CORTEX (Memory Unit)

0. Canonical role

CORTEX is the Heptagon memory unit for long-term behavioral context in governed agent execution.

1. Classification in the ecosystem

Behavioral memory and anomaly-detection layer.

CORTEX should be understood as the component that tracks historical agent behavior over time and contributes contextual risk signals to the runtime decision path.

2. What this project is

CORTEX is not presented here as a generic consumer memory product or a stand-alone memory cloud. In the ABS ecosystem, its documented role is to learn baseline tool-use patterns per agent, detect deviations, and feed anomaly signals into the policy evaluation flow.

3. What the current technical material shows

The current technical documentation describes a baseline built from authorized actions, with normal tools, normal sequences, and normal frequencies tracked per agent.

The current anomaly model documents four signal types:

• novel tool,
• unusual sequence,
• frequency spike,
• and time anomaly.

These signals contribute to an anomaly_score, which is added to risk_score before policy evaluation.

4. Functional role

CORTEX should be described as:

• a long-term behavioral memory layer for agents,
• a detector of behavioral drift across sessions,
• and a contextual risk input to governed authorization decisions.

5. What it is not

CORTEX should not be described here as a generalized Memory-as-a-Service business, a semantic graph platform, or a universal RAG infrastructure layer unless those capabilities are separately proved and documented. That positioning is broader than the current technical material clearly supports.

6. Deployment posture

CORTEX fits the broader ABS model of self-hosted or customer-controlled deployment. It should be presented as part of a governed runtime stack rather than as a public SaaS memory product.

7. Documentation discipline

When documenting CORTEX, prioritize the verified flow: sequence capture, baseline establishment, anomaly detection, retention, and contribution to runtime risk. Broader roadmap ideas should be clearly marked as experimental or future work.