Skip to content
FAS Localize

Vietnamese translation and audit API

The FAS Vietnamese API translates from 34 languages into Vietnamese, audits the result against 118 cited standards rules and gives every segment a calibrated confidence, so your team reviews only what needs a person.

  • Translate
  • Audit
  • Triage
  • Early access

Every segment lands in one lane

  • Auto-approve AUTO

    confidence 0.85 or higher, no hard flag

  • Quick review QUICK

    confidence 0.60 to 0.85, or a soft flag

  • Must review MUST

    below 0.60, a counted audit error, or a high-risk segment

Default thresholds. Numbers, money, dates and regulated text always land in must review.

Three jobs, one call

  • Translate into Vietnamese

    From 34 source languages, English, Chinese, Japanese and Korean among them. Your translation memory is checked first. In a batch job each segment is then routed by risk: short UI strings go to one fast model, the rest to a four-model ensemble whose drafts an anonymised arbiter picks from, edits or rewrites. Small requests can take the fast path: one model, answered in the same call.

  • Audit against Vietnamese standards

    118 rules with legal and academic citations, run in layers from file integrity to document-type compliance. Every issue comes with its rule, its place in the text, a suggested fix where one exists and the source it rests on.

  • Know what to review

    Model agreement, cheap pre-checks and an independent judge model combine into one calibrated confidence per segment, which sorts it into auto-approve, quick review or must review. Numbers, money, dates and regulated text always go to a person.

How a segment moves through it

The batch path, step by step. Small requests on the fast path skip the ensemble and the arbiter: memory, one model, then the same audit.

  1. 1

    Translation memory

    Exact approved matches are reused without a model call; close matches become hints.

  2. 2

    Risk routing

    Each segment gets a tier: memory, low, standard or high (numbers, money, legal, medical).

  3. 3

    4-model ensemble

    Claude, Gemini, GPT and Qwen draft in parallel, one per vendor family.

    • Claude · Anthropic
    • Gemini · Google
    • GPT · OpenAI
    • Qwen · Alibaba
  4. 4

    Anonymised arbiter

    Sees shuffled, unlabelled drafts and picks, edits or rewrites the best one.

  5. 5

    Standards audit

    118 cited rules across layers L0 to L5, plus an MQM-style score.

  6. 6

    Confidence and triage

    One calibrated score per segment: auto-approve, quick review or must review.

Six audit layers

L0 Integrity
Encoding, tags and placeholders, untranslated leftovers, length limits.
L1 Mechanical
Spelling and diacritics (hỏi/ngã), capitalisation, numbers, units, dates, punctuation, place names.
L2 Normative
Standards rules that need judgement, answered by a model on request.
L3 Equivalence
Numbers, dates and units that changed between source and target; omissions and additions.
L4 Consistency
Terminology, names and forms of address across the whole document.
L5 Compliance
Document-type profiles: administrative documents, contracts, labels, ads, accounting.

Rules you can trace

118 rules (64 core, 54 for document types) carry 137 citations to 72 legal and academic sources. Examples:

  • Decree 30/2020/NĐ-CP on administrative documents (format, capitalisation, abbreviations)
  • Decree 78/2025/NĐ-CP as amended by Decree 187/2025/NĐ-CP (format and capitalisation of legal documents)
  • Accounting Law 2015 (88/2015/QH13), Article 11 (numbers and currency in accounting)
  • Advertising Law 2012 (16/2012/QH13), Article 18 (Vietnamese in advertising)

Reports come with an MQM-style score and a pass, warn or fail verdict per segment. A finding flags possible non-conformity; it is not legal advice. Try the rules now in the free Vietnamese Standardization Checker.

JSON in, JSON out

Base URL https://api.faslocalize.com. Authenticate with your key as a Bearer token. Keys are issued during early access.

Translate two strings (curl)
curl https://api.faslocalize.com/v1/translate \
  -H "Authorization: Bearer $FAS_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: 3f1c0d2e-order-page" \
  -d '{
    "sourceLang": "en",
    "targetLang": "vi",
    "segments": [
      { "id": "checkout.title", "source": "Review your order" },
      { "id": "settings.save", "source": "Save changes" }
    ],
    "options": { "score": true }
  }'
Audit a translation (JavaScript)
const res = await fetch("https://api.faslocalize.com/v1/audit?format=json", {
  method: "POST",
  headers: {
    Authorization: `Bearer ${process.env.FAS_API_KEY}`,
    "Content-Type": "application/json",
  },
  body: JSON.stringify({
    sourceLang: "en",
    targetLang: "vi",
    segments: [{ id: "1", source: "Order on 03/04/2026", target: "Đặt hàng ngày 03/04/2026" }],
    options: { profiles: ["advertising-marketing"] },
  }),
});
const report = await res.json();
// report.summary: { segments, issues, bySeverity, byLayer, mqmScore, passRate }
// report.segments[0]: { segmentId, verdict: "pass" | "warn" | "fail", issues: [...] }
Response from /v1/translate, trimmed
{
  "documentId": "…",
  "segments": [
    {
      "id": "checkout.title",
      "source": "Review your order",
      "target": "Xem lại đơn hàng",
      "tier": "low",
      "score": {
        "segmentId": "checkout.title",
        "confidence": 0.9,
        "triage": "AUTO",
        "reasons": ["…"]
      }
    },
    { "id": "settings.save", "target": "Lưu thay đổi", "tier": "low", "score": { "…": "…" } }
  ],
  "audit": {
    "rulesetVersion": "1.0.0",
    "summary": { "segments": 2, "issues": 0, "mqmScore": 100, "passRate": 1 }
  },
  "providersUsed": ["…"],
  "requestId": "…",
  "stats": { "tmReused": 0, "tmHinted": 0, "scored": 2 }
}

Example values for illustration, not measurements. Each segment also carries its risk tier, and batch jobs add the arbiter’s verdict (pick, edit or rewrite) and the candidate drafts.

Endpoints

POST/v1/translate
Fast path: up to 20 segments, synchronous. Memory, one model, audit.
POST/v1/jobs
Job path: up to 2,000 segments, asynchronous. Ensemble, arbiter, audit, triage.
GET/v1/jobs/{id}
Job progress and, when finished, the result.
POST/v1/audit
Audit existing translations. JSON report, Markdown or review rows.
POST/v1/score
Calibrated confidence and triage lane for a chosen translation.
POST/v1/tm/query
Search your translation memory; upsert and TMX import next to it.
GET/v1/rules
The rule catalogue with legal citations, filterable by layer and profile.
GET/v1/usage
This month's and last month's usage for your key.

Every POST accepts an Idempotency-Key: a retry within 24 hours returns the first answer instead of running again. Errors return a stable code, an HTTP status and a request id; rate limits answer 429 with Retry-After.

Your files, left intact

The API takes segments as JSON. The command line, the GitHub Action, the MCP server and the connectors read these formats and write translations back in place, without reflowing your markup, placeholders or plural forms.

  • JSON

    flat, nested, i18next, next-intl

  • XLIFF

    1.2 and 2.0

  • Gettext PO

    plural forms kept

  • TMX

    memory import and export

  • CSV

    source, target, key, notes

  • HTML

    text and attributes

  • Markdown

    code blocks untouched

  • Apple

    .strings and .xcstrings

  • Android XML

    strings, arrays, plurals

  • Subtitles

    SRT and VTT, line limits

  • Plain text

    one segment per line

Plans for early access

Pick what fits and request access: we confirm the plan and its terms with you before issuing a key. For human translation by our linguists, see translation pricing.

Translation API

Pay per word translated. Every tier includes the standards audit.

  • AI only

    Fast, synchronous translation for strings and short content.

    Early access
    • Translation memory, then one model
    • Audit layers L0 to L5
    • Calibrated confidence per segment
    • Up to 20 segments per call
  • Ensemble economy

    The full ensemble for work that can wait for the providers' batch queues.

    Early access
    • Four model families plus arbiter
    • Full audit and triage lanes
    • Batch jobs up to 2,000 segments
    • Runs on provider batch queues
  • Ensemble priority

    The same output, started at once for work that cannot wait.

    Early access
    • Four model families plus arbiter
    • Full audit and triage lanes
    • Batch jobs up to 2,000 segments
    • Started immediately, no batch queue
  • Human in the loop

    The ensemble, then a FAS native reviewer on every segment.

    Early access
    • Everything in the ensemble
    • Native Vietnamese review, effort set by triage
    • Audit report with citations
    • Invoiced directly as a service

Website and app plans

A monthly allowance of new or changed words. Repeats from your memory do not count.

Checker API

The cited rule set on its own: audit Vietnamese you already have, no translation.

  • Free

    Try the checker API on your own content.

    Early access
    • 5,000 words a month
    • Rule layers, no model calls
    • JSON, Markdown or review rows
    • Instant key, usage dashboard
  • Developer

    The checker in your product or pipeline.

    Early access
    • Up to 2 million words a month
    • Rule set updates
    • Every layer that needs no model call
  • Team

    Higher volume and every document-type pack.

    Early access
    • Up to 20 million words a month
    • All 6 document-type packs
    • Change log with legal citations
  • Platform

    For TMS vendors and agencies building it in.

    Early access
    • Up to 200 million words a month
    • Use inside your own product
    • Annual contract, direct invoice

API questions

What is the FAS Vietnamese API?

An API that translates from 34 source languages into Vietnamese and audits Vietnamese text against 118 cited standards rules. Each segment comes back with a calibrated confidence and a review lane, so your team reads only what needs a person.

How is it different from a machine-translation API?

A plain MT API returns one draft. Ours checks your memory first, routes by risk, has four model families from different vendors draft the harder segments, lets an anonymised arbiter choose, then audits the result against Vietnamese standards and tells you how much to trust each segment.

Is it the same as the free checker on this site?

The free Vietnamese Standardization Checker checks one text at a time against the same standards database. The API adds translation, memory, the whole-document consistency layer, confidence scoring, larger batches and machine-readable reports for your own tools.

Does an audit make a translation legally compliant?

No. Each finding points to the text it rests on so an editor can decide. It flags possible non-conformity; legal conclusions need qualified counsel, and certified or notarized translations remain a human service.

Which source languages and file formats are supported?

34 source languages into Vietnamese, including English, Chinese, Japanese, Korean, French, German, Thai and Indonesian. The API takes segments as JSON; the CLI, GitHub Action, MCP server and connectors read and write JSON, XLIFF, PO, TMX, CSV, HTML, Markdown, Apple and Android strings, SRT and VTT, and plain text.

How fast and how accurate is it?

We publish speed and accuracy figures only once they are measured on real projects. During early access we share the measurements from your own pilot with you.

How do I get a key?

For the free checker plan (5,000 words a month, rule layers), sign up with your email at /developers/signup and create a key in your dashboard straight away. For translation, more volume or a contract, send the request form on this page: we reply within 24 hours, agree on the use case and limits with you, and issue a key scoped to what you need.

Can a person review the output?

Yes. Your own reviewers can work the must-review and quick-review lanes, or FAS native linguists can review them for you as a human-in-the-loop service.

Request API access

The FAS Vietnamese API is in early access. Tell us what you want to translate or check, and we set you up.

  1. We read your request

    A person at FAS replies within 24 hours, Vietnam time (UTC+7).

  2. We agree on the scope

    Use case, source languages, volume, the integration you need and how your data is handled.

  3. You get a scoped key

    Only the permissions you need, with a monthly limit, and our help with the first integration.

Prefer email? business@faslocalize.com

Integrations you are interested in

We reply within 24 hours · Vietnam time (UTC+7)

We use your details only to answer this request. See our privacy policy.