Facet limits
Keyword facets support up to 5 fields per request and 1–100 returned buckets per field (default 10). size: 0 is allowed only when at least one facet is requested.
Aggregation is bounded by 10,000 distinct (field, value) buckets and 262,144 UTF-8 bytes of distinct values across all requested fields. The same value on two fields counts twice toward these limits. Exceeding either limit returns HTTP 400; reduce the matching population or request fewer fields. Lowering the returned bucket count alone does not reduce the aggregation population. Facets remain inline even when documents use docsUrl.
Separate memory admission limits apply to aggregation work and distributed responses. An individual request over its estimated memory budget returns HTTP 400; reduce requested fields or selected partitions. Temporary budget exhaustion from concurrent requests returns HTTP 429; retry with backoff. See Limits and cost for why narrowing a query may not reduce the memory estimate.
Large results
Query results do not have a hard 6 MB total-result limit. When a response reaches the inline payload threshold, LambdaDB stores the returned documents separately and responds with HTTP 200, isDocsInline: false, an empty docs array, and a presigned docsUrl. The inline threshold leaves room below 6 MiB for transport overhead; other query limits still apply.
For direct REST requests, check isDocsInline before reading docs. When it is false, send a GET request to the complete docsUrl without changing its query parameters or forwarding your LambdaDB API key. The downloaded JSON is an array of result items, not an object containing a docs field. An empty inline docs array in this case does not mean there are no results.
Fetch and list responses use the same download mechanism. The high-level collection methods in Python SDK 0.9.0, JavaScript/TypeScript SDK 0.5.1, and Go SDK 0.4.0 or later download these results automatically.
To reduce result size and download overhead, choose a lower size value,
or includeVectors=false (Python: include_vectors=False) to exclude vector values from the result.
