Skip to main content

​
AEON Logic Model

OpenAEON defines AEON as:
  • Open: inspectable, extensible, and operator-controlled
  • Eternal Evolution: continuously adaptive over long-running sessions
This page describes the implemented runtime logic, not aspirational messaging.

​
Core loop (implemented)

AEON runs as a verifiable five-stage loop:
  1. Perceive
    Ingest session state, runtime telemetry, task intent, and model/tool outputs.
  2. Adjudicate
    Apply guardrail and policy decisions (maintenanceDecision, guardrailDecision, reasonCode) with epistemic labels.
  3. Act
    Execute agent/tool work under policy intensity and safety constraints.
  4. Persist
    Record delivery outcomes (persisted / persist_failed), memory checkpoint signals, and decision logs.
  5. Trace
    Expose structured state for operators and UI via AEON control APIs.

​
Safety-first decision semantics

Decision state is emitted as typed, inspectable data:
  • selfKernel: identity continuity, drift, calibration, integrity state
  • epistemic: confidence labeling and uncertainty framing
  • impactLens: multi-scale benefit/risk and reversibility
  • decisionCard: why / why-not / counterfactual / rollback
This keeps optimization constrained by trust and reversibility rather than raw autonomy.

​
Runtime inspection APIs

The Control UI and operators read AEON through:
  • aeon.status (schema v3 with compatibility mirrors)
  • aeon.memory.trace
  • aeon.execution.lookup
  • aeon.thinking.stream
See Control UI and RPC reference for request/response usage patterns.

​
Session durability and operator outcomes

AEON is designed to avoid “worked overnight but no result” failures by making delivery and traceability first-class:
  • explicit delivery state transitions
  • persisted artifact references for lookup
  • memory persistence metadata (lastDistillAt, checkpoint, totalEntries, lastWriteSource)
  • UI-visible runtime source/timestamp for telemetry blocks

​
Memory logic (implemented)

AEON memory is intentionally layered:
  1. In-memory working tail
    A bounded cognitive log supports fast runtime/UI interaction.
  2. Persisted cognitive stream
    Events are appended to per-scope JSONL storage (session + agent), enabling replay and audit.
  3. Distillation checkpointing
    Distillation advances checkpoint state and appends checkpoint markers instead of erasing MEMORY.md.
  4. Memory trace contract
    Runtime memory state is exposed via aeon.status and aeon.memory.trace.
This balances short-latency operations with long-horizon continuity.

​
Relationship to evolution log

The evolution log is the historical timeline.
This model page is the system contract.