A target tool invocation method and device, electronic equipment and storage medium

By using simplified view matching and dynamic replacement of complete views in a large language model, the problems of resource waste and interference when calling massive tools are solved, thus improving calling efficiency and security.

CN122364384APending Publication Date: 2026-07-10NINGBO TELIAN INFORMATION TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
NINGBO TELIAN INFORMATION TECH CO LTD
Filing Date
2026-04-10
Publication Date
2026-07-10

AI Technical Summary

Technical Problem

In existing technologies, large language models suffer from resource waste and interference when calling massive amounts of tools, resulting in high inference costs, increased latency, and reduced execution security and robustness.

Method used

By performing tool retrieval in a pre-built vector database, a simplified view of candidate tools is obtained and injected into the context of a large language model. Only required parameters are matched. After the target tool is determined, the full view is replaced and invoked. This ensures that the similarity of tools in the vector database is below the threshold and reduces the character usage in the context window.

Benefits of technology

It achieves high-accuracy target tool invocation, reduces token consumption and inference speed, improves the robustness and security of the intelligent agent system, and avoids resource waste and information interference.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122364384A_ABST
    Figure CN122364384A_ABST
Patent Text Reader

Abstract

This application relates to the field of large language model technology, and provides a method, apparatus, electronic device, and storage medium for invoking a target tool. The method involves obtaining a query statement, retrieving tools from a pre-built vector database based on the query statement to obtain at least one candidate tool corresponding to the query statement; obtaining a simplified view of each candidate tool and injecting each simplified view into the context of a large language model, driving the large language model to match the query statement with the simplified views to determine the target tool from at least one candidate tool; obtaining a complete view of the target tool, replacing the simplified view in the context of the large language model with the complete view, and generating and executing the target tool invocation instruction. In the initial screening stage, only simplified views are injected into the large language model, compressing the character space occupied by the context window. After the target tool is identified, the view is then replaced with the complete view.
Need to check novelty before this filing date? Find Prior Art