Implementing online learning on event driven-neuromorphic systems requires (i) an event-driven learning local algorithm that calculates the weight updates on streaming data, (ii) mapping the weight updates onto limited bit precision memory, and (iii) an always-on function that does not need to separate between the training and inference phase. Recent neuroscientific studies have shown how dendritic compartments solve these problems in biological brain. Inspired by these studies and with the aim to build low-power online learning hardware, in this paper, we introduce ALIVE. ALIVE is a prototype neuromorphic chip designed and fabricated in 180,nm technology that implements always-on stochastic dendritic-based online learning. We present an algorithm-circuits co-design approach and show circuits simulations that implement the learning rules. We validate our system by performing system-level simulations which reach above 85% accuracy on MNIST using only one layer network and 4-bit weight precision weights.