Customs classification copilot

Customs classification you can defend.

Leviqa reads the shipment and suggests the HS/HTS code with the rule that backs it. When the shipment doesn’t say enough, it asks for what’s missing instead of guessing. You review and approve.

No document at hand?Look up a code from a product description
The instrument

Every item that crosses the border needs a code.

Origins worldwide converge on the importing market — and Leviqa classifies each good with the rule cited.

OriginsImporting markets
today

Type and guess

Codes filled in by hand, without certainty. Every shipment is hours of paperwork and the doubt of whether the code was right — or not.

with leviqa

Review and approve

The suggestion arrives ready, with confidence and the cited rule. You confirm in seconds — and every correction tunes your tenant’s model.

the cost of getting it wrong

Two similar codes. Very different bills.

Misclassifying isn’t a technical detail — it’s retroactive charges, fines, and held cargo. And the bill lands on the importer, even with a broker in the middle.

That’s why Leviqa never gives a verdict without grounding. Every suggestion comes with the rule that backs it — so every approval can withstand an audit.

  • $22.8Msettlement over misclassified imports (False Claims Act)
  • $400M+in unpaid duties uncovered by CBP through EAPA
  • 2,100+penalties issued under 19 U.S.C. § 1592

Public cases and primary CBP figures.

unchecked code
100W USB-C cable · same invoice
8544.70.0000
Import duty rate25%
Landed cost$61,240

+ $83K charged retroactively across six shipments.

grounded code
100W USB-C cable · same invoice
8544.42.9090
Import duty rate2%
Landed cost$49,000

Classified by the cited rule — GRI 1, Note 8 of Ch. 85.

Compare the same item under two HS codes

Drag the divider — the goods are the same; only the code changes.

It speaks the language of customs. Every code comes with the rule that backs it.

You review and approve. The AI never decides on its own.

GRI
the general rules of interpretation
Chapter notes
the legal text that decides
Knows when it doesn’t know
low confidence goes to review, never auto-suggested

how quality is measured

Against real shipments verified by authority — a CBP ruling (CROSS) or a licensed broker. Including the hard pairs: 2% vs. 25%. And every correction a broker makes becomes a new verified case: the bar rises with use.

From document to declaration, without the grunt work.

AI takes typing and guessing out of the path. Scroll to watch the shipment advance — each step delivering something ready to check.

  1. 01

    Documents

    Invoice, packing list, and BL come in as they are — PDF, photo, or spreadsheet.

    shipment SHA-2291receiving
    • commercial-invoice.pdf184 KB
    • packing-list.xlsx32 KB
    • bill-of-lading.jpg1.2 MB

    PDF, photo, or spreadsheet — they come in as they are, no manual formatting.

  2. 02

    Extraction

    Items are read and structured automatically, no typing.

    extracted itemsreading
    ItemQty
    100W USB-C cable1,200
    65W GaN charger800
    7-in-1 USB hub450

    Structured automatically — zero typing.

  3. 03

    Classification

    The AI suggests the HS/HTS code and cites the rule that backs it.

    classificationsuggesting
    100W USB-C cable
    8544.42.9090
    97%
    Confiança alta

    GRI 1 · Note 8 of Ch. 85. Conductor with a connector — classified by function, not by material.

    Every suggestion comes with the confidence level and the rule that backs it.

  4. 04

    Landed cost

    Duty, Section 301, MPF/HMF, and freight summed — a planning view of the import cost.

    landed costsumming
    Customs value
    $48,000
    Duty (2.6%)
    $1,248
    Section 301 (25%)
    $12,000
    MPF + HMF
    $226
    Freight + insurance
    $3,174
    Real cost
    $64,648

    The number that decides sourcing — before you buy, not at the border.

  5. 05

    Declaration

    The declaration arrives pre-filled, ready to review and approve.

    declarationready
    Importer
    Acme Imports LLC
    Regime
    Formal entry
    Classified items
    3 of 3
    Declared total
    $64,648
    Review and approve

    Pre-filled and auditable — you confirm in seconds.

Trade in motion

Every route carries a decision.

Leviqa turns global flows into traceable classification decisions — from document to declaration.

Preparing the film

Decision infrastructure for trade that never stops: local rules, visible evidence, and confidence in every market.

What people usually ask before getting started.

Does Leviqa file the declaration with the government?

Today Leviqa assists the decision — it extracts the items, suggests the code with the cited rule, and pre-fills the declaration. Official filing with the agency is the product’s next frontier.

Can I override the AI’s suggestion?

That’s the whole point — Leviqa proposes, you decide. You can accept, correct, or replace any suggested code, and the change is recorded in the append-only trail with who made it and when. Your corrections also tune the model for your tenant only.

What happens when Leviqa isn’t sure?

It says so, instead of guessing. Every suggestion carries a confidence level, and low confidence is routed to mandatory review — it is never auto-suggested. When the description doesn’t support a code, Leviqa asks for the missing attribute instead of picking one for you; when nothing supports it, it returns no code at all. And when the national tariff line is ambiguous, it presents the candidates for you to decide instead of assuming a duty rate.

Where does the confidence in each suggestion come from?

Every code comes with a confidence level and the rule that backs it (GRI and chapter notes), plus the alternative codes that were ruled out. Codes come from the tariff schedule we ingest, with the version and effective date recorded on the decision — not from open-ended generation. Never a verdict without grounding.

How do you measure the quality of the suggestions?

Against a golden set: real items whose correct code was verified by an authority — a CBP ruling (CROSS), a licensed broker, or an already accepted entry. The set is frozen and the engine runs blind, and every model or prompt change re-runs against it, so a regression shows up before it ships. It includes the hard pairs (2% vs. 25%) on purpose, and it grows from real corrections. We are not publishing an accuracy figure yet — our sample is still too small to state one honestly.

Does it work outside the United States?

Today Leviqa is available for imports into the United States (HTS). Each new country goes through its own validation before going live — we’d rather cover well what we’ve validated than promise untested coverage. More countries are on our roadmap.

How is my data protected?

Per-tenant isolation with Row-Level Security, encryption in transit (TLS) and at rest (KMS). Every correction you make tunes only your tenant’s model.

The brand

Light on the operation, exact on the decision.

Classify with confidence.

Stop guessing the code. Start approving with proof.

Bring your operation’s customs classification up to the pace of review and approve.