Multi-agent operation management and control method and electronic equipment

By constructing target runtime domains and shadow runtime domains for pre-simulation verification and resource binding proof, the problems of resource mismatch, memory pollution, and streaming response consistency in multi-agent systems are solved, thereby improving the stability and response performance of the system and reducing the risk of anomalies.

CN122472228AActive Publication Date: 2026-07-28国家超级计算天津中心
6 Cites 0 Cited by

Patent Information

Application Number
CN202610977205.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-07-02
Publication Date
2026-07-28
Estimated Expiration
2046-07-02

AI Technical Summary

Technical Problem

Existing multi-agent systems suffer from resource mismatch risks, memory pollution spread, lack of consistency in streaming responses, and lack of pre-verification and traceability issues in multi-user, multi-session, and multi-agent concurrent scenarios, resulting in poor system stability and response performance.

Method used

By constructing a target execution domain at the logical level and a shadow execution domain at the physical level, pre-run verification and resource binding proof are performed to ensure the resource configuration and context consistency of the target intelligent agent. Resource binding proof is generated and verified on the business side to ensure the unified representation and verification of the execution environment.

Benefits of technology

It improves the stability and responsiveness of multi-agent systems in high-concurrency scenarios, reduces the risk of abnormal content entering model output or long-term memory, and enhances the security and traceability of the system.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention provides a multi-agent operation control method and electronic equipment, and the method comprises the steps: receiving a user request, and determining a current task identifier, a target agent and an operation resource bound with the target agent according to the user request; constructing a target operation domain of a logic level; before the target agent is called, constructing a shadow running domain of a physical level, and verifying whether a resource entity loaded in a running environment is consistent with resource data limited by the target running domain or not; generating a resource binding proof, and verifying whether a binding relationship declared in an agent registry exists among an assistant identifier transmitted by a service side, an agent actually loaded at a bottom layer and resources of the agent; and if the verification is passed, calling the target agent to execute model reasoning. The problem that the execution environment in the multi-user, multi-session, multi-thread and multi-agent concurrent service is difficult to uniformly characterize and verify is solved, and the stability and response effect of multi-agent operation are improved.
Need to check novelty before this filing date? Find Prior Art