This paper introduces a novel approach to real-time health monitoring of rolling bearings using a diffusion-based model. We propose a method that leverages Denoising Diffusion Implicit Models (DDIM), exploiting their intrinsic denoising capability, to process frequency spectra of vibration signals and construct a health index without prior system information. Our approach employs a continuous-time DDIM to model the complex distribution of frequencies in bearing vibration signals, enabling the detection of subtle spectral changes indicative of incipient faults. The method operates in real-time, adapting to specific bearing characteristics and learning normality over time. The health index (HI) is constructed using the reconstruction error between input features and their denoised versions and standardized using Welford's algorithm for online mean and variance estimation. We validate our method on the XJTU-SY bearing dataset. Results demonstrate the efficacy of our diffusion-based indicator, particularly in complex degradation scenarios. The method's ability to capture complex distributions and early degradation signs without prior system information presents a promising tool for predictive maintenance.