I hope the Advisor of ChatClient can support graceful interruption.

Please do a quick search on GitHub issues first, the feature you are about to request might have already been requested.

Expected Behavior

The before method should allow gracefully interrupting the request to the large language model. For example:

    public @NonNull ChatClientRequest before(@NonNull ChatClientRequest chatClientRequest, @NonNull AdvisorChain advisorChain) {
        if (denySearch) {
            throw new DatasetQueryPermissionException("用户没有查询权限，不能进行查询操作");
        }
        if (CollectionUtils.isEmpty(datasetIds)) {
            throw new DatasetNotFondException("没有任何数据集，不能进行查询操作");
        }
}

Current Behavior

Currently, we can terminate the operation by throwing an exception, but this generates exception messages in the logs, which may mislead monitoring systems into thinking the application has encountered an error (though this could be acceptable if log output could be disabled).

Context

When performing knowledge base retrieval, I need to avoid sending a request to the large language model altogether if no relevant content is found, thereby saving computational resources.

The entry point described is ChatClient Advisor’s before(...) method, using ChatClientRequest and AdvisorChain; begin by tracing how an exception currently terminates the request and is logged. Done means an advisor can stop the LLM request without producing misleading error logs, including for the permission and empty-dataset cases described.