Continuous Interaction Diffusion: A Diffusion-Native Runtime for Asynchronous Tool-Augmented Reasoning

Continuous Interaction Diffusion (CID) integrates external tool calls directly into the iterative denoising process of diffusion language models. Unlike autoregressive stop-and-resume patterns, CID enables asynchronous tool-augmented reasoning by separating model-read-only facts from continuous refinement. This architecture prevents premature tool calls and reduces redundant execution during parallel output generation.