KiHustle Cosine Similarity (Bag-of-Words)

Use this endpoint when you need a lightweight, fast, lexical text similarity score without embedding model overhead. It is ideal for quick deduplication checks, keyword-overlap scoring, or scenarios where semantic nuance is less important than token-level overlap. Prefer this over embedding-based similarity when cost, latency, or simplicity matters most, and when the texts share common vocabulary.

• Empty or missing text_a or text_b fields may return an error or zero similarity
• Very long texts may time out or produce unexpected results
• Non-string input types will likely cause a validation error
• Identical texts always return 1.0; completely disjoint vocabularies return 0.0 which may be misleading
• The generic response schema ({result, status}) may not expose the numeric score directly — caller must parse carefully