Which API for general real-time LLM agents [D]

A Reddit user inquired about the optimal interface for developing custom real-time LLM agents, noting that existing "real-time APIs" from vendors are primarily for coding. They highlighted the "AsyncLLM preprint 2609.35427" as the closest relevant work, which proposes using asyncio for coroutines with shared memory blocks to facilitate agent communication. The discussion centers on finding a suitable interface for building asynchronous agents in general, drawing from concepts like System 1 LLMs and various interaction models.