Gemini-embedding-001 not respecting free quota limits

[VooDisss]

App Version

3.23.10

API Provider

Google Gemini

Model Used

gemini-embedding-001

Roo Code Task Links (Optional)

Deleted all of vector databases from qdrant through "localhost:6333/dashboard"
In Roo pressed "Clear index data" > selected "gemini-embedding-001" > start indexing
3072 sized vectorbase appeared in http://localhost:6333/ but yet Roo still failed finishing beyond 95/815 due to free quota limits!

🔁 Steps to Reproduce

Windows 10
Just go to use the gemini-embedding-001 model with a gemini API key which is limited by free quota limits.

💥 Outcome Summary

Expected embedding model to complete without problems (like it does, when using text-embedding-004 model (with the same key, just after gemini-embedding-001 model failed)). Got a message that it failed:

📄 Relevant Logs or Errors (Optional)

Error - Failed during initial scan: Indexing partially failed: Only 95 of 815 blocks were indexed. Failed to process batch after 3 attempts: Failed to create embeddings after 3 attempts: HTTP 429 - 429 Quota exceeded for quota metric 'Batch Embed Content API requests' and limit 'Batch embed contents request limit per minute for a region' of service 'generativelanguage.googleapis.com' for consumer 'project_number:999999999999'.

[dosubot]

Hi @VooDisss! I'm Dosu and I’m helping the Roo-Code team.

Roo Code has features to help manage API quotas, like per-profile rate limits and a "Max Requests" setting for auto-approved actions, which you can configure in your API profile settings. These let you set a minimum delay between requests or limit the number of consecutive API calls to avoid hitting provider limits. Roo Code also has smarter retry logic for Gemini APIs, using the retry timing from Gemini's error responses and exponential backoff for other errors, which should help with transient 429 errors docs docs docs.

However, there’s no documentation on model-specific quota handling or differences between gemini-embedding-001 and text-embedding-004. The error you’re seeing suggests a per-minute quota is being exceeded for gemini-embedding-001, which might be stricter or handled differently than text-embedding-004.

Try setting a higher delay between requests for the gemini-embedding-001 profile, or reduce the batch size if possible. Also, check your Google Cloud Console to confirm your quota and usage. If the problem persists, more detailed logs or developer input may be needed, since there’s no model-specific guidance in the docs.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}   

[SatoshiReport]

Try setting a higher delay between requests for the gemini-embedding-001 profile

I am using OpenAI for embedding and my only setup in roo for OpenAI is in the Indexing section. How do I add rate limits for embedding models?

Thank you @dosu

[SatoshiReport]

I submitted a PR which adds a short delay between embedding calls.

#5715