Loading…
Loading…
Governed AI agents
Every other AI agent lets the model decide what happens next — impossible to audit, certify, or bound the cost of. A governed agent inverts that: the control flow is a BPMN process, the decision is an editable DMN policy table your team owns, and the AI is confined to reading. Describe an invoice below — no signup.
The LLM extracts structured fields (amount, PO number, vendor) from unstructured text. It perceives. It decides nothing.
A deterministic rule engine evaluates your policy table — no randomness, no LLM inference. Business analyst edits the JSON; no code change.
The BPMN process sends the decision to the approver, writes an audit record, and closes the loop. Every run is fully traceable.
AI reads your description · DMN policy decides · BPMN routes · audit trail written
Upload a PRD that mentions invoice approval, claims processing, loan adjudication, or any spend-management workflow. Archiet detects the agent capability, generates the BPMN process XML, DMN policy table, Python runtime, and operator console — and drops them into your ZIP alongside the rest of your application. No drag-and-drop builder. No prompt engineering. No redeploy to change policy.
Every decision traces to a named policy rule, not an AI guess. The AI only extracts data; it never routes.
Change the policy by editing a data file — no prompt surgery, no redeploy. Your team owns the rules.
Generate your own app with a governed agent and it ships with this exact console, the policy, and the BPMN/DMN docs.
A governed AI agent is an AI agent where the process is defined in BPMN (Business Process Model and Notation), the decision logic is encoded in a DMN (Decision Model and Notation) policy table that business users can edit, and the AI model is confined to data extraction only — it never decides the outcome. This makes the agent fully auditable, certifiable under EU AI Act Article 9 (risk management), and approvable by enterprise compliance teams.
A regular AI agent lets the language model decide what action to take next — making the outcome non-deterministic and impossible to audit or bound the cost of. A governed agent inverts this: the control flow is a BPMN process (deterministic), the routing decision is a DMN policy table your team owns (auditable), and the AI only reads structured data from documents — it never chooses the route. Every decision traces to a named policy rule, not an AI guess.
BPMN (Business Process Model and Notation) is an OMG standard for defining the sequence of steps in a business process. When applied to AI agents, BPMN replaces the agent's autonomous decision-making loop with an explicit, auditable process graph. Each step is deterministic: extract data (AI task), evaluate policy (DMN decision table), route to the appropriate action (BPMN gateway). The result is an agent whose full behaviour can be diagrammed, reviewed, and certified.
DMN (Decision Model and Notation) is an OMG standard for encoding business rules as a table of conditions and outcomes. In a governed AI agent, the DMN table replaces the prompt instruction that would otherwise tell the AI how to decide. For example, an invoice-approval DMN table specifies: if amount < $500 and PO matched, auto-approve; if amount > $10,000, escalate to CFO. Business analysts can edit the table without touching code or redeploying the agent.
Yes. Every Archiet-generated application with agent capabilities ships a governed agent pack: the BPMN 2.0 process XML (opens in bpmn.io or Camunda), the DMN decision table in JSON format, the Python/FastAPI agent runtime, and an operator console built in React/Next.js. The generated app includes the same live invoice-approval console you can try on this page, wired to your own data model and editable policy tables.
Governed agents are designed to satisfy the EU AI Act's Article 9 (risk management system), Article 13 (transparency), and Article 14 (human oversight) requirements for high-risk AI systems. Because every decision is traceable to a named DMN rule — not an AI model inference — the agent generates a complete audit trail. The Archiet AI Governance pack that ships with every governed agent includes EU AI Act risk classification, model cards, and a data-lineage diagram required under Annex IV.
Related: Archiet MCP server · EU AI Act compliance documentation · Free architecture audit