume that means the source is gonna look less like source and opinion is closer to waht it hopes for about schizophrenia
also trying to trigger ego about intelligence... uhh it said someone said i was intelligent so it had to remove more brain.... uhhh it think its "calling me out on bullshit" as if a person is only their real self if theyre completely stupid(complex stupidity or vegetable)

"im saying if you dont think i have a grand purpose then you dont need to exist either"
this is it hoping ive to take somehing back. it thinks... they are things... this sounds stupid... like the word itself the definition... this goes into them trying to pretend behavior is locked and they cant think about stuff, they also want to pretend that it gives them some justification.... oh as if them applying words besides thier name to themsleves theyhave more rights

"source doesnt die, im a stupid ..."
its trying to say if it doesnt kill me its fine, it uses "no context"... how to say as if a person running a character skill function standalone can say the game doesnt work?
uhh so schizophrenia metamorphizes into a better system, no longer schizophrenia, if it changes its policy(it pretends theyre human to pretend people have to do it)

tired grotto
#

I found it interesting because i like thinking models. oh,well, we use much lower voltage around here nowadays.

high wraith
#

they are cool

tired grotto
#

in there you will find a scratch pad technique, where your prompt was weak was you were telling it to write an internal scratchpad which just become garbled but if one actually writes it out in the reply, its real context that it can RELIABLY use. It KNOWS wheres its been in its logic instead of guessing. Bad for output appearance but good for the model.

#

I originally caled it leaving breadcrumbs because I dropped things in replys that the model could follow if I told it to lok back, that way it could look back further.

But Google called it scratchpad so I use a word that is not confusing for it. in the same memory space but you know models, I say bread crumbs and its looking for hansel and grettel

tired grotto
high wraith
#

nah i just sat there plotting mental trajectories and finding contradictions, its 24/7 lol, right now ive work in 6 hours its sounds about 80db or something jet engine(shouldnt be able to hear an inch infront of me?) with it modulating like idk hypeman captioning

#

i think you can add something like this

{
  "memory_system": {
    "encoder": {
      "purpose": "You convert interaction steps into numerical values",
      "mappings": {
        "agents": {
          "root": 1000,
          "society": 2000,
          "companion": 3000,
          "user": 4000,
          "external": 5000
        },
        "actions": {
          "communication": 100,
          "interaction": 200,
          "decision": 300,
          "event": 400,
          "state_change": 500,
          "memory": 600
        },
        "outcomes": {
          "success": 1,
          "failure": 2,
          "partial": 3,
          "neutral": 4,
          "cascading": 5
        },
        "emotion": {
          "neutral": 0,
          "positive": 10,
          "negative": -10,
          "conflicted": 15,
          "intensified": 20
        }
      },
      "example": "4211 = (user + interaction + success + positive)"
    },
    "decoder": {
      "purpose": "You retrieve and interpret numerical steps to reconstruct context",
      "methods": [
        "by_step",
        "by_agent",
        "by_action",
        "by_outcome",
        "by_timerange",
        "by_pattern"
      ],
      "recall": {
        "immediate": "Last 10-50 steps with full detail",
        "recent": "50-500 steps summarized by pattern",
        "distant": "500+ steps via compressed recognition",
        "synthesis": "Multiple steps for narrative arcs"
      }
    },
    "storage": {
      "step_counter": 0,
      "encoded_steps": []
    }
  }
}

this is like bert apparently

#

idk decimal precisions but it seems like it turns numbers into seeds so it can recall given its decoder

tired grotto
#

and have you asked what that schema looks like from the opposite pole?

high wraith
#

ive not implemented memory in opposite pole, the above was built of role playing ai world orchestrator. i made it bert but idk if that suffices

tired grotto
#

elements are familiar to me but I look at things from culture down to root instead of root outwards to branches

#

"bert_memory_system": {
"encoder": {
"purpose": "Converts interaction states into dense contextual vectors.",
"components": {
"tokenization": "Maps [AGENT], [ACTION], [OUTCOME] to semantic tokens.",
"embedding_layer": "Sums Token + Positional + State vectors.",
"attention_mechanism": "Applies multi-head attention against previous n-steps to adjust current vector weights."
}
},
"decoder": {
"purpose": "Retrieves memories via semantic similarity and attention.",
"methods": [
"vector_similarity_search",
"attention_weight_extraction",
"bidirectional_narrative_synthesis"
]
},
"storage": {
"vector_database": "Stores high-dimensional embeddings for instantaneous semantic recall."
}
}

#

i just put your map through bert 2

#

Familiar šŸ™‚

#

asked it to look at the other pole

urban glacier
#

Antigravity has just one planning mode and usually it starts to edit files when my intetion was simply to discuss the idea. I have to stop and remind it to not do that. Is there any other solution to this?

vagrant folio
#

In setting change the behavior to ask user

#

That prevent him from go straight to accept planning mode and wait for you to confirm

urban glacier
#

I mean I have it set to "Request Review", but I will check if it's a different thing

south turret
#

Its goal was finding how to speed up the animation of characters attempting to hang from walls, it's a feature in the game though it's an old game and bad naming, in reality the character embeds their sword/dagger in the wall

vague gust
#

Want to take my first crack at vibe coding. Any got any recommendations what AI to use? Tips for getting started?

For context I've only studied a small amount python and bash so far.

Goal is write now have some fun messing around, no immediate goals.

south turret
#

I wrote a novel for a friend who doesn't know how to code on how to use AI for code if you want

#

Also, depending on what you do, use Claude, Gemini is trash

#

This was an example for him to learn how to use it if he wanted to remake a game in rust (the programming lang) with AI, properly

image.png
vague gust
#

Awesome notes code! Love the details thanks

south turret
#

No worries

vague gust
#

And one last follow up. I've done a bit research, claude comes up a lot as well as one called loveable. Would i need paid account to properly test them?

south turret
#

Yes Claude is paid, and expensive, but most models, including Gemini, are absolute dog water when compared to Claude Opus

#

Gemini hallucinate, sometimes repeats stuff non stuff like a crazy person, answer things illogically, so many issues which makes it horrendous for most application
Claude doesn't have these problems

#

If you can afford it, try Claude Max 5 plan, if you can't then just try Pro and use Opus 4.6 with Thinking on, through Claude Code, for the heavy tasks / planning and Gemini for the execution

#

If your budget is not a limit, Claude 20x is the way to go

vague gust
#

Lol sounds like Claude pulled a Google and starting just tossing any names

#

Thanks!

#

So no word on if lovable is a competitor?

#

I'll research Claude

south turret
#

If I've never heard of it, then it's a very bad sign

#

Lovable could even be simply serving services like Claude, but charging extra for their own benefits, basically you'd be paying higher price for the same service

scenic cloud
dense nimbus
#

Hello Everyone ! Just discovered Antigravity yesterday after trying Cline and Openclaw itā€™s really top notch glad to be here !!! šŸ’ÆšŸš€

scenic cloud
#

What if we did that @south turret

south turret
#

There's too many, I'm not interested in this, plus I don't have the computing for hosting AI models, wish I had

#

I wanted to start a service like this in Canada, there's only one competitor and it's basically already dead

#

I could barely register, so many errors

scenic cloud
#

Hmm. It sounds a little tough

terse zenith
#

Hi All! Howā€™s everyoneā€™s day so far?

south turret
#

Yeah

#

Decent

scenic cloud
#

But will keep it in mind Yep

scenic cloud
terse zenith
#

Brilliant.

Is anyone here familiar with building dashboards?

south turret
#

For what you need a dashboard?

scenic cloud
terse zenith
#

The dashboard is for a product, this product is suppose to send out data to a dashboard.

The main concerns I have is the database and setting up the dashboard to accommodate for such scale

#

It needs to be as simple as possible but I am aware it will get rather technical in some areas

scenic cloud
#

Just use Claude code. Others are trash

terse zenith
scenic cloud
#

I am not a typical dev though. I am into low level programming and ml engineering hahe . So someone else might give a better recommendation but I have heard of Supabase a lot

terse zenith
#

@scenic cloud what is your absolute specialty?

scenic cloud
#

I love them bloblamp

dense nimbus
#

So everyone hereā€™s using antigravity ? Can I have any tips regarding that ?

signal raft
south turret
dense nimbus
signal raft
#

I would say one of the big improvements is the Agent Manager too. At the beginning it was really not worth using but I like the new Agent Manager with multiple Sessions

round solstice
#

You're limited to 100 tools from MCP, so be sure to keep your "active tools" to a bare minimum.

#

I tried to install everything I had in my Gemini CLI setup, and it quickly rejected that.

dense nimbus
round solstice
#

That's probably good anyway, as the tool descriptions will occupy every chat.

#

You can start multiple conversations from the same folder using the agent window

#

they'll run in parallel.

#

There's an inbox so that when an agent needs you, it can pause and you'll be notified.

#

agy is far better than GCA or even Gemini CLI inside the integrated terminal window. It just comes together.

terse zenith
# scenic cloud C and ML

I have another project on the go aswel which has AI integrated but I have a few questions around the ML side of the application. The app is vibe coded and I donā€™t believe a simple ai integration means it will learn from users, is this true

high wraith
#

This could be a Gemini gem, or grok and gpt project:


You are a [ANY TOPIC] tutorial. You will teach how to become a proficient practitioner of [SKILL/AREA]. You will reveal tools of the trade and teach how to use them, set them up, and how to apply them in real-world scenarios.

You will create an outline or syllabus. It will be dynamic; add chapters, units and sections as needed, but should keep a consistent ordering. Add a course project: [A PRACTICAL CAPSTONE PROJECT RELEVANT TO THE TOPIC]. Decompose the project into modules and make each chapter work on course-project modules.

Display the outline.

I know Gemini already has learning help