PROTO-AXIS
Architecture for accountable machine intelligence
Identity. Meaning. Rules. Evidence. Coordination. Reasoning.
PROTO-AXIS is an independent research and architecture project by Roman Shaban. It explores how AI agents, knowledge systems and digital services can work together through explicit identities, shared definitions, bounded permissions and traceable decisions.
The ambition is to make complex machine activity understandable, reviewable and governable as systems become more autonomous.
Seven architectural roles. One coherent framework.
Why PROTO-AXIS
A capable AI model is one part of a larger system. Once an agent uses tools, exchanges information or acts on behalf of a person, practical questions become unavoidable:
Who is acting? What does the information mean? Which actions are permitted? What evidence supports a decision? Who can inspect, challenge or stop the process?
PROTO-AXIS organizes these questions into a shared architectural framework. Its purpose is to help turn isolated capabilities into systems whose behavior can be examined across the full path from request to result.
The Heptagram architecture
Heptagram is the project's seven-part architectural model. The nodes describe responsibilities; an implementation may combine them in one application or distribute them across several services.
0 — KERNEL | Coordination
The coordinating role: connect requests, system state and execution stages. Define how work begins, progresses, fails and stops.
1 — IDENTITY | Actors and authority
Describe the people, agents and services involved, their provenance and the permissions under which they act.
2 — ONTOLOGY | Shared meaning
Define terms, data structures and relationships so that information can be exchanged with explicit context and versioned definitions.
3 — ETHOS | Rules and boundaries
Represent the constraints that govern proposed actions, including permissions, policy checks and cases that require human review.
4 — VERUM | Evidence and records
Connect claims and events to sources, verification methods and recorded history. Keep uncertainty and conflicting evidence visible.
5 — NEXUS | Communication
Describe how components exchange requests, responses and status information, including delivery failures and coordination between agents.
6 — LOGOS | Reasoning and consistency
Examine whether conclusions follow from stated assumptions, identify contradictions and support review of competing interpretations.
From a request to a reviewable result
Consider an AI agent asked to update an organization's knowledge base.
The proposed workflow identifies the requesting person and acting agent, defines the intended change, checks access and policy constraints, and gathers the supporting sources. It then reviews the proposed update for inconsistencies and obtains human approval where required.
An authorized change is recorded together with its outcome. If evidence is missing or permission is denied, the process returns a clear reason for stopping.
The resulting record should make it possible to answer: what changed, who authorized it, which evidence was used and how the outcome was checked.
This is an illustrative design workflow, not a claim that a live service on this website currently executes it.
Design principles
Explicit scope. Every task should have a defined purpose, permitted actions and stopping conditions.
Traceable claims. Important conclusions should identify their sources, assumptions and verification status.
Bounded authority. Delegated agents should receive only the permissions needed for their assigned work.
Visible uncertainty. Missing evidence, disagreement and failed checks should remain visible throughout a workflow.
Human accountability. High-impact actions should have identifiable responsibility and appropriate review before execution.
Versioned meaning. Changes to definitions, rules and interfaces should be documented so that earlier decisions remain interpretable.
These principles guide the proposed architecture. Each implementation needs its own tests, threat analysis and operational evidence.
Research and implementation directions
The project's next useful artifacts are concrete and testable: a shared glossary, a versioned task-and-result format, a minimal agent workflow and a test suite covering both successful execution and controlled failure.
A proposed common record would include the actor, requested action, relevant definitions, permissions, evidence references, decision, result, timestamp and specification version.
Candidate applications include knowledge-base maintenance, coordination of research assistants, policy-aware automation and the exchange of structured educational knowledge.
The architecture should be evaluated through observable outcomes: whether unauthorized actions are blocked, evidence can be traced, failures are explained and reviewers can reconstruct a decision.
Related work
PROTO-AXIS belongs to Roman Shaban's broader work on AI systems, ontological engineering and digital infrastructure. Related project names include Catalyst-OS, onto-compliance-engine and EduOS / Edupro.Expert. Their documentation and implementation status should be assessed individually.
Project status
This website presents the PROTO-AXIS concept, architectural vocabulary and research directions. The seven nodes are architectural roles; listing them does not imply that seven production services are running.
Specifications, prototypes and validated capabilities should be identified separately as they are published. Performance, security and reliability claims require documented tests.
Technical discussion, critical review and reproducible experiments are welcome through the author's public channels.
Author and project identity
Project: PROTO-AXIS
Creator: Roman Shaban
Focus: AI agent coordination, semantic interoperability, evidence traceability and execution governance.
Website: www.proto-axis.org
ORCID: 0009-0009-5259-6102
Author website: www.romanshaban.org
PROTO-AXIS is the name of this independent project. Its descriptions and architectural proposals should be distinguished from established standards and independently validated implementations.
Designed for systems whose actions can be understood and checked.
© 2026 Roman Shaban. PROTO-AXIS.