Research / prototype; some modules have internal dashboards.
Cross-brand research case
AI multi-agent autonomous governance
Research into how multiple AI agents divide work, propose actions, review one another, and execute only with human approval. The focus is governance and traceability, not a claim of an autonomous company or unattended operations.
Ownership and role disclosure
The core R&D lead directed research and prototypes. This is a cross-brand research case; CPIN does not claim operation or ownership.
Public technical caseResearch / prototype; some modules have internal dashboards.Personal-led · ownership pending
Verified status
Separate development, deployment, commercial, and legal status.
This page does not equate a test pass, website launch, or research prototype with full commercial operation or CPIN ownership.
Production deployment not verified within this website scope.
No public commercial operation or revenue data.
Pending entity, ownership, financial, and automated-decision governance review.
Screenshot permission
No public screenshots; interface mockups are not used as substitutes.
Positioning and market
What engineering problem does this case address?
Multi-agent governance research centred on roles, permissions, approvals, and audit, exploring controlled AI collaboration inside organisational workflows.
Enterprises and research partners needing AI workflow governance, role permissions, and traceable automation.
Public-safe evidence
- EV-AI-002Document inventory of multi-agent roles, governance, and internal dashboards. Raw paths, model output, and sensitive data are withheld.
Evidence IDs map to the internal evidence index. Raw paths, customer data, keys, and full reports are withheld.
Completed capabilities
Not generic slogans, but explainable modules and engineering stages.
Capabilities describe research and engineering scope. Unapproved entities, customers, revenue, or performance are excluded.
Role separation
Separate agents for research, proposal, execution, and review so one agent does not both propose and approve.
Human approval gates
High-risk actions require human approval; the research does not claim unattended autonomous operation.
Task board and audit records
Traceable records for task state, agent output, decisions, and exceptions.
Tool and permission boundaries
Explicit tool interfaces and permission boundaries constrain what agents can execute.
3D office prototype
Hermes3D is a visualisation and interaction prototype; art and public scope are not approved.
Governance and exceptions
Research on agent conflicts, retries, escalation, and human takeover.
Architecture
A conceptual architecture of the system layers.
The diagram omits deployment details, vendors, keys, and sensitive interfaces. It is not a production screenshot.
Input
TaskKnowledgeTools
Agent layer
Research agentExecution agentReview agent
Governance
RolesApproval gatesAudit log
Interface
Task boardDashboard3D prototype
Cross-cutting: permissions · records · testing · data minimisation · human approval
Known limits
State the limits before the next step.
- Not a registered company and not claimed to operate autonomously.
- No public evidence of production deployment, customers, or revenue.
- 3D office art, agent output, and internal dashboards are not approved for publication.
Next stage
Proceed within the approved scope.
Work with us
Want to discuss this research case?
Tell us the purpose, public scope, and topics needed. Where other brands or sensitive data are involved, we confirm permission and review procedures first.