Agent-Assisted Authoring

Describe a KPI in plain language and review a validated draft proposed by the authoring agent (early access).

Instead of building the calculation spec by hand, you can describe the KPI you want in plain language and let the Fleetera authoring agent draft it for you. The agent reads your population's live variables, consults the operator library, and proposes a validated definition — which you then review, save, preview, and enable exactly like a hand-authored one.

Agent-assisted authoring is an early access capability. The agent proposes; a human approves. It has no ability to save, enable, or change anything in your organization — every draft goes through the same review, preview, and enable gate as manual authoring, and every agent action is journaled.

Drafting with the agent

The Agent panel sits on the left of the New KPI page.

Pick the population first (asset categories, optionally a site). The agent drafts against the population you selected — it inspects those assets' variables, so it cannot start without one.
Describe the KPI in the message box — for example, "Average state of charge per hour for these modules" — and click Start with the agent.
The agent works server-side for the whole turn: it inspects your population, checks existing definitions, drafts a spec, and validates it. Turns take on the order of tens of seconds up to a minute — the panel shows a working indicator while the agent consults live data, and breadcrumb badges for each lookup it performs.
When the agent presents a proposal, it is applied straight into the builder on the right: name, KPI key, unit, cadence, the spec in the JSON view, and the agent's plain-language rationale. Only proposals that passed the platform's validation ever reach you.
Review and adjust anything you like, then continue with the normal flow: Save draft (disabled), preview real values, and Enable.

To iterate, keep chatting — ask for changes and the agent refines the proposal against the same population. Start over discards the session and begins fresh. Sessions are stored server-side, so a page reload resumes the conversation where you left it.

What the agent can and cannot do

The agent operates against a deliberately read-only set of platform lookups:

Lookup	What it provides
Population context	The matched asset count and the telemetry variables available on those assets (keys, units, coverage).
Operator reference	The calculation grammar — operators, their parameters and ports, and the semantic rules the engine enforces.
Existing definitions	Your organization's current KPIs, so the agent avoids duplicates and can reuse them as KPI-on-KPI building blocks.
Validation	Server-side validation of a draft against your population — spec shape, unit codes, KPI key uniqueness, input-variable coverage — without saving anything.

The agent cannot save, enable, disable, or delete definitions, and it cannot write data of any kind. There is structurally no path from the agent to persistence: its only artifact is a proposal, and turning a proposal into a definition is always your action in the builder.

Accountability

Saving an agent proposal records the revision with author type agent and you as the approving user. The Revisions table on the definition detail page marks such revisions with an agent badge.
Every agent session is journaled — each of your messages, every lookup the agent performed, and every proposal version, in order. For definitions authored this way, the full Agent journal renders on the definition detail page, so anyone reviewing the KPI later can see exactly how it came to be.
Agent behavior is exercised continuously against an automated evaluation suite that runs nightly, and failures feed new evaluation cases.

Agent-Assisted Authoring

Drafting with the agent

What the agent can and cannot do

Accountability

On this page