Execute a semantic similarity search using dense vector embeddings.

This endpoint retrieves items by meaning rather than exact keyword matching. It supports two modes:

Text Mode:

• Provide a text string (minimum 5 characters)
• The server generates a 768-dimensional embedding using the multilingual-e5-small model
• Results are ranked by cosine similarity to the generated embedding

Vector Mode:

• Provide a pre-computed 768-dimensional vector
• Useful when you have embeddings from another source or want to cache/reuse them
• Results are ranked by cosine similarity to your provided vector

You must provide either text OR vector, but not both.

Result Ordering: Results are ordered by similarity score (highest first). Scores represent cosine similarity between the query and document vectors, ranging from 0.0 to 1.0 where 1.0 is identical.