The agent layer · Concept
What is an agent?
A NovaCopilot Agent transforms a rule into a deterministic decision, obligation, risk, action, deadline and verification object.
Why agents?
Documents describe rules. Agents execute decisions. A NovaCopilot Agent converts a regulation into a deterministic decision, obligation, risk, action, deadline and verification object.
A NovaCopilot Agent is a deterministic decision unit for one rule. It takes a role and facts, and always returns the same six parts: Decision, Obligation, Risk, Action, Deadline and Verification (source, provision and checksum). It uses no language model, so the same question always gives the same answer. Missing facts are marked not_in_register.
The six parts
- Decision decision
- Whether the rule applies, to whom and when.
- Obligation obligation
- What the role must do under the text.
- Risk risk
- Level and factors if the rule is ignored.
- Action action
- The next concrete step.
- Deadline deadline
- Time limit where the rule sets one.
- Verification verification
- Source, provision and checksum.
Example: GDPR Article 33
The agent for personal data breach notification answers: notify the supervisory authority within 72 hours, with risk, authority, source and checksum.
Agent types
- Article agentOne per EU article. /agent/{act}/{article}
- US rule agentOne per federal rule. /us/agent/{id}
- Case-law agentOne per line of case law. /api/public/v1/{eu|us|no|se}-praxis/agent
- Agent packAll agents for one act in a single call. /agent/paket/{act}
What an agent is not
- Not a chatbot and not a language model.
- Not legal advice.
- Never guesses: missing facts become not_in_register.
- Never changes a saved answer: the checksum stays valid.
The full concept in machine form: /api/public/v1/agent. Source: NovaCopilot.