requests per minute and day limiting

[joetheshmoe]

since roo-cline works so quickly most API's will quickly limit you. once this happens you have to manually press retry often. this could be paired with rotating keys/providers to make everything faster

[h4k4s4m]

want this for another reason, GitHub is banning people for overconsumption using the GitHub copilot api.

i dont want to get my account suspended 😳

[TonyCollett]

@h4k4s4m Do you have any proof to this claim? I just spent about half an hour trying to find anything on this, but I found nothing.

[h4k4s4m]

@TonyCollett https://www.reddit.com/r/RooCode/comments/1i6wkmo/copilot_account_suspended/

[GontrandL]

Hello,

I have the same limit issued by anthropic at 50.000 words by minute to not get an error. Please implement the settings (or make it automatic from know provider limit ?)

[paravastup]

if you use API keys from openrouter, you dont have such low limits.. or spend more, build more and get to tier 4 on anthropic :)

[mrubens]

Does the per-profile rate limiting help enough with this request?

[ericfitz]

Some API responses actually tell you exactly when to retry.

Here's a "429 Too Many Requests" error body from Gemini (I cut and pasted from the Roo Code pane, and formatted):

{
  "error": {
    "message": {
      "error": {
        "code": 429,
        "message": "You exceeded your current quota, please check your plan and billing details. For more information on this error, head to: https://ai.google.dev/gemini-api/docs/rate-limits.",
        "status": "RESOURCE_EXHAUSTED",
        "details": [
          {
            "@type": "type.googleapis.com/google.rpc.QuotaFailure",
            "violations": [
              {
                "quotaMetric": "generativelanguage.googleapis.com/generate_content_paid_tier_input_token_count",
                "quotaId": "GenerateContentPaidTierInputTokensPerModelPerMinute",
                "quotaDimensions": {
                  "location": "global",
                  "model": "gemini-2.5-flash"
                },
                "quotaValue": "1000000"
              }
            ]
          },
          {
            "@type": "type.googleapis.com/google.rpc.Help",
            "links": [
              {
                "description": "Learn more about Gemini API quotas",
                "url": "https://ai.google.dev/gemini-api/docs/rate-limits"
              }
            ]
          },
          {
            "@type": "type.googleapis.com/google.rpc.RetryInfo",
            "retryDelay": "56s"
          }
        ]
      }
    },
    "code": 429,
    "status": "Too Many Requests"
  }
}

Note that it specifically tells you when to retry:
"retryDelay": "56s"

[LittleVoidGames]

Had a similar issue and found this feature request.

I have an implementation idea for this issue. Made a short write-up. Not sure how feasible it is, but it may help. I hope it addresses the user's issue. I think it may address mine.

Waterfalling Providers Per Mode.md

[ericfitz]

That change looked way more extensive than simply monitoring http 429 errors and honoring the suggested retryDelay.

[ericfitz]

@mrubens Would you take another look at this feature request?

Summary: monitor API responses for http 429 "too many requests" responses from LLM provider. If present, use the numeric value in the retryDelay parameter (in error/message/error/details[]) in the API response as the number of seconds to wait before retrying the API again.