Search Document Sentences

Search through document sentences using semantic similarity.

This endpoint performs a semantic search at the sentence level, which is best for:

• Finding very specific information or facts
• Getting precise answers to questions
• When you need the most granular search results

Performance Considerations:

• Slower search performance due to more vectors
• Highest memory usage
• Best for smaller, focused searches
• Consider impact on rate limits

Best Practices:

• Use only when specific facts are needed
• Keep document set focused using metadata filters
• Consider caching frequent queries
• Use smaller top_k values (5-10 recommended)
• Use min_relevance to filter out low-quality matches

Sentence-level search provides the highest precision but with minimal context. Each result is a single, focused statement or fact from the document.

You can provide either:

• A text query (which will be converted to embeddings)
• A pre-computed embedding vector (1536 dimensions)

Filter options:

• min_relevance: Set a threshold (0-1) to only return results above a certain relevance score