Skip to main content

Web Search

Web search allows LLMs to access real-time information from the internet during inference. This enables the model to provide up-to-date answers based on current information rather than relying solely on its training data.

Overview​

To enable web search in your LLM requests, you need to:

  1. Initialize a client with API key authentication
  2. Add web_search: true to the extra_body parameter in your chat completion request
  3. The model will automatically search the web and incorporate relevant information into its response

Basic Usage​

examples/3-web-search.ts
loading...

How It Works​

When web search is enabled:

  1. The model analyzes your query to determine if web search would be beneficial
  2. If needed, it formulates search queries and retrieves relevant information from the web
  3. The retrieved information is incorporated into the model's context
  4. The model generates a response that combines its knowledge with the web-sourced information

Use Cases​

Web search is particularly useful for:

  • Current events: Getting the latest news and updates
  • Real-time data: Accessing current prices, statistics, or live information
  • Recent developments: Information about topics that have evolved since the model's training cutoff
  • Fact verification: Cross-referencing information with current sources
  • Dynamic content: Queries that require up-to-date information

Important Notes​

  • Web search adds latency to requests as the model needs to fetch and process external information
  • The quality of responses depends on the relevance and accuracy of web sources found
  • Not all queries will trigger a web search - the model determines when it's necessary
  • Web search works with both API key and delegation token authentication flows