Overview of the Osiris Beta Platform

The Osiris platform contains a number of critical components that work together to enable a decentralized network of AI services to flourish. The core components include many architectural components that allow for a functional, scalable, and extensible system. We arrived at this architecture through a careful process guided by a few key decisions governing blockchain interactions, AI service integration, and abstraction and by the goal of building an AI marketplace that is both open and compliant with regulatory and legal requirements.

First, we made the conscious choice to minimize our dependence on our current blockchain, Binance smartchain. Both conceptual and practical issues motivated this decision. Conceptually, we desire to be blockchain-agnostic and will consider building our own consensus algorithm based on reputation. The speed, reliability, and costs of Binance blockchain interactions dictate that any scalable system built on top of it must minimize gas costs and the delays introduced by block-mining time. These decisions are reflected in our use of tools to abstract away all blockchain interactions (the daemon, CLI, and SDK) and in our use of a multi-party escrow contract and atomic unidirectional channels for payments.

Second, on AI services integration, we wanted to abstract away as much of the network as possible, from an AI developer's perspective, in order to reduce the learning curve and minimize the overhead associated with providing AI services via the network. Moreover, we wanted to achieve this abstraction with a single flexible tool that also helps us provide scalability, robustness, and distribution and management features. This is achieved by the daemon, which is a sidecar proxy used to communicate with services and the network and which will soon also allow services to very easily find and call other services.

Finally, to make our marketplace compliant with regulations without compromising on openness, we implemented a fully decentralized registry of AI services available on the platform. The AI marketplace supplements that registry (a smart contract) with a centralized source for which services are curated; that is, they have gone through a due-diligence process covering the service owners and the nature of the service being offered.

The diagram below depicts the key components along with auxiliary components and their roles.

For a developer who wants to offer an AI service over the network, the most crucial component is the Osiris daemon, an adapter and proxy that abstracts away interactions will all other components. The daemon handles interactions with smart contracts and payments, takes care of client request validation, and does other useful tasks, allowing AI developers to focus almost exclusively on the AI-related aspects of their server-side software and services. The daemon is a sidecar proxy, so one daemon instance is deployed next to each AI service instance.

For end users who want to purchase access to the AI services available in the platform, the most important component is the Marketplace DApp, through which they can search and browse a collection of curated services (i.e., services approved by the Osiris Foundation as relevant and high quality and whose owners have signed user and data-privacy agreements) for a large and ever-growing variety of AI tasks. The Marketplace DApp also handles payment for services (through MetaMask integration) and service ratings.

For application developers who want to use the network’s intelligence in their applications, the key component is the Osiris SDK, which automatically compiles client-side code for interacting with the platform and with specific services, allowing service requests to be coded in a straightforward way and supporting payment and interactions with the blockchain.

The Binance blockchain is used to host two critical smart contracts: the Registry and the Multi-Party Escrow.

The Registry is where AI service providers register on the platform, which involves providing text descriptions and tags to allow users to discover their service, pricing information, and information such as gRPC models and endpoint locations to allow users to call their services.

The Multi-Party Escrow contract handles payments through escrow accounts for each user (end users and applications) coupled with atomic unidirectional channels for faster and cheaper transactions.

Those are the core components of the platform. Two key support components are also worth mentioning:

● The AI developer– and owner-oriented CLI (command-line package) provides command line APIs for a number of crucial service developer and service owner tasks: registering and managing identities, publishing services, updating registration information, notifying the platform of new endpoints, managing payment channels and balances, and calling services.

● The Request for AI Portal (RFAI) is a DApp through which end users and application developers can request specific AI services they would like added to the network and stake OSR tokens as a reward for high-quality solutions.

Last updated