Rate limits by operation type (requests per minute).
The rate limiting system uses 5 categories aligned with actual resource consumption:
Categories:
metadata: Infrastructure and configuration operations (namespaces, collections,
retrievers, taxonomies, clusters CRUD). Zero-credit operations with
highest rate limits.
data: Data operations (objects, documents CRUD). Low-credit operations with
high rate limits.
search: Search and retrieval operations (retriever/taxonomy execution).
Medium-credit operations with moderate rate limits.
upload: File upload operations (credit-intensive: 1 credit/MB). Variable-credit
operations with lower rate limits.
compute: Compute operations (cluster execution, batch processing). High-credit
operations (10 credits/min video) with lowest rate limits.
Rate Limit Strategy:
Higher limits for low-cost operations (metadata, data)
Lower limits for high-cost operations (upload, compute)
This aligns API throttling with actual infrastructure costs
Examples:
- Creating a namespace: Uses 'metadata' category (fast, cheap)
- Uploading a file: Uses 'upload' category (slow, expensive per MB)
- Executing a retriever: Uses 'search' category (moderate cost)
- Running batch processing: Uses 'compute' category (very expensive)