AI Semantic Search

AI semantic search is an intelligent search method provided by Cloudreve based on Meilisearch. Unlike traditional full-text search, semantic search returns results based on the meaning and context of the query, rather than relying solely on keyword matching.

How It Works

Semantic search generates vector embeddings through LLM providers (such as OpenAI, Hugging Face, etc.), converting document content and search terms into semantic vectors. It then finds semantically relevant search results by comparing vector similarity.

When semantic search is enabled, Cloudreve returns both full-text search and semantic search results by default (i.e., hybrid search), complementing each other to provide a more accurate search experience.

Full-text Search vs AI Semantic Search

	Full-text Search	AI Semantic Search
Matching	Exact keyword matching	Semantic similarity matching
Use Case	User knows the exact keywords, needs precise matching	Query is vague or lengthy, needs semantic understanding
Resource Usage	Low	May incur third-party model costs
Search Example	Searching "travel plan" only matches documents containing that keyword	Searching "travel plan" also matches documents containing "trip arrangement", "vacation itinerary", and other semantically related terms

Prerequisites

1. Full-text search is already enabled in Cloudreve.
2. Choose a suitable embedding model and provider, such as OpenAI, Hugging Face, etc. For guidance on choosing a model, refer to the Meilisearch documentation.

Enable AI Hybrid Search

Navigate to Cloudreve admin panel -> Filesystem -> Full-text Search -> AI Semantic Search to enable AI semantic search, and fill in the embedding configuration based on your model provider:

Currently, OpenAI offers the following three mainstream embedding models:

• text-embedding-3-large: 3,072 dimensions
• text-embedding-3-small: 1,536 dimensions
• text-embedding-ada-002: 1,536 dimensions

Fill in the following embedding configuration in Cloudreve:

{
  "source": "openAi",
  "apiKey": "<OpenAI API Key>",
  "dimensions": 1536,
  "model": "text-embedding-3-small"
}

Where:

• apiKey is your OpenAI API key.
• dimensions is the dimension of the embedding model, e.g., 1536.
• model is the name of the embedding model, e.g., text-embedding-3-small.

Reference: Meilisearch OpenAI Embedding Configuration

After saving the settings, there is no need to rebuild the index. Meilisearch will automatically generate embedding vectors for existing indexes using the embedding model.