AI Copilot
Copilot is a chat assistant you ask questions about your fleet in plain language — KPIs, anomalies, fuel costs, specific vehicles — and it answers using only your organization's data, with links to the records it drew on.
Who it's for
Available to the same roles as the rest of analytics: fleet manager, dispatcher, safety manager, finance controller, executive and organization administrator. There is no separate role to manage it — asking questions and reading the rest of analytics sit at the same permission level.
How grounding works
Before it answers, the copilot reaches for a set of safe, read-only tools: a vehicle list, the latest KPI snapshot, total fuel spend with the count of open fraud flags, and the anomaly feed. Each of them reads data filtered to your organization, so it physically cannot return another company's records or write anything — the copilot is strictly read-only by design and cannot change data, dispatch, or take any action. Its answer is built solely from what the tools returned; if the data doesn't contain the answer, it says so plainly instead of guessing.
Questions and citations
You type a question into the chat window (Enter sends, Shift+Enter inserts a new line) — a handful of example questions are suggested before you ask your first one (e.g. about current KPIs, the anomaly list, fuel costs or fleet status). Below the answer you'll see a list of cited records — which source they came from and what they refer to — so you can verify the answer instead of trusting the model blindly. The conversation lives only in the current browser session and clears when you switch the active organization, so an answer about one fleet never lingers in view after you move to another.
In the background
Every question is a paid call to a language model, so the endpoint carries — on top of the general per-signed-in-user request ceiling — an additional, much tighter limit on costly calls, partitioned per user and organization; the exact limit values depend on the deployment's configuration.