Gemini seems to be hit for no reason, when the agent isn't even active

[bcardarella]

When the agent has completed and I can verify there are no other agents that are runnnig I will see the Gemin auth notification. It seems to be doing something as I will hit my rate limit with Gemin now (I'm an Ultra AI account) pretty quickly.

[junhoyeo]

Thanks for the report. A couple clarifications to help us pin this down.

When you say “Gemini auth notification”, do you mean a Gemini/Google-side usage or rate-limit warning?

If so, one possibility is delayed usage attribution on Google’s side. Requests made earlier can show up later, which can look like “phantom” traffic after the agent has finished. (like in Claude etc)

[bcardarella]

5XT6FQYncP9acEzC.mp4

OMO was doing this despite no agents actually running

[junhoyeo]

From the screenshot, it looks like you’re being rate-limited (at least now). Could you share the exact details of what you were doing when it happened?

It’s also possible a background task is still running even after the main agent run completes.

• The exact steps right before you see the auth prompt (what command/action/tool calling, which agent, which model)
• Whether this happens after every run or only after a specific agent/model
• Any logs or error messages around the time it happens?

[bcardarella]

The problem isn't the rate limiting, the problem is that it is attempting to hit the Gemini API over and over despite no work being doing. I suspect that a rate limit was hit and now it's in a backoff period but the agent stopped. The timeout expires and attempts again, rate limit, backoff, over and over.

[junhoyeo]

Totally hear you. I’ll look into the backoff loop and see how it’s implemented. In the meantime, do you have any additional logs you can share?

Also, in your case did the run actually complete successfully, or was it aborted/canceled? I want to confirm whether the repeated Gemini calls happen after a clean finish, or after an interrupted run (including any background tasks, if any)

[bcardarella]

Sorry I don't have the logs, this was a few hours ago. If it happens again I'll try to capture for you

[381181295]

@junhoyeo
also ran into the rate limiting issue, but i think in my case the agent that got stuck was hitting an anthropic endpoint, because it drained almost 50% of the session quota in like 5 minutes until i quit the opencode session
https://opncd.ai/share/M9TelR0z
session-logs.log

[381181295]

downgraded to v2.13.0 and still happened again,
seems to me like its the same issue as #418

[mrjared16]

@junhoyeo also ran into the rate limiting issue, but i think in my case the agent that got stuck was hitting an anthropic endpoint, because it drained almost 50% of the session quota in like 5 minutes until i quit the opencode session https://opncd.ai/share/M9TelR0z session-logs.log

I've been experiencing the same behavior for the last two days. If I don't watch the session, the quota runs out in just a few minutes.

[mrjared16]

could be related to #472

[i-smile]

I encountered the same problem. After a sub-agent system completed, the main agent also finished and stopped. However, the rate limit prompts kept popping up in the upper right corner of the interface and continued for a long time. I was worried it would keep calling the model, so I exited.

[bcardarella]

I am wondering if this is actually an issue in the antigravity auth plugin?