Stack
Five Constitutional Layers
Top to bottom: direction, governance, cognition, continuity, execution.
Human Intent
Direction, scope, responsibility.
ARG starts from human initiative rather than autonomous model goals. Human intent defines
what the system is allowed to pursue, what counts as success, and where responsibility lies
when the system acts.
Human intent is not a prompt and not a one-time instruction. It is the standing principle
that the system operates for a human reason that can be questioned, revised, or withdrawn.
Current public route: AI governance records
Chimera
Governance and coordination.
Chimera enforces stability conditions, refusal logic, and operational boundaries. It is the
deterministic layer that prevents the system from drifting into actions that violate human
intent, and it stabilizes multi-model work under pressure.
Chimera is not a content filter and not a list of forbidden topics. It is a layer of structural
rules with consequences: actions that violate the rules do not happen.
Current public route: Chimera publication
Meta-Fleet
Distributed cognition across multiple models and roles.
Different models contribute different kinds of cognition: search, synthesis, critique,
generation, and validation. Meta-Fleet coordinates those contributions under one frame,
with clearly defined roles.
Meta-Fleet is not an ensemble and not a voting system. A model assigned to critique does not
also vote on the final synthesis; cognitive diversity is structured, not averaged away.
Current public route: I.I.L. publication
Mnemosyne
Persistent memory and continuity.
Research persists across sessions. Mnemosyne names the long-duration retention layer: how
findings, decisions, and revisions accumulate over time and re-enter future work instead of
dissolving when a session closes.
Mnemosyne is not a chat history and not personalization. What enters memory must be curated,
retrievable, and subject to revision; memory does not substitute for intent.
Current public route: ARG Knowledge Graph
Agent Engineering
Bounded execution under constraint.
Agent Engineering is the task-facing layer where AI systems actually do things: write code,
run analyses, retrieve data, and draft documents. It operates with structured outputs,
defined tool use, and boundaries enforced by Chimera above.
Agent Engineering is not unconstrained generation. Agents do not pursue goals beyond their
assigned scope, and the execution layer never writes the law that governs it.
Current public route: AR-LAB artifacts