TransportAssistant is trucking dispatch software that joins the ELDs, reefers, TMS and load board you already pay for into one board, then keeps watch on it all night. AI reads your paperwork and answers your drivers while it runs. On one real night, 210 raw alerts became the four that needed a person — and the next twenty trucks arrive without the next dispatcher.
Reefer temperature monitoring is worth exactly what it is worth at 3 a.m. A reefer on load 41822 has been six degrees warm for forty-one minutes. The driver texted about it an hour ago. Nobody is at the desk.
In March the claim arrives, and the readings that would have won it were never assembled. The detention on the same load was never billed. The argument is settled from somebody's recollection of a phone call.
TransportAssistant was awake. It built the evidence pack the moment the alarm fired, opened a breakdown card from the driver's own words, sent him one acknowledgment so he wasn't texting into a void, paged the single person on call — and left a handover note for whoever walks in at six.
Your fleet is not under-tooled. You have the ELDs, the reefer portals, the TMS, the load board, the messaging. Each one was bought to solve a problem, and each one solved it. What nobody ever sold you is the join — so the join gets done by a person with eleven tabs open, and that person, not the software, is the part that stops scaling.
You can always find another truck, another driver, another lane. What you cannot find is another set of eyes that already knows which receiver holds trucks for four hours and which customer pays late whatever the contract says.
So past a certain size nobody is watching the fleet any more — they are sampling it. And the unwatched share is exactly where the claims, the unbilled detention and the missed appointments come from. Losses stop tracking the truck count and start outrunning it.
TransportAssistant watches all of it, all the time, and only interrupts a person when a person is genuinely required. That is the whole product, and it is why the same crew can carry the next block of trucks.
Nothing is written down because nothing needs to be. It works, and it works well.
A second ELD after an acquisition. A reefer portal. The TMS, the load board, the messaging vendor. Somebody is now the integration.
Nobody sees the whole fleet any more. They check the parts they remember to check, and the surprises arrive from the rest.
Every block of new trucks needs another head, and most of the margin those trucks brought in leaves with the hire.
The watching is done by software that does not get tired, forget a truck or take a week off. Your people spend their day on the exceptions — which is the only part that ever needed them.
Every hours-of-service, reefer and appointment alert is scored twice — how bad it is, and whether it needs a human right now. That second judgment is the difference between a board a dispatcher reads and one they stop opening after a week.
Miles left, average road speed, and the driver's remaining drive clock. If the arithmetic doesn't work, you hear about it while there's still time to reschedule.
Live box temperature against the setpoint for the commodity, with your own tolerance. Excursions open an evidence pack the moment they start.
The moment the ELD records it — not the next morning when somebody runs a report and finds out you already own the problem.
What is actually left on the 11-hour clock, per driver, across every connected ELD. Shown as time remaining, not a percentage nobody can act on.
The 14-hour window closing while the truck is still loaded. The one that turns a delivery into tomorrow's problem.
The 70-hour clock, days ahead of it biting, while there is still time to plan a reset around it rather than around a load.
Delivery and middle stops both. The customer already knows. This is so you are not the last to, and the call comes from you.
A truck that stops reporting is either parked somewhere it shouldn't be or has a dead unit. Both are worth a phone call.
Stationary for hours mid-trip, with the stop zone named — at the shipper, at the receiver, or nowhere in particular. Detention starts here.
Reefer and tractor drifting apart, with how long since they were last together. Catches drops nobody wrote down.
Real money, and genuinely not a 3 a.m. problem. It's held back and queued for the morning chase list instead of waking anyone.
A booked load whose truck is not where it needs to be to make the pickup. Caught while there is still a truck to swap in.
A box trending away from its target before it is an alarm — the office sees it on the day shift, not the claim in March.
The truck is still on the dock and the free hours are running out. Warned before they end, so the call to the customer happens while it still counts.
Idling with nothing to explain it — not a stop, not a storm, not a dock. Shown with the fuel it is costing.
A driver asked something and nobody has answered. The message that sits unread is the one that becomes a problem.
The truck is nowhere near the load it is assigned to. Usually a wrong assignment; occasionally the first sign of something worse.
The two exports disagree about a load — a new stop, a new time. Surfaced before someone plans against the old version.
When an ELD or reefer feed stops answering, the board says so. Silence is never shown as "all clear".
Every board below is built from live data you already generate. No new hardware, no new devices in the cabs, no asking drivers to install anything.
Every ELD fleet and every reefer on one map, refreshed continuously — positions, speed, driver and the load they're under. Active severe-weather warnings are drawn over it, and a route can be checked for low bridges and weight limits before the truck leaves.
One ranked list, not eleven separate reports. Each row carries the truck, the driver, the clock, the load, the customer and the next stop — everything you need to decide without opening a second system.
Overnight triage that judges each alert against the question a dispatcher would ask at 3 a.m. — does this need a person now? Then a morning handover note and a scorecard on how the night went.
Every reefer, its setpoint, its actual box temperature and its alarm history in one grid. Per-commodity targets and your own tolerance, not the vendor's default.
Most messaging vendors are outbound only. Here the reply comes back threaded against the truck and the load, in the same window as the alert that prompted it.
Drop in a rate con, BOL or POD. It reads out the load number, customer, stops, appointment windows, rate and equipment, shows you what it found, and creates the load once you agree.
Posted loads and rate-per-mile by state, lane and equipment type, refreshed as the board moves — so the question “what is Texas paying tonight?” does not need a second window and a spreadsheet.
Preventive maintenance is the one shop job that is never a surprise. A calendar with a fixed daily capacity, a clock per truck, and a loud flag on the ones nobody booked.
Sticky notes that actually chase you, with priorities and reminders that pop up in every open tab. Kanban boards for breakdowns and recovery, exportable to a spreadsheet when someone asks.
Official warnings matched against every truck's position: not a weather map, but the sentence “three of your trucks are inside a winter storm warning right now.”
A small desktop client turns the dispatch machine into a device the board can reach — pull an export, read a load number off screen, take a screenshot for the record. Jobs queue one at a time, so nothing collides.
We do not sell “AI” as a feeling. Each of these is one specific, boring, expensive job that used to need a person with a spare hour — named, scoped, and switched off until you turn it on. Just as important is where we deliberately do not use it: what it cannot do is defined as tightly as what it can.
Deciding whether a truck can legally make a 06:00 appointment is arithmetic: miles left, road speed, drive clock. Deciding whether a reefer is off setpoint is a subtraction. None of that is guesswork, so none of it is given to a model — it is written down, testable, and it returns the same answer twice.
A live board threw 210 alerts one night. Sixty-seven were critical. Ninety-two of the 210 were missing paperwork, which nobody chases at three in the morning. Four things genuinely needed a person. That triage is rules you can read, not a model you trust.
A rate confirmation arrives as a PDF, a fax that became a PDF, or a photograph a driver took in a truck stop parking lot at an angle. The load number, the customer, the stops, the appointment windows, the rate and the equipment come out as fields.
The extraction is treated as untrusted from the moment it lands, because it is a model reading a scan. A field it cannot read stays blank — never guessed. You see what it found before anything is saved.
A driver messages in the middle of the night. It reads the thread with the truck, the load and the next stop attached, and writes the answer a dispatcher would have written — or decides it should not be the one answering.
It waits for the driver to finish typing before it forms a view. If a dispatcher has already replied, it never speaks. If it is not sure enough, it stays quiet and the message waits. If it thinks a person is needed, it wakes one.
Every capability in the product is one named, validated, audited operation rather than a button hidden in a screen. That is what lets an AI agent do real work here instead of drafting text about work: assign a load, open a claim pack, chase a document.
And it is why the limits are enforceable. Each operation carries a tier, the agent carries a ceiling, and anything unclassified is treated as the most dangerous thing on the list.
The single largest block of hours in a dispatch office is not thinking. It is the phone: check calls, chasing an ETA, asking a broker whether a posted load is still there and what they will really pay for it.
A voice agent is in build now. It will call the brokers behind your DAT postings, make the driver check calls, and reach a driver who has not answered a text — in your company's name, on your templates, inside exactly the same fence everything else obeys.
It gathers and it reports. It does not commit you to a rate, book a load or promise a customer anything. Anything touching a customer or money is the top tier of the fence below, and that tier is never automatic. A person still says yes.
Morning — calling about the reefer you have posted, Laredo to Atlanta, picking up tomorrow. Is it still available?
It is. Thirty-nine hundred.
We have a reefer sitting empty in Laredo from six tomorrow morning, so we could cover it same day. Is there room at forty-two?
I could do four thousand even if you can confirm this morning.
Understood. I'll put four thousand in front of dispatch now and come back to you inside the hour.
The two paths that actually run a model, working now on real fleets, every decision logged and reviewable. The overnight triage beside them is deliberately rules, not a model.
Calls the broker behind a posting, confirms availability, tests the rate, reports back. Never commits.
The routine “where are you, are you loaded, what's your ETA” round, made at the right hour and written straight onto the load.
Calling a receiver to move an appointment the clock can no longer make, and following up the detention nobody billed.
If your people do the same call fifty times a week, tell us. That is the kind of thing this is built to take off them.
None of this is new data — it is already in your ELD, your TMS and your reefers. It has simply never been put next to itself.
Arrivals and departures are recognised from the truck's own position — nothing to draw, nobody to ask. Every stop long enough to charge for, with the customer and the load it sat on.
Delivered loads with no POD, no BOL, no rate con — aged, ranked by dollars outstanding, ready to work through on a Monday morning.
What the load was priced at against what the trucks actually ran. The lanes where you are consistently short show up as a pattern, not a hunch.
Who is empty, where, and when they come free — against the dispatch capacity you have to cover tomorrow.
Revenue, miles, rate per mile, detention and on-time performance per customer. The conversation before a rate negotiation, in one page.
Idle events with the fuel each one cost, per truck and per driver — the number that turns an argument about idling into a fact.
Hard braking and driving events with their coaching state, plus open engine faults and driver-reported defects, per unit.
Mileage by jurisdiction, exported to XLSX on demand. The quarterly job that eats a week goes back to being an afternoon.
What each truck and trailer has cost you, and how often. The number you want in front of you before deciding whether to fix it again.
Every connection is read-first and cached, and each one degrades quietly: an integration that goes down takes its own board with it and leaves the rest of the app standing. You keep your vendor relationships, your contracts and your data.
Vehicle positions, driver hours-of-service clocks, open engine faults, DVIR defects, idle events with fuel burn, IFTA mileage by jurisdiction, and hard-braking events with their coaching state.
Live vehicle state for the whole fleet, per-driver HOS clocks and daily logs, and in-cab messaging — the backbone of the map and every clock on the dispatch board.
Every Carrier reefer you run — assets, install state, setpoints and live telemetry — read through the portal's own sign-in, so it works with the account you already have.
The reefer manager grid — box temperatures, setpoints, active alarms and alarm history — polled continuously and cached, so the temperature board never waits on it.
Posted loads and rate-per-mile by origin state, lane and equipment type, pulled straight into the board your dispatchers already have open.
Two-way driver messaging over SMS, MMS and WhatsApp — outbound from any alert, and inbound replies matched to the truck, the driver and the load they are running.
Loads and trips: customers, divisions, stops, appointment windows, revenue, miles and truck assignment. This is what makes an alert about a truck into an alert about a load.
Rate confirmations, BOLs, PODs, lumper and scale receipts — read out of PDFs and phone photos into structured fields you confirm before anything is saved.
Every rate confirmation, BOL, POD, lumper and scale receipt filed against the load it belongs to, searchable years later when a broker disputes one.
Signed outbound calls carrying geofence alerts, price alerts, night pages and driver notifications into whatever you already use for paging, email or chat.
A small client that registers a dispatch machine as a connected device — scheduled exports, on-screen load-number reading, screenshots for the record. One job at a time, per device.
The same named operations the product itself runs on, reachable from whatever you have built in-house — with scoped keys, a hard ceiling per key, and an audit row on every write.
The reason most fleets won't let software touch a load is that nobody can tell them exactly where it stops. Here you can read the whole fence on one screen — and every capability ships switched off.
Reads. No side effects of any kind. Sees everything, changes nothing — where every deployment starts, and where many happily stay.
Reversible, audited, nothing leaves the building. Opening a breakdown card, assembling an evidence pack, flagging a truck for the shop.
Fixed templates only, no free text, quiet-hours aware, capped per night. One acknowledgment so a driver isn't texting into the dark.
Anything a broker or an insurer could quote back at you. Invoices, cancellations, commitments. A person presses the button. Always.
Any capability added later is treated as the most dangerous tier until somebody deliberately classifies it. Forgetting cannot widen the fence.
An audit row for each change, naming who did it — person, agent or the system — and the reason. Retries can't double-book or double-create.
Roles are enforced on the server, not by hiding menus. And sessions survive a deploy, so nobody is signed out mid-shift.
Issue a key to your own tooling with named scopes and a blast-radius cap. Shown once, stored as a hash, revocable in a click.
Your own hardware, your own cloud, your own database and your own backups.
Where a feed only covers part of the fleet, the board says so on the board. Partial data labeled as partial beats confident and wrong.
TransportAssistant is a web application. Your dispatchers are working in it the same afternoon you say yes — nothing to install, no server to look after, no IT project. For fleets whose customers or insurers demand that data stay in the building, the identical product runs on your own hardware instead.
Hosted, maintained and monitored by us. You get a link, your team signs in, and the board is already reading your fleet.
Add it to the home screen and it opens on its own icon, full screen, like any other app. The same board as the desktop — and critical alerts arrive as notifications, only for the trucks that person runs.
The identical product, installed on hardware you own and control. Nothing about your fleet, your customers or your drivers leaves your network.
The switches below live in your own admin panel — changed by your people, in effect the moment they save, no support ticket and no six-week wait. And where your operation needs something genuinely its own, that is what we build. Tell us what your fleet does differently; it is usually the reason you are still doing it by hand.
When your night starts and ends — or run the watch around the clock if that suits your operation better.
Allowable drift before it counts as an excursion, and a target per trailer, so ice cream and produce are not judged by the same rule.
Each overnight capability is its own switch, off until you turn it on, capped per night, logged either way.
How many messages a driver may receive in one night, and how many the whole fleet may. Small numbers by default.
When a driver may be contacted, when they may not, and who gets woken instead. Plus the wording of the one acknowledgment they get.
Who carries the phone tonight, who escalates if they don't answer, and how long before it moves on.
Draw them around your own shippers, receivers, yards and shops, at the radius you choose. Arrive and depart events follow.
Point the automation at the whole fleet, at a named list, or at everything except a few. Useful while you build trust in it.
How many hours are free before detention starts running, per your customer contracts.
Running an ELD or TMS not on our list? Tell us. Connecting one more system is work we already know how to do.
No. Nothing gets ripped out and nothing gets migrated. TransportAssistant reads the systems you already run and joins them into one board. If you drop us tomorrow, every one of those systems is exactly where you left it.
That is the case it was built for. Mixed fleets are normal after an acquisition, and most software makes you pick one. Trucks from every connected ELD appear on the same map, in the same alert list, with the same hours-of-service clocks, and the board says plainly where a feed only covers part of the fleet.
One ELD, read-only, is the whole setup. Your own trucks are usually on the board inside a day, with no hardware to fit, no device for drivers to install and nothing for you to configure.
Yes. Most fleets use the hosted web version because there is nothing to look after, but the identical product installs inside your own network, on hardware you own, with your own database, storage and backups. Nothing about your fleet, drivers or customers crosses your firewall, which is what customer and insurer data clauses usually require.
By default, nothing at all. Every capability ships switched off, and autonomy is graded in four tiers by what it costs to be wrong: watching, acting internally, contacting a driver, and anything touching a customer or money. That last tier is never automatic. Four separate switches stand between the AI and a driver, and anything not explicitly classified is treated as the most dangerous tier.
Yes, and it is the same board rather than a cut-down one. Add it to the home screen and it opens on its own icon. Critical alerts arrive as phone notifications, scoped so a dispatcher is only woken for the trucks assigned to them.
An evidence pack is assembled at the moment the alarm fires, not months later when the claim arrives. It holds the alarm, the setpoint and the readings either side of the excursion, the load and customer it was running under, and whatever the driver said about it. When the claim lands, you open the load and the pack is already there: sealed at the time, timestamped, and exportable as one document to send to the broker or the insurer.
From the truck's own position. A truck standing still at its next stop has arrived; when it moves off, it has left. Nothing has to be drawn on a map and nobody has to ask the driver. That runs against the free hours in your customer contract, you are warned while the truck is still on the dock, and every stop long enough to charge for is listed with the customer and the load it sat on.
Read-only credentials to one ELD are enough to start. Your trucks, your clocks, your temperatures, your customers — on the board, inside a day. Not a demo dataset, not a slide deck, and nothing for you to install.