This work presents a lightweight and physical layer-based key agreement system specific for resource-constrained and ubiquitous IoT devices. Unlike traditional cryptographic approaches based on mathematical difficult problems, this approach exploits inherent randomness in the wireless physical channel to derive shared secret keys between device nodes. An adaptive multi-bit quantization algorithm that improves entropy while avoiding repetitive patterns, and a lightweight reconciliation scheme inspired by Cascade algorithm, optimized via recursive hash verification are proposed. Various quantization and reconciliation algorithms were evaluated using metrics such as Bit Disagreement Rate (BDR), Shannon entropy, autocorrelation, and Hamming distance. The keys are validated through a simplified battery of NIST statistical tests adapted for short sequences, confirming their high randomness. In addition, we assess security against passive eavesdropping attacks and analyze resilience under adversarial models constrained by spatial decorrelation. Experimental results confirm that the system achieves secure key agreement using only physical-layer information, with minimal computational overhead and without relying on conventional cryptographic infrastructure, providing a viable solution for trusted communication in future IoT environments. This lightweight and infrastructure-independent architecture is well-suited to address the tight constraints in Ambient Intelligence and mHealth applications, where nonintrusive and efficient security is a must.