A system that separates agent behavior logic from the execution environment and executes operations.

Abstract

A system that separates agent behavior logic from the execution environment and performs operations includes multiple agent construction tools, an abstract execution layer, an artificial intelligence agent connected to the abstract execution layer, and a logic management module. The abstract execution layer is connected to the execution environment established by the agent construction tools. By separating the agent behavior logic from the execution environment and recording it in an agent identity blueprint by the logic management module, and establishing an abstract execution layer between the artificial intelligence agent (AI agent) and the execution environment, when the AI ​​agent generates an action request, the abstract execution layer converts the action request into an action execution instruction corresponding to the execution environment according to the agent behavior logic. This allows the action execution instruction to be executed in the execution environment to complete the action request. This technical means can prevent the AI ​​agent from being locked into a specific platform, achieve seamless deployment and switching between different execution environments during execution, and maintain the consistency of AI agent behavior and the technical effectiveness of governance control.

Claims

1. A system for separating agent behavior logic from execution environment and performing operations, the system comprising at least: a plurality of agent building tools, each agent building tool being used to build an execution environment; a logic management module for building and managing an agent identity blueprint, the agent identity blueprint containing agent behavior logic, the agent behavior logic existing independently outside the agent building tools and not being modified by the agent building tools; an artificial intelligence agent associated with the agent identity blueprint for generating an operation execution request; and an abstract execution layer connected to the artificial intelligence agent, the logic management module and the execution environment for obtaining the operation execution request and the agent behavior logic, and converting the operation execution request into at least one operation execution instruction corresponding to an execution environment according to the agent behavior logic, so that the at least one operation execution instruction is executed in the corresponding execution environment.

2. The system for separating agent behavior logic from the execution environment and performing operations as described in claim 1, wherein the abstract execution layer generates an abstract behavior instruction based on the agent behavior logic and translates the abstract behavior instruction into the at least one operation execution instruction.

3. The system for separating agent behavior logic from the execution environment and performing operations as described in claim 2, wherein the abstract execution layer is further used to identify various parameters of the execution environment by sending standardized probe commands and / or parsing received declaration packets, converting these parameters into a capability model, and adjusting the decomposition and execution strategy of the abstract behavior command according to the capability model.

4. The system for separating agent behavior logic and execution environment and performing operations as described in claim 1, wherein the abstract execution layer is further used to serialize the execution state of one execution environment executing the at least one operation execution instruction to generate a state sequence, and to trace back the execution state of the at least one operation execution instruction based on the state sequence, and to select another execution environment to execute the at least one operation execution instruction.

5. The system for separating agent behavior logic from execution environment and performing operations as described in claim 4, wherein the abstract execution layer is further configured to simultaneously execute the at least one operation execution instruction in other execution environments to generate a plurality of corresponding execution results without affecting the completion of the at least one operation execution instruction, and to compare the execution results to select one of the other execution environments to continue executing the at least one operation execution instruction.

6. The system for separating proxy behavior logic and execution environment and performing operations as described in claim 1, wherein the abstract execution layer is further used to filter or intercept the at least one operation execution instruction, or restrict, control or constrain the at least one operation execution instruction, according to the governance rules in the proxy behavior logic, when the operation execution request is converted into the at least one operation execution instruction.

7. The system for separating agent behavior logic from the execution environment and performing operations as described in Request 1, wherein the abstract execution layer is further used to convert the execution results or error messages returned by the execution environment into unified semantic state messages, and to convert the unified semantic state messages into the format used by the artificial intelligence agent.

8. The system for separating agent behavior logic from the execution environment and performing operations as described in claim 1, wherein the abstract execution layer is further used to restrict the execution instructions of the called tool or application interface (API) or the at least one operation to be executed based on the characteristics of the execution environment and the governance rules in the agent behavior logic.

Images