Technical overview · evidence as of 2026-10-09
This page describes the documented organizational architecture for the AI Clouds Company OS candidate. It is a conceptual model of roles, controls and capabilities—not proof that every role, connector or workflow is currently running.
Measured / documented
The five operating layers, one primary reporting line per role, bounded task contracts, human approval boundaries, and tool or connector ownership are described in the local CompanyOS organization and registry files.
Unknown / not independently verified
A role contract is not a permanently running agent. Capability declarations are not live connector evidence, and this candidate does not claim production readiness, full autonomy, commercial deployment or direct Claude product usage.
Documented inventory
The source outline records 23 role contracts, 18 policy areas and 16 operating-workflow stages. These are documented design inventories for review; they do not prove that 23 independent agents, all policies or every workflow stage are currently active, connected or production-ready.
Semantic hierarchy
The hierarchy below is intentionally plain HTML. It communicates accountability and boundaries without implying a live systems diagram or autonomous execution.
Direction and approval
Owns direction, budgets, policy, price exceptions, customer commitments and production-domain decisions. Publication and other protected external decisions remain explicit human authority.
Coordination and gates
Inspects company state, chooses the next objective, assigns one accountable department, enforces gates, resolves blocked workflows and reports exceptions to the CEO.
Accountable outcomes
Own department outcomes, decompose objectives, select permitted tools, validate specialist work and return one structured report to Hermes. The role registry defines one primary reporting line for each manager.
Bounded execution
Perform bounded tasks and return evidence or artifacts. They cannot alter policy or approve their own critical output. A documented specialist may be a temporary worker, reusable skill or deterministic procedure rather than a permanent agent.
Capabilities and boundaries
APIs, browser automation, database, storage, analytics, scheduling and model providers supply capabilities. They do not manage departments, decide company policy or create authority on their own. Live availability and external results remain to be verified at runtime.
How accountability moves
The intended flow is inspect → assign → execute within a bounded contract → validate → report. Independent QA or security review may block critical output, while protected actions return to the appropriate human authority.
Read the named, registry-backed role directory or return to the customer-facing overview.