# AI-native architecture

Tranexio’s AI-native architecture treats AI as a core capability: business modules call a single AI gateway, and every call is logged, explainable and reviewed by a person. HS classification suggestions are already used in customs, while document recognition, intelligent alerts and other capabilities will follow the roadmap.

## Key points

- **A single AI gateway**：Every AI call goes through one entry point, governed by tenant switches, roles, daily call quotas and cost limits, and it degrades gracefully when an external service is unavailable without disrupting business workflows.
- **Every call logged**：Each call records an anonymised input summary, the output, the rationale and the user, in an append-only log that can be traced by business document on the audit trail page.
- **Human in the loop**：AI only makes suggestions. Key decisions are accepted or overridden by business users, and review records and reasons are stored together to meet audit and regulatory requirements.
- **HS classification suggestions**：Enter a product name and its declaration elements to receive the top three HS codes with confidence scores, rationale and distinguishing points; once a customs officer confirms, the result is written to the product compliance record.
- **Roadmap clearly labelled**：Document recognition and comparison, intelligent alerts and a conversational assistant are planned and will be released along the roadmap; until then they are not offered as current features.

## The problem

In trade, product classification, document checking and risk judgement depend heavily on individual experience. Plugging in general-purpose AI tools directly brings its own problems: data leakage, unexplainable results and no accountability. Tranexio builds the governance foundation first and then releases capabilities one by one: AI in the core, not bolted on.

## What the platform does

| Layer | Available now |
|---|---|
| AI gateway | Single entry point; tenant switches, role authorisation, daily call quotas and cost limits; automatic fallback when unavailable |
| Logging and review | Append-only call logs with anonymised inputs and summaries; human review records; audit trail page |
| Business capability | [HS code](https://tranexio.com/en/resources/glossary/hs-code/) classification suggestions: top three codes, confidence score, rationale and distinguishing points, confirmed by a person |
| Governance documentation | Model cards, explainability notes and model lifecycle management |

## Capability roadmap

The following capabilities are **planned** and will be released along the roadmap; they are not current features:

- Document recognition and multi-document comparison (planned)
- Intelligent checking of L/C terms and compliance (planned)
- Intelligent risk and anomaly alerts (planned)
- Conversational business assistant (planned)

## Key controls

- AI output never overwrites business data directly; it must be confirmed by a person.
- Every suggestion can be traced to the model, input summary, rationale and the person who accepted it.
- Gateway switches and limits are controlled by tenant administrators and can be turned off at any time.

## FAQ

### What can Tranexio’s AI do today?

Today it provides HS classification suggestions in customs: the top three codes with confidence scores and rationale, confirmed by a customs officer and logged. It also provides AI governance, including a single gateway, call logging, human review and quota controls. Document recognition, intelligent alerts and a conversational assistant are planned.

### Are AI classification results used directly in customs declarations?

No. HS classification suggestions are only candidates. A customs officer must accept or override them before they are written to the product compliance record and used in declarations, and every acceptance, override and reason is logged for later review and audit.

### Will our business data be used to train external models?

Nothing is sent out by default. AI calls go through the single gateway, and inputs are anonymised before a summary is stored. Whether an external model channel is used is controlled by a tenant administrator switch, and private deployments can keep data entirely within the company’s environment.

### What does “AI-native” mean, and how is it different from an AI plug-in?

AI-native means AI is embedded in the platform core through a shared capability interface, governed alongside permissions, audit and business documents. Plug-ins typically bypass business permissions and audit trails. In Tranexio every AI call happens in a business context and can be traced to a specific document and user.

---
Source：https://tranexio.com/en/platform/ai/ · Updated 2026-10-06 · © 上海赛淘数据技术有限公司 / Shanghai Saitao Datatech Co., Ltd