ARTFEED — Contemporary Art Intelligence

AstronOS: Unified Execution Model for Long-Horizon Agentic Systems

ai-technology · 2026-08-18

A new study on arXiv introduces AstronOS, an advanced framework designed specifically for long-term agent systems. This framework maintains a consistent identity and a versioned authoritative state for tasks, addressing the common issue of fragmentation in systems that prioritize individual dialogues or model interactions. Each step's inputs are matched with a specific state version and fresh data, and state changes only happen after careful validation. AstronOS uses Cases, Tasks, and Scenario Packs for both central and local execution. The research tested five strategies to integrate a software update plan into a new session, involving 150 executions across ten controlled tasks. You can find the paper on arXiv with the identifier 2608.16381.

Key facts

  • AstronOS is a unified execution model and runtime for long-horizon agentic systems.
  • It maintains persistent identity and versioned authoritative state across calls.
  • Each step receives input scoped to a specific state version and new material.
  • Results advance state only after validation and recording.
  • AstronOS implements selected paths using Cases, Tasks, and Scenario Packs.
  • Five strategies were compared for carrying a software-version update plan into a fresh model session.
  • Ten controlled tasks were run under all five strategies with three repetitions.
  • The study yielded 150 included executions.
  • The paper is available on arXiv with identifier 2608.16381.

Entities

Institutions

  • arXiv

Sources