Skip to main content

​
Basic Settings

1. Enable AI

  • Locate the “Use AI” toggle in settings
  • Switch it to the enabled state (purple)
Supported model providers:
  • OpenAI: Official models and API
  • AzureOpenAI: Microsoft Azure platform
  • Anthropic: Claude series models
  • DeepSeek: DeepSeek series models
  • Gemini: Google Gemini models
  • Grok: xAI Grok models
  • Ollama: Local deployment models
  • OpenRouter: Multi-model aggregation platform
  • Choose corresponding AI model based on provider
  • Recommended models:
    • OpenAI: gpt-4o, gpt-4o-mini
    • Anthropic: claude-3-5-sonnet
    • DeepSeek: deepseek-chat
  • Enter API key in the “API Key” input field
  • Ensure key is valid and has necessary permissions
  • Different providers require different key formats
  • Custom API endpoint address
  • Used for proxy or self-deployed services
  • Leave empty to use default endpoint

​
Embedding Settings

Embedding Model

  • Select model for generating text embeddings
  • Support independent embedding API key and endpoint
  • Recommended models:
    • OpenAI: text-embedding-3-small, text-embedding-3-large
    • Voyage: voyage-3-lite, voyage-large-2
  • Rebuild: Incremental update of embedding index
  • Force Rebuild: Complete regeneration of all embeddings
  • Index rebuild required after changing embedding model
  • Set the dimension size of embedding vectors
  • Auto-detect dimensions for common models
  • Manual setting required for custom models

​
Search Optimization Settings

[Search Settings Screenshot Needed]
  • Number of most relevant chunks returned by embedding search
  • Range: 1-20 (recommended: 3-5)
  • Higher values include more context
  • Minimum similarity score threshold for embedding search
  • Range: 0.0-1.0 (recommended: 0.4-0.7)
  • Higher values ensure more relevant matches
  • Use large language model for result reranking
  • Improve relevance and accuracy of search results
  • Note: Currently requires large language models, AI SDK framework doesn’t support dedicated rerank models yet
  • Rerank Top K: Number of results processed by reranking
  • Rerank Score: Minimum score threshold after reranking
  • Use Embedding Endpoint: Whether to use independent embedding API endpoint

​
Proxy Settings

  • Enable HTTP proxy to access AI services
  • Configure proxy host, port and authentication
  • Suitable for network-restricted environments

​
Test Connection

  • Click “Test Connection” to verify configuration
  • Check validity of API key and endpoint
  • Ensure network connection is working
Important Notes:
  • Rerank model currently requires large language models (like GPT-4)
  • Future versions will support dedicated rerank models
  • Must rebuild index after changing embedding model
⌘I