ASSIST-IoT will provide an innovative reference architecture, envisioned as a decentralized ecosystem, where intelligence is distributed among nodes by implementing AI/ML close to data generation and actuation, and hyperconnecting nodes, in the edge-cloud continuum, over softwarized smart network. Smart network will be realised by means of virtualized functions, with clear separation of control and data planes, facilitating efficient infrastructure programmability. Moreover, the action will follow a DevSecOps methodology to ensure the integration of security, privacy, and trust, by design, in all aspects of the envisioned ecosystems. ASSIST-IoT will be supported by several pillars: (i) innovative IoT architecture, to adapt to the NGI paradigm, with three-dimensional approach, including intelligence, security and privacy by design, supporting decentralized collaborative decision-making; (ii) moving from semantic interoperability to semantically-enabled cross-platform, cross-domain data transactions, within decentralized governance, DLT-anchoring transaction security, privacy and trust; (iii) development and integration of innovative devices, supporting context-aware computing, to enable effective decision making close to events; (iv) introduction of self-* mechanisms, supporting self-awareness and (semi-)autonomous behaviours across IoT deployments, and (v) Tactile Internet support for latency applications, like AR/VR/MR, and human-centric interaction with IoT components. Results of the action will provide foundation for a comprehensive practice-based methodology, for future designers and implementers of smart IoT ecosystems. Finally, to validate research results, and developed solutions, and to ensure their wide applicability, extended pilot deployments with strong end-user participation will take place in: (i) port automation; (ii) smart safety of workers, and (iii) cohesive vehicle monitoring and diagnostics, bringing about domain-agnostic aspect of the approach.