AI Agents — Overview

AI agents are managed in roleALPHA like employees: they fulfill roles, support value streams, use competences — and are bound by policies (guardrails). This produces a machine-readable context bundle that real AI agents pull over MCP.

Why this is a business advantage

Enterprises deploy more and more autonomous AI agents — and quickly lose the overview: Who built which agent for what? What may it access? Which rules apply to it? Who is accountable?

roleALPHA answers these questions by placing agents into the same organizational model your people already live in. An agent is no longer a loose script but a governed actor with:

• Grounded accountability — the agent derives its responsibility from the role it fulfills (including that role’s accountabilities, visibilities and policies). Authority is derived, not freely granted.
• Traceability — every context retrieval is logged. You can always prove which guardrails were delivered to an agent at retrieval time (relevant to, among others, the audit and oversight obligations of the EU AI Act).
• Guardrails in business language — binding policies are versioned, approvable and human-authored (via the Policies module), not buried in model prompts.
• Control at scale — many agents can be overviewed, grouped and governed centrally in the org graph, with the same RBAC and visibility mechanisms as the rest of the organization.

What roleALPHA AI Agents is

The governance and org-context layer for the hybrid workforce — people and AI agents in one model. Concretely:

• A system of record for your agents: purpose, persona/instructions, status and the links to roles/value streams/competences/policies.
• A context provider: over MCP (get_agent_context) and REST, roleALPHA delivers a self-describing, versioned context bundle — exactly what a real external agent needs to operate.
• A guardrail declarator and auditor: roleALPHA states which policies bind an agent and logs their delivery.

What it is not

Equally important — the deliberate boundary:

• Not an agent runtime / execution engine. roleALPHA does not run the agents itself. Your agent platforms (e.g. cloud agent services) do. roleALPHA feeds them context.
• Not a model safety filter and not an observability/tracing tool.
• Not a replacement for rAlph, the built-in assistant. rAlph uses the model; an AI agent here is a governed definition with context.
• At this stage roleALPHA does not yet enforce the guardrails. It declares and audits them; enforcement is the consuming agent’s responsibility. A policy marked as a guardrail must not be downgraded to mere context by the consumer — its only latitude is to comply or not (and non-compliance is provable via the audit trail).

Getting started

The step-by-step guide (create an agent, link it, view context, retrieve over MCP) is under AI Agents.

Related

• AI Agents — The “AI Agent” entity type in detail
• Roles — Roles an agent fulfills
• Policies — Policies as guardrails
• Relationship Types — The relationship types AGENT_FULFILLS_ROLE, AGENT_BOUND_BY_POLICY, etc.