In craft ai, you create, at least, one agent per user. Each agent is independent and knows about the context of its user: it only takes personalized decisions.

Based on the history of an agent’s context, craft ai uses a range of machine learning techniques to generate a decision tree defining which decision needs to be taken in a given context; this model is no blackbox: it is easy to understand and debug.

craft ai does the heavy lifting in the cloud: the context history analysis and learning. However, the decision-taking part can be embedded, even in low power devices, to reduce latency and be resilient to network hiccups.