Unlike keyword search, semantic search understands query meaning, finding related content without exact matches.
How it works: Documents chunked → Vectorized via Embedding → Stored in HNSW index → Queries vectorized and matched.
Ask AI naturally: “Based on the knowledge base, what are the advantages of XX method?” — AI auto-retrieves and answers.
Embedding Models
| Model | Best For |
|---|---|
| text-embedding-v4 (Bailian) | Chinese |
| text-embedding-3-large (OpenAI) | English |
| nomic-embed-text (Ollama) | Local/offline |