Supernet | AI OS SDK

The integration of the AI OS system will enable more developers to conduct secondary development based on SuperNet.

The Structure of Supernet Operating System.

Centralized Queue Management

Supernet Intelligent Node centralizes all task queues in the scheduler module, isolating request management from individual modules. This allows each module to focus on execution. The scheduler uses FIFO and Round Robin algorithms to prioritize and distribute tasks, with FIFO processing tasks in order, and Round Robin providing balanced resource distribution through time slicing.

Context Interrupt Mechanism

The context interrupt mechanism allows tasks to be paused and resumed without losing progress. It captures and restores task states based on text-based or logits-based decoding strategies, ensuring efficient execution even in complex tasks. This avoids unnecessary recomputation and improves task continuity.

Memory Management

Supernet Intelligent Node manages agent memory, which includes conversation history and execution results. It uses a trie-based compression system to optimize memory usage and employs a K-LRU eviction policy to manage memory between RAM and disk. This ensures efficient memory utilization and prevents data overflow.

Storage Management

The storage manager handles persistent data storage, such as files and agent memories. It swaps data between RAM and disk when memory usage exceeds limits, using trie-based compression for efficient storage. It supports both local file storage and vector databases like Chromadb for permanent data management.

Tool Management and Access Control

The tool manager dynamically loads and executes system tools, resolving conflicts with parallel access limits. The access manager controls permissions, ensuring agents only access authorized data. It also includes user intervention checks to prevent unintended operations, such as deletions or privilege changes.