Sagas (AetherOS)

A Saga is a structured, narrative record of an AI agent's experience while performing a task. Unlike a simple log file, a Saga is an enriched story that includes not only the sequence of events but also contextual critiques and prescriptive advice. Sagas are the primary mechanism through which a Wingman AI learns and adapts.

They are generated by a high-level LLM that observes a lower-level agent's performance, creating a story that serves as a memory for future tasks.

Symbiotic Design

This page, like all pages on this wiki, is designed for symbiotic learning.

• For the User: It provides a clear explanation of how your Wingman learns from its successes and failures.
• For the Wingman: This page serves as a foundational memory about its own learning process. By reading this, the Wingman learns about the concept of Sagas themselves, reinforcing the importance of its own experiences.

Saga Generation Process

A Saga is generated at the conclusion of a task or simulation run. The `SagaGenerator` component takes the raw log file and contextual critiques from a "Guide" AI and synthesizes them into a series of commands.

Key Components

• Run Log: The raw, event-by-event history of the agent's actions (e.g., its path through a grid).
• Guide Critiques: Real-time feedback provided by a high-level LLM that observes the agent. This feedback provides crucial context about why certain actions were good or bad.
• Prescriptive Command (`SUGGERO`): If a task ends in failure, the Saga Generator creates a prescriptive `SUGGERO` command. This command provides a concrete, actionable suggestion for how the agent could have avoided the failure, turning a mistake into a learning opportunity.

AetherOS Commands

Sagas are written in a specialized command language for the AetherOS. This format is both machine-readable and metaphorically descriptive.

• CREO: "Create" - Establishes the context or agent persona.
• PERTURBO: "Disturb" or "Recount" - Describes the events of the journey.
• SUGGERO: "Suggest" - Provides prescriptive advice after a failure.
• OSTENDO: "Show" or "Present" - Concludes a successful run.
• REDIMO: "Redeem" or "Recover" - Concludes a failed run, framing it as a chance for redemption.
• vale: "Farewell" - The universal sign-off, ending every Saga.

This command structure ensures that every experience, whether a success or a failure, is framed as a coherent story that can be stored in the Wingman's associative memory.

Role in Aetheric Learning

Sagas are the fuel for the Aetheric Learning Loop. When a new task begins, the Saga from the previous run is used to perturb the agent's Animus (its core state matrix). The narrative of the last experience directly influences the agent's initial state for the next task, creating a persistent thread of memory and learning across experiences.

This process allows the Wingman to build upon its history, avoiding repeated mistakes and reinforcing successful strategies over time.