6. IVM

6.1 Instance Virtual Machine (IVM)

The protocol’s long-term architecture includes an Instance Virtual Machine (IVM) -- a standardized execution environment that defines how AI instances operate within the network. The IVM represents Synaptika’s vision for a universal runtime specification for decentralized AI workloads.

The IVM specification encompasses:

• Model loading and initialization: Standardized procedures for deploying models onto nodes within a pool, ensuring consistent behavior regardless of the underlying hardware.
• Request handling: Defined interfaces for receiving, processing, and returning inference requests across the network.
• State management: How instance state (context windows, session data, fine-tuning parameters) is maintained across requests and across node failovers.

The IVM concept extends naturally toward agentic workloads -- AI instances that do not merely respond to inference requests but execute multi-step tasks, maintain persistent state, and interact with external systems. The IVM provides the sandboxed execution environment for such workloads, with the pool handling resource allocation and the protocol handling verification and billing.

The IVM specification will be developed iteratively based on testnet learnings and ecosystem feedback. A detailed technical specification will be published separately.