Interface ToolSearchStrategy

All Known Implementing Classes:
SimpleToolSearchStrategy, VectorToolSearchStrategy
@Experimental public interface ToolSearchStrategy
Strategy for discovering tools dynamically by an LLM during an AI Service invocation.

A ToolSearchStrategy allows reducing token usage by avoiding sending all available tools to the LLM upfront. Instead, the LLM is initially exposed only to one or more tool search tools, which it can call to discover relevant tools on demand.

Once the LLM performs a tool search, the matching tools are included in subsequent requests, typically accumulating until the corresponding search result message is evicted from the chat memory.

LangChain4j provides out-of-the-box implementations such as SimpleToolSearchStrategy and VectorToolSearchStrategy, but custom strategies can be implemented to support alternative discovery mechanisms.

Since:
1.12.0
See Also:

  • Method Details

    • getToolSearchTools

      List<ToolSpecification> getToolSearchTools(InvocationContext invocationContext)
      Returns one or more tool specifications that are exposed to the LLM and are used to perform tool searches.
      Parameters:
      invocationContext - the AI Service invocation context
      Returns:
      list containing one or more tool search tools

    • search

      ToolSearchResult search(ToolSearchRequest toolSearchRequest)
      Executes a tool search request generated by the LLM and matching it against searchable tools.
      Parameters:
      toolSearchRequest - the tool search request
      Returns:
      the search result containing matching tool names