Test llm_reduce with images and GROUP BY clause.

Test llm_reduce with multiple input columns.

Test llm_reduce with audio transcription using OpenAI. The audio content says: 'Flock transforms DuckDB into a hybrid database and a semantic AI

Test llm_reduce with batch size configuration.

Test llm_reduce with GROUP BY clause.

Test llm_reduce with image data integration.

Test llm_reduce with custom model parameters.

Test llm_reduce with special characters and unicode.

Test llm_reduce with structured JSON output.

Test that Ollama provider throws error for audio transcription in llm_rerank.

Test basic llm_rerank functionality without GROUP BY.

Test llm_rerank behavior with empty table.

Test error handling with empty prompt.

Test error handling with non-existent model.

Test error handling with missing required arguments.

Test llm_rerank with multiple images in batch processing.

Test llm_rerank with images and GROUP BY clause.

Test metrics for llm_rerank aggregate function

Test llm_rerank with multiple ranking criteria.

Test llm_rerank with audio transcription using OpenAI. The audio content says: 'Flock transforms DuckDB into a hybrid database and a semantic AI

Test llm_rerank with batch size configuration.

Test llm_rerank with GROUP BY clause.

Test llm_rerank with image data integration.

Test llm_rerank with custom model parameters.

Test llm_rerank with special characters and unicode.

Test that both scalar and aggregate functions are tracked separately

Test that multiple aggregate function calls get sequential numbering

End-to-end: one request over the per-minute cap waits for the window to roll, then succeeds. Unit tests cover limiter math and wiring with mocks;

Test that multiple calls of the same function get sequential numbering

Sync batch: cumulative completion_tokens from provider usage exceeds completion_tokens_limit.

Async with batch_size 16 and 8 rows: all rows fit in one HTTP request.

Async with batch_size 16: one batch (16 rows) fits, two batches (20 rows) exceed the quota.

Sync batch_size 1: sequential requests accumulate tokens until the quota is exceeded.

Sync batch: cumulative prompt_tokens from provider usage exceeds prompt_tokens_limit.

A single completion stays under a generous cumulative token quota.

Native-specific helpers