OpenAILLM(LLMBase) does not use AsyncOpenAI; AsyncMemory still runs sync LLM via asyncio.to_thread

[cainiaowenhou]

🚀 The feature

Inside AsyncMemory, the LLM call is still synchronous and only wrapped with:

asyncio.to_thread(self.llm.generate_response)

This means the LLM workload is executed in a thread pool, not asynchronously,

Motivation, pitch

Thanks

[cainiaowenhou]

class Memory(MemoryBase) should initialize its client using:

self.client = OpenAI()

because it operates synchronously.

Meanwhile, class AsyncMemory(MemoryBase) should instead use:

self.client = AsyncOpenAI()

to ensure that asynchronous memory operations rely on a true async OpenAI client rather than wrapping synchronous calls with asyncio.to_thread. This change would allow AsyncMemory to support genuine non-blocking LLM calls and improve concurrency performance in async applications.