Use Case: I’m using WorkBuddy, an AI agent platform, for Agentic Workflow development. A single task typically involves multiple rounds of reasoning plus tool calls (file operations, web search, etc.), which generates dense API requests in short bursts. The current 40 RPM limit frequently triggers 429 errors, disrupting normal workflow.
Current Limit: 40 RPM
Requested Limit: 200 RPM
Expected Usage: Personal development only, not intended for production-level high concurrency.
NVIDIA cannot manually increase rate limits for free, personal developer accounts. The 40 RPM limit is a global hard cap enforced across the evaluation tier to maintain system stability for everyone.
If your workflow is hitting a bottleneck, you have two options depending on your budget:
Optimize your code (Free): Implement request throttling or exponential backoff (e.g., using time.sleep()) to pace your script within the 40 RPM limit.
Scale your infrastructure (Paid): If your project genuinely demands a 200 RPM production workload, you will need to host the NIM container locally on your own hardware or upgrade to an NVIDIA AI Enterprise tier.
The evaluation API is designed strictly for basic prototyping. If your requirements have outgrown the free tier sandbox, it is time to build a robust local pipeline or budget for a commercial license.
Hugs and best of luck!