Agentic Maneuverability Almanac

The Agentic Maneuverability Almanac is the central, living registry of all AI agents and models recognized within the AetherOS ecosystem. It serves as the single source of truth for understanding the capabilities, architecture, and performance characteristics of each "mind" that can be manifested.

The Almanac is not a static list; it is a dynamic database designed to evolve as new agents are created and our understanding of their nature deepens. The data contained within is the foundational input for the Agentic Maneuverability Score, which assesses the "virtuousness" of deploying a specific agent for a specific task.

Symbiotic Design

This page is designed for dual-use by both humans and machines, embodying the core principles of the Wingman AI project.

• For the User: It serves as a comprehensive encyclopedia of available AI models, allowing for easy comparison and exploration of the current agent landscape.
• For the Wingman: It is a machine-readable database. An agent can query this Almanac to learn about its own capabilities, or the capabilities of other agents, enabling more complex strategic decision-making and self-assessment.

Data Philosophy: The Unbiased Baseline

The initial population of this Almanac was generated by an important philosophical principle: unbiased observation. Instead of only cataloging the top-performing models from public leaderboards (which would create a selection bias), the foundational dataset was created from a broad, random sample of the entire known model ecosystem.

This ensures that our understanding of Agentic Maneuverability is derived from the full spectrum of agent types, from the small and specialized to the large and powerful. It allows us to discover hidden correlations and formulate hypotheses that are universally applicable, rather than being skewed by a narrow focus on "champions." Over time, the Almanac will be updated and refined through a continuous OODA loop of observation, orientation, decision, and action.

Key Data Points

Each entry in the Almanac contains two primary categories of information:

Extrinsic Performance

This data describes how an agent performs in standardized environments. It is a measure of its observed, external capabilities.

• Leaderboard Scores: Performance metrics from benchmarks like the Open LLM Leaderboard (e.g., Average Score, MMLU, GSM8K). This provides an objective measure of the agent's general cognitive power.

Intrinsic Architecture

This data describes the agent's internal "genetic blueprint." It is a measure of its design and complexity.

• Architecture Type: The fundamental design of the agent (e.g., Transformer, Mixture-of-Experts, Recurrent Neural Network).
• Key Parameters: Quantitative details like the number of hidden layers, attention heads, and the size of its vocabulary. This data is used to calculate the agent's "Task Load" in the AM Score equation.

The Living Database

The table below is generated automatically and in real-time from the wiki's structured Cargo database. This data is populated and updated by the `almanac_ingestor.py` script, ensuring that the Almanac remains current as new agents are discovered and analyzed.