Space
It knows where everything is.
Facility, zone, rack, position, container, package. Every unit has an address, a container and a path to it, whether a person or a robot fetches it.
The warehouse that earns its way to lights-out.
WarehouseAI keeps a factual model of every package, container and position, and its Brain decides what should happen next. People, equipment and robots work from the same tasks with the same evidence. Each zone becomes more autonomous only when its measurements show it is ready.
Eight lines reserved in one transaction from grid bins. The ready-time promise comes from queue length and port capacity, not a fixed SLA.
Illustrative replay from the WarehouseAI specification. Not customer data.Promised12:20Predicted12:20
For warehouse owners and operators: small-parts 3PL, quick commerce and dark stores, courier and parcel depots, distribution, cross-dock and specialist storage.
Four things a warehouse system has to get right before it can run a building.
Most warehouse software records what people say happened. WarehouseAI is built to know where everything is, how sure it is, what it may decide and who or what can do the work.
Twin
It says how it knows.
Observed, inferred, planned, disputed and unknown are different states. “P123 in B08 at 10:02” is a fact about the last scan, not a promise that it is still there.
Brain
It decides within authority.
Policies handle the routine cases. Reasoning is used only where they run out. Each action type has its own authority: observe, recommend, approve or act.
Execution
Anyone, or anything, can do the work.
Scanner users, forklifts, goods-to-person ports and robot cells all work from the same tasks and must produce the same evidence.
Person or robot, the record is the same.
A put-away done with a handheld scanner and one done by a grid robot give the same receipt, movement, ledger posting and completion. Only the evidence differs. That is what lets you automate a zone without retraining the rest of the operation.
Task PA-2201Put away tote T501 from decant station D-2 to grid position B08
| Record | Done by an operator | Done by the grid robot |
|---|---|---|
| Receipt state | Accepted | Accepted |
| Movement | D-2 → B08 | D-2 → B08 |
| Ledger posting | −1 D-2 · +1 B08 | −1 D-2 · +1 B08 |
| Task state | Completed | Completed |
| Evidence | Handheld scan HS-14, 10:02:11 | Grid controller ack GC-1, 10:02:09 |
| Certainty | Scan-confirmed | Controller-confirmed |
Every zone climbs at its own pace.
Six autonomy levels, measured per facility, zone and process. A zone moves up only after a shadow run, sustained thresholds, a rollback plan and a named approval. It moves down automatically when the numbers slip.
L4
Remotely supervised
No routine on-floor labour in the zone. Exceptions are resolved by automation or remotely.
- People
- Supervise from a remote operations workspace and attend on call.
- WarehouseAI
- Resolves exceptions with versioned playbooks and escalates only when none applies.
One facility, six zones, four different levels
Autonomy is set per facility, zone and process, so a robot grid can run beside manual loading. The highlighted zones are the ones at the level you picked.
ReceivingL2
DecantL3
Storage gridL4
Pick portsL3
PackingL1
LoadingL1
A Brain that knows what it is allowed to decide.
Routine cases run on deterministic policy. Reasoning is used only for novelty, conflict or uncertainty. Every action type has its own authority, and every consequential decision leaves a receipt.
Observe onlyThe Brain watches and records. Nothing is proposed.
RecommendIt proposes an action with evidence. A person decides.
Approval requiredIt prepares the action. A named role approves it before anything moves.
Bounded automaticIt acts inside limits and reports what it did.
Decision receipt DR-4471, illustrativeAct
Lane D2 is blocked. Route R42 departs in 30 minutes and its packages are staged for D2.
AuthorityBounded automatic
RecordedInputs, options, policy, outcome
Not done by the BrainMoving any equipment
One core for anything that uses a warehouse.
Segments are operating profiles and add-ons on one core, not forks of it. A new kind of warehouse business is configured from extension points. A new add-on is needed only for genuinely new equipment or integrations.
Small-parts 3PL
Many clients sharing space and automation, each with their own goods, data and commitments.
CoreContract logisticsGoods-to-personRobotic handlingParcel carriersBillAI
Quick commerce and dark stores
Minute-level promises on your own stock, with honest availability and no silent substitutions.
CoreQuick commerceGoods-to-personSpecial handlingCommerce
Courier and parcel depots
Parcels are promised to recipients before they arrive. The depot’s job is to sort them and send them onward on time.
CoreCourierConveyors and sortationTransport
B2B distribution and wholesale
Case and pallet flows with appointments, manufacturer replenishment and returns.
CoreTransportParcel carriersSpecial handling
Cross-dock
Goods move from receipt to outbound with little or no storage.
CoreConveyors and sortationTransport
Specialist storage
Cold chain, dangerous goods, high-value, bonded and regulated goods.
CoreSpecial handlingRegional template packs
WarehouseAI runs the building. Everything else connects by contract.
Orders come in as JSON, packages go out to LogisAI or any carrier, and verified usage flows to billing and the books. Each system owns its own decisions. None of them needs a third system to coordinate.
Order in
Product codes and quantities as JSON
Allocate
Stock reserved once, never twice
Pick and pack
By people, ports or robot cells
Label
Applied, then verified
Load and hand over
Custody evidence per package
Route and deliver
Routes, drivers and outcomes
Measured
Verified usage: storage hours, picks, labels, handovers
Rated and invoiced
Invoices to your own clients, from your own BillAI seller account
Booked
Documents, payments, warehouse costs and statements
WarehouseAI measures. BillAI rates and invoices. AccountsAI keeps the books.
Written down before it is built.
WarehouseAI is in development. The platform is specified as 409 requirements, each with an owning component, a delivery stage and a testable acceptance criterion. Seven stages take it from platform foundation to lights-out.
1. Platform foundation
Invariants proven by automated tests.
2. Directed manual operations
L1Audited manual journeys for each operating profile.
3. Integrations and segments
End-to-end 3PL, quick-commerce and courier scenarios pass.
4. Equipment and hybrid automation
L2–L3Automated transfers confirmed, recoverable and equivalent to manual.
5. Autonomous operations
L4No routine on-floor labour in promoted zones.
6. Optimisation and intelligence
Measured improvement over baseline.
7. Lights-out operations
L5Lights-out readiness criteria met.
Bars show how many requirements are first due in each stage. Each stage ends when its exit evidence passes.
161 invariants. Rules that must hold on every path, including retries, concurrency and add-ons. Each is tested by trying to break it.
20 reference workflows. These run from provisioning a tenant to recall and lot trace. Each one runs manually first, then on emulated equipment, then on commissioned equipment.
Design partners shape the order. The recommended first profile is small-parts 3PL on goods-to-person storage, followed by quick commerce and then courier. Partners’ floors decide whether that holds.
Bring us the warehouse you have.
Design partners get the first operating profiles built and tested against their own floor, SKUs and order history. You start with directed manual work and climb from there.