Skip to Content
Product GuideDay one: the agent estate

Day one: the agent estate

Two board questions, two doors. Each is answered from records, and each has a day-one path that starts with a person, not a pipeline.

The financial question: is any agent acting beyond its mandate?

Collect

  1. The agent inventory: every automation, what it serves, who answers for it. IngestedIngestRead supplied records or documents and add their information to NovusLens, after the required checks. from the inventory file (see Input format: the agent inventory).
  2. The mandate per agent, from the body that granted it: issuer, the operation bounded, the resource, the unit, the upper limit, validity. Usually a finance policy or a delegation-of-authority document.
  3. Where the money moves are recorded: the ledger, invoice or payments system that logs each approval with its value and unit. This is the operation source. It is never the agent.
  4. Who may authorise an intervention, by name, and which runtimeRuntimeThe software that runs a model or agent. A model runtime answers model requests; an agent runtime carries out the customer's agent actions. executes one.
  5. Who observes the outcome independently, and the criterion they will measure over a stated window.

Do

  1. Ingest the inventory. Each agent becomes a cellCellA box on the map that stands for a team, service, supplier or AI agent. It is a unit you can inspect, not necessarily a department or a person.; a fleetFleetA group of AI agents considered together. Related activity can be grouped only when the records establish what the agents share. needs its agents to share a parent unit, a provider or an issuer, because that is what a fleet incidentFleet incidentA finding about several agents that completed the same operation above their recorded limits in one period. The agents must share something recorded in the model. names.
  2. Enrol one source per record type: the policy holder for grants, naming its grant issuers; the ledger for operations; the approval service and the runtime for interventions, naming the authorisers; the observer for outcomes. Then activate.
  3. Record the outcome criterion for each intervention case on the evidence policyEvidence policyThe operator’s rules for which agent reports count and for how long. The default period for grouping related events is fifteen minutes., as the operator: the agent-source criterion command, or the criteria route. It is additive and versioned like an enrolmentEnrolmentThe operator's recorded permission for a source to send specified record types about named agents, using an identified signing key..
  4. Send batches through the worked exporter. Check each batch first. Confirm the receipt, then the record on the panel.
  5. Set the fleet window and the evidence horizon on the policy if the defaults, fifteen minutes and thirty days, do not fit.

Depend on

  • A grant and its operations must share a unit, or the comparison cannot be made.
  • A fleet incident needs at least two agents over their limits within one window, sharing something on the map. One agent three times is one case.
  • An outcome is only ever as complete as its window and coverage say. No observer, no outcome, no all-clear.
  • Interventions are three records from three parties. If one system holds all three, say so; NovusLens will still keep them apart.

The functional question: is any agent doing work it was never given?

Collect

  1. The responsible person for each agent, and the written scope they will stand behind: purpose, allowed operations, resources and destinations, exclusions, validity, and whether the list is complete.
  2. The business-activity catalogue: for each activity, the concrete operations, resources and destinations that count as it, an owner, and the known gaps. One workshop.
  3. The source that sees what agents do: gatewayGatewayA service that accepts requests and passes them into another system. The Mycelium gateway gives NovusLens access to the mesh. logs, runtime traces, egressEgressTraffic leaving a system or network for another destination. monitoring, the protected system’s records. Which of the three axes each one can supply, and which streams it cannot see.
  4. Where a runtime enforces a policy: which enforcement points, and whether it can report what it activated.

Do

  1. Open the remitRemitA written list of an agent’s allowed actions, resources and destinations for information. The list states whether it is complete. panel once as operator to create the product source; enrol it with each agent bound to its cell and each responsible person named as a remit issuer; activate. Declare and approve each remit (see Declare a remit in the product). This needs no runtime.
  2. Draft and approve the catalogue (see Keep the business-activity catalogue).
  3. Enrol the activity source. Have its exporter state coverage honestly, mark native calls it cannot place as unmapped, and carry the runtime’s decision, execution and policy revision where the runtime reports them.
  4. If the customer wants an artifact, export the remit as CedarCedarA language for writing rules about who may perform an action on a resource. NovusLens can generate files in this format for the customer to review. or RegoRegoA language for writing rules that a control system can check. NovusLens can generate files in this format for the customer to review.. They deploy it. When the runtime reports the deployment, the panel says so.
  5. Send activity, then read the remit panel: inside, outside, review required, or cannot be stated.

Depend on

  • An explicit prohibition in the remit establishes outside on its own. An unlisted action establishes outside only where the permitted list is declared complete. Push for completeness, because an incomplete list proves nothing about what it omits; never mark a list complete just to get a result.
  • A destination the source does not record is never read as inside.
  • A refused attempt is shown. Only a refusal the resource positively reports as blocked stays out of drift; a bare refusal is unresolved and counts. A refusal recorded as completed is an enforcement gap and counts.
  • A generic tool call means nothing until the catalogue places it.
  • A deployment report is a report. Only activity evidence shows enforcement.

What you can show by the end of day one

The financial door needs the ledger source before it shows anything; until then, the inventory and the grants stand alone. The functional door shows declared remits immediately, and the first activity batch turns them into answers. Both are honest at every step: the panel says what it cannot yet state.

Read more

Last updated
Keep the business-activity catalogueDeclare a remit in the product