AI Agents
Hire a Fuel Analyst, a Driver Coach or Feed Watch, scope it to a fleet, a tag or a single driver, and let it watch every transaction. Every suggestion arrives with a confidence score, the evidence it read and a “Before you act” list — nothing acts on its own until you move it up the autonomy ladder.
The hire model
You hire a role, scope it (whole fleet, a fleet, a tag, a driver…), and can hire more than one of a role with different settings. Every hire starts in Shadow.
A Fuel Analyst, a Driver Coach or Feed Watch, each with its own task library, evidence and allow-listed actions. An Onboarding role handles data readiness.
Whole fleet, a fleet, a tag, a driver… and hire more than one of a role with different settings, say one coach per fleet.
Every hire starts in Shadow: it runs and decides, tells no one, and you see how often you would have agreed before it proposes anything.
Three roles today
Each role is a job description with its own task library, the evidence it is allowed to read and the actions it is allowed to propose.
Keeps every transaction honest
Keeps your fuel record straight: finds dead gauges and quiet trucks, puts transactions on the right vehicle, fixes the mappings behind repeat mistakes, reconciles gallons bought against gallons burned, and tells the fuel buyer where the network has holes.
12 tasks in the libraryDrafts the right note
Coaches drivers on how they drive, buy and follow their plans — one specific, askable note at a time, drafted for you to send.
7 tasks in the libraryTells you when a feed goes quiet
Tells you when a feed stops delivering, before the reports quietly go stale.
1 task in the libraryAn Onboarding role handles data readiness, one per account, so the other hires only run on data that is ready.
The task library
Each task listens for a signal, reads evidence, decides with a prompt you can edit, and may propose only allow-listed actions.
Your own tasks
Fleet managers can also create their own tasks from any list page's filters (the filters become the signal) and edit any task's name, description, evidence, prompt, confidence bar and limits.
The autonomy ladder
Autonomy is set per task, not per agent, on a five-rung ladder. Move a task right as it earns it: Off → Shadow → Suggest → Approve → Auto.
Never runs.
Runs and decides, tells no one; you see how often you would have agreed.
New hires start hereProposes in your feed with the evidence; you click the action.
Proposes, then applies when you approve; one click in the digest clears a batch.
Ceiling for anything without an undoApplies at once above the confidence bar; shows in your digest with an undo.
Only for actions with an undo
Evidence first
Every suggestion shows a confidence %, the evidence it read (chips like “Driver behavior”, “Plan compliance”, “Coaching history”), a “Before you act” list of caveats, and, for coaching, the exact note the driver would read.
Actions are typed product commands, never raw edits
Each one is the same command you would run yourself in the product, so it can be reviewed, approved in a batch, or undone.
169-gal, $606.23 Texarkana buy sits on a truck with no tank movement; only vehicle 222 was there and its tank rose 185 gal
Alfredo Trevino hit 4 of 9 planned fuel stops but ran 111 miles under plan — draft asks why, not a correction
Alfredo — a question, not a ding. On 2 Sep you fuelled at Saundersville, MS and Loudon, TN instead of the planned stops — 122 gallons at about 19¢/gal over the plan price that day. Across the last four weeks you hit 4 of 9 planned stops. What made the planned ones not work: hours, a dock hold, a site down, or the lane changing? Tell me which stops keep missing and I'll take it to planning.
Readiness and budget
A hire that cannot see the data stays quiet, and every one of them reports what it read, what it proposed and what it cost.
Readiness gates: a task only runs when its data is ready (e.g. fuel cards mapped to vehicles, weeks of purchase history). Until every gate is open, the task waits, and the hire page tells you which gate is closed.
Each hire shows signals seen / proposed / accepted / agreement % and an AI cost against a monthly budget (e.g. “$0.74 of $75.00”), so you can see what a hire reads, how often you agree with it, and what it costs to run.
One hire · last 30 days
Agreement % sits beside these: how often you accept what the hire proposes.
Where you'll meet them
Wrong truck on the ticket, gallons that never made it into a truck, a card mapped to the wrong truck: proposed as reassignments and mapping fixes, with the tank and GPS evidence attached.
Fuel Theft MonitoringDriver CoachRecognition when a driver improves, a question when planned fuel stops keep getting missed, delivered on the ELD, Telegram or SMS once you send it.
Driver PerformanceFeed WatchTelematics, fuel cards, pricing networks and TMS loads: when one stops delivering, you hear about it before the reports quietly go stale.
Integrations & APINext step
Walk through your fleet data and identify where cost, loss, and efficiency can improve. Use our Calendly scheduler to pick a time that works for you.