Today was my first time using Antigravity in about two months. I’m on the free plan and was only working on a small personal website project with some simple HTML, CSS, and PHP edits.
After just 1 - 2 coding tasks and roughly 5 minutes of use, my quota was completely exhausted. What confused me even more is that all available models immediately showed a refresh time of around 6 days instead of a short cooldown.
The requests were very basic website edits, not large codebase analysis, long conversations, or anything particularly demanding. Just a few folder with files in. A couple of months ago I could spend much longer working on similar coding tasks before hitting any limits, so this came as a surprise.
Has anyone else experienced this? Were there major changes to the free plan quotas, or does this sound like a bug with quota calculation?
I get the same problem. So I built a proxy for this. You can get it here: GitHub - 12errh/antigravity-proxy: A TypeScript proxy that intercepts Antigravity's Gemini API and routes to NVIDIA or OpenRouter — chat, tools, browser automation, all working. · GitHub . What it does is intercept the Google Antigravity 2.0 request and route it through a custom provider. Currently, I am using it with NVIDIA because it has good models. But you can also use it with OpenRouter or local models.
I’m not sure if I can make that work but I’ll read through it and try for sure, thank you!! 
Try 3.5 Pro Low or 3.5 Flash low.
Does it get your account banned? There are many similar proxy services like 9router, MITM Server, etc., but they all get your Antigravity account banned when used.
Thanks a lot mate, its just what I needed… but… can someone confirm if using this can get you banned or not?
Maybe you where using Opus or Sonnet? they have very short limit, even in AI Pro subscription.
I tested the free plan a couple weeks before, and i could work for hours before hitting the limit, my workflow was: Sonnet for planning > Gemini Flash for implementing.
Having a good understand of how tokens and context works and building a workflow with that in mind, and using “RTK + Caveman” also saves a lot of tokens…
I was acutally using the better models and after 5 minutes of working my quota was fully gone and I have to wait 6 days.. I’m not sure what’s happening.. Months ago I could use that for hours, maybe it’s cause I’m on the free plan now.
Yeah, that makes sense. My quota with Claude models was hitting the limit very fast in the free tier too.
My workflow was basically:
Plan the implementation with Sonnet > create a documentation with structured workflow, splitting the context into small files that will be read only when needed > change the model to Gemini Flash and start actually coding. That way I could work for some hours even when using the free tier.
No, it’s not, because it just intercepts chat-related requests, and all other requests, like auth, image generation, and voice, go to Google servers. It also generates signing keys, so you don’t need to worry. I am using it since I made it. I don’t get any kind of problem.
Look at the screenshot: I burned 17 million tokens in a day, all free, no account ban, no limits. And that 38 dollars is just a testing number. Its dashboard has a feature where you can set up input and output cost per million tokens, so I was just testing.
This is because of the context that every request has. For example, when I was reverse engineering the requests, I saw a single request had thousands of tokens; some are agent guides and some are skills that are getting injected into every request. So, I stripped out unnecessary parts and compressed the agent context. My preference is that if you are using NVIDIA, then use Step 3.7 Flash; it works well, and I am currently using it for my work.
That’s completely normal, considering those are the free tier account limits for regular users. You need to upgrade your Google account to the Gemini AI Pro plan, so that your limit will be higher and reset every 6 hours.
It suffers from dementia. Cost me $63due to a loop. Just can’t trust ir!
Couldn’t I just use a Google cloud api key
