Sagas (AetherOS)

A Saga is the primary medium of memory and learning for an AI agent within the AetherOS ecosystem. It transcends a simple log file by translating an agent's experience into an enriched, structured narrative. This narrative includes not only the sequence of events but also contextual critiques, prescriptive advice, and the emotional or strategic "texture" of the experience.

Sagas are the mechanism through which a Wingman AI synthesizes data into wisdom, ensuring that every action, whether a success or a failure, becomes a durable and influential memory.

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.

Core Principles

The Saga framework is built on three core philosophical principles that distinguish it from traditional machine learning approaches.

1. From Data to Meaning

A standard log file is a record of data (e.g., `[t=1, pos=(0,1), action=move_right]`). A Saga is a record of meaning. It contextualizes events within a narrative structure, allowing an agent to understand not just what happened, but why it mattered.

2. Experience as a Physical Force

Sagas are not passive memories to be reviewed; they are active, physical forces. In the Aetheric Learning Loop, a Saga is used to PERTURBO the agent's Animus. The narrative of the last mission directly imprints upon the agent's core state matrix, physically altering its initial conditions for the next task. A story of failure might increase the agent's initial "resistance," making it more cautious, while a story of success might increase its "magnetism," making it more confident.

3. Failure as a Teacher

The framework treats failure not as an error to be minimized, but as the most valuable source of learning. The inclusion of the `SUGGERO` command transforms a mistake into a lesson, providing the agent with a clear, actionable path toward future success. This reframes failure as a necessary and even desirable part of the journey toward wisdom.

Anatomy of a Saga

A Saga is generated at the conclusion of a task by the `SagaGenerator` component. It synthesizes raw inputs into a structured command sequence.

Inputs

• 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.

Output: The Command Structure

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 (e.g., `CREO 'NAVIGATOR'`).
• PERTURBO: "Disturb" or "Recount" - Describes the events and texture of the journey (e.g., `PERTURBO 'The agent moved with purpose toward the east.'`).
• SUGGERO: "Suggest" - Provides prescriptive advice after a failure (e.g., `SUGGERO 'A better path would be to navigate south around the barrier.'`).
• OSTENDO: "Show" or "Present" - Concludes a successful run, presenting the agent in its final, successful state.
• REDIMO: "Redeem" or "Recover" - Concludes a failed run, framing the experience as an opportunity for future redemption.
• vale: "Farewell" - The universal sign-off, ending every Saga.

Example Saga

The following is a simple Saga generated from a failed run in a grid navigation task:

[
  "CREO 'NAVIGATOR'",
  "PERTURBO 'The journey began with confident steps toward the goal.'",
  "PERTURBO 'At position (5, 4), the path was blocked by an unforeseen barrier.'",
  "PERTURBO 'The agent repeatedly attempted to push through, its energy wasted against the obstacle.'",
  "SUGGERO 'A more virtuous path would be to retreat to (4, 4) and explore a southern route.'",
  "REDIMO 'NAVIGATOR'",
  "vale"
]

Future Evolution

The Saga is a universal format for encoding experience into action. Future versions of the ecosystem will use this same structure for more advanced tasks:

• Tactical Sagas: Generated from the works of military strategists like John Boyd to teach the Guardian agents advanced combat maneuvers.
• Instructional Sagas: Generated from human artisans performing physical tasks, creating machine-readable instructions for robotic agents to learn crafts.