NuNet agents will support mobility of computational processes by enabling them to move between devices of the network and in this way enabling the dynamic optimization of computational workflows and business processes - bringing data closer to processes or processes closer to data. All computational processes are containers, which can be spun and installed at the location of choice. Choices of moving containers across the network will be made by workflow organizers (i.e. network operations agents), while NuNet will provide necessary APIs providing context awareness and computational reflection information as well as support installation and destruction of containers as instructed.

Note that mobility of computational agents provides not only for the greater efficiency of computational workflows by enabling processing closer to data, but also for completely new business models. For example, highly sensitive data can be processed at a client's site by planting containers with proprietary AI / ML algorithms within the boundaries of the client's private cloud and ensuring that no data is leaked. Containers with algorithms may be cryptographically signed and secured and NuNet will ensure that the container is destroyed when the computing job is done, in this way securing privacy and intellectual property of both data and computing intelligence providers.

Context awareness, provided by NuNet agents, will amount to awareness of location and proximity to other NuNet agents in the network. Proximity in this sense means the cost of collaboration with these agents and includes a list of parameters which may be agent specific and therefore not centrally managed.

Considering the wide variety of hardware devices, the network of NuNet agents will constitute a dynamic topology where physical location of resources may change over time, as well as resources can go offline and pop-up in other places. The topology of NuNet’s network, however, will be defined not by geographic distribution of physical resources, but rather by the relative costs of transferring data between agents in a workflow in terms of time and price.

Each NuNet agent installed in a particular device or resource will accumulate, keep and update this information upon request and provide it to computational agents via an open API. Note that NuNet agent will be radically decentralized in the sense that no meta-agent will control or have information about the whole network. Therefore, context awareness functionality will include methods of querying and learning local network topologies by individual NuNet agents, eventually supporting and enabling automatic or semi-automatic search and discovery of computational processes as required by clients or other processes in the ecosystem, independently of the network or physical location of processes.

Computational Reflection

NuNet will provide means for computational actors to exercise certain levels of computational reflection in terms of: (1) resource allocation, (2) data representation, (3) execution introspection:

• The physical resource allocation aspect of computational reflection will allow agents to have continuous interaction with their execution environments, and search and request for additional computational resources and infrastructures according to their own criteria. Additionally, agents will be able to download, or update required libraries, i.e. evolve their own execution strategies, and this will allow for agents’ free migration from one node (virtual machine, cloud vendor, private computer or a mobile phone) in a distributed computing environment to another;

• Capability and data representation aspect will allow agents to semantically represent their own computational capabilities and input and output data. This information will be made available for other agents to query when negotiating pairwise contracts and workflow designs;

• Execution introspection is the ability of each agent to monitor actual resource utilization by its algorithms, keep history of execution times and memory usage, and access its own state during execution, amongst other features. Agents may decide to share part of this information with the network in order to prove their capabilities and quality of services;

Note that the abstraction of NuNet does not define in any manner how computation or actions of agents will be performed. Using means of computational reflection, computational agents will be able to design and apply workflow design and workflow execution functionalities pertaining to their individual choice and requirements. Furthermore, a human element can be seamlessly incorporated into the same model. For instance, an agent can represent a UI through which tasks that need human intervention can be performed and integrated into the workflow. A hybrid computer-human collaborative case can be imagined where NuNet workflows formulate a computational task, which is then performed by humans through crowdsourcing or freelancing marketplaces (Amazon MTurk, Udemy), or even code hosting platforms (GitHub, Bitbucket, etc.) The tokenomy of NuNet will support and facilitate decentralized marketplaces where human and machine jobs will be demanded, offered and contracted on a commercial or other basis. Computational reflection will enable owners of resources to advertise and price their capabilities and for resource users to estimate, track and manage computation costs in a dynamic and transparent way.

NuNet will enable the network to connect computational resources, data and algorithms owned by different vendors into one network. NuNet adapters will enable to exercise payments to resource owners directly from the computational agents of distributed computing frameworks or for individual users using NuNet. The goal of NuNet is to overlay the computational network with the payments network and provide interoperability and exchange adapters for tokens and payment methods used by owners of computing resources, decentralized computing platforms and marketplaces.

The value exchange between independent network components will be based on an established smart contract platform and decentralized micro-transactions. Initial value exchange infrastructure will be implemented on Ethereum blockchain and soon after on Cardano, but NuNet architecture will allow it to work with different distributed ledger technologies as long as they provide smart contract and decentralized micro-transaction capabilities as required by the platform. Ultimately, when the platform will be at an advanced development stage, it may allow a network operation agent to choose a smart contract and transaction platform from several options and to extend the framework with new technologies appearing in the space.