AI for freight forwarding operations.

The category is often sold as "AI for logistics." That phrase does not help an operations manager. What they need is AI for freight forwarders that sits on live operational data — and that refuses to silently rebook freight.

Airline messages contain AWBs, flight numbers, dates and event types in inconsistent prose. UpliftOS extracts those fields — MAWB, HAWB, booking reference, flight, roll versus delay versus cancellation — and treats them as structured input, not as a paragraph to summarise.

Extraction without matching still leaves a search. The software associates the message with the shipment that owns that AWB or booking, then opens work against that file. Matching is the difference between a language model and air freight operations software.

Once the shipment is known, the event becomes a disruption with impact. That is how air freight disruption management stays connected to AI rather than becoming a separate module. Rolled cargo, delays, missed connections and missing documents all land in the Operations Inbox.

Recoveries create work: rebook, chase a DGD, confirm collection, notify the customer. UpliftOS can open those tasks in context so the AI output is operational, not a chat transcript.

For a cargo roll, the recommendation is a next flight with ETA impact, not a generic suggestion to "contact the airline." Flight-cancellation recoveries use a deterministic planner over live shipment and capacity data. The operations agent can explain a plan, but write actions that change bookings go through approval.

Drafts are generated from shipment state after recovery is accepted. Sending is explicit. That keeps freight forwarding automation honest: the software removes composition labour, not accountability.

Operators can ask questions grounded in the workspace:

Answers come from tools over Convex data, not from an unconstrained model inventing AWBs. Without an LLM key, the demo still runs on mock tools against live records so the workflow can be evaluated.

Observe → reason → recommend → approve → execute. Automatic execution inside policy is a later option, not the current default. If a vendor tells you the AI will run the branch unattended, ask what happens when the recovery flight has DG restrictions or the consignee SLA cannot absorb another day.

Alongside CargoWise coexistence, this is the point of the product: an intelligence layer, not an unsupervised replacement for the operator or the TMS.