[Proposal]: Standardize Non-blocking Behavior for ChatModelBase Implementations

[wuji1428]

AgentScope-Java is an open-source project. To involve a broader community, we recommend asking your questions in English.

Description

Currently, the five primary implementations of ChatModelBase (OpenAIChatModel, AnthropicChatModel, DashScopeChatModel, GeminiChatModel, and OllamaChatModel) exhibit inconsistent behavior when handling non-streaming calls.

The Core Issue

In non-streaming mode (stream=false), different implementation classes exhibit different blocking behaviors towards the calling thread:

Implementation	Non-streaming
OpenAIChatModel	Blocking
AnthropicChatModel	Non-blocking
DashScopeChatModel	Blocking
GeminiChatModel	Non-blocking
OllamaChatModel	Blocking

Why This Needs to be Addressed

1. Semantic Inconsistency: The same client code behaves differently depending on the underlying model. This leads to unpredictable application performance and inconsistent API semantics.
2. Developer Experience: Users expect a unified Reactive interface (returning Flux/Mono) to follow consistent non-blocking patterns.
3. Maintenance Complexity: It increases the cognitive load for developers integrating multiple models.

Proposal: Standardize on Non-blocking Mode

I propose that all ChatModelBase implementations should be updated to ensure that non-streaming calls do not block the calling thread.