Azure Language PII compatibility

Move supported Azure text, conversation, and document PII workloads to ShinrAI.

Migrating from Azure Language PII

Use your ShinrAI API key in Ocp-Apim-Subscription-Key and point the client at the ShinrAI API base URL. The text PII adapter implements the supported request and response contracts below; model predictions and full Azure feature coverage are separate concerns.

curl 'https://api.shinrai.innovius.io/language/:analyze-text?api-version=2023-04-01' \
  -H "Ocp-Apim-Subscription-Key: $SHINRAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"kind":"PiiEntityRecognition","analysisInput":{"documents":[{"id":"1","language":"en","text":"Email emma@example.com"}]},"parameters":{"modelVersion":"latest","stringIndexType":"Utf16CodeUnit"}}'

{
  "kind": "PiiEntityRecognitionResults",
  "results": {
    "documents": [
      {
        "id": "1",
        "redactedText": "Email ****************",
        "entities": [
          {
            "text": "emma@example.com",
            "category": "Email",
            "offset": 6,
            "length": 16,
            "confidenceScore": 1.0
          }
        ],
        "warnings": []
      }
    ],
    "errors": [],
    "modelVersion": "shinrai-v1.3"
  }
}

Synchronous text PII allows up to 5 documents of 5.120 grapheme clusters each. Asynchronous jobs allow up to 25 documents and 125.000 Unicode 8 text elements in total, with a 1 MiB request limit. Follow the returned operation-location URL for polling. Job access and cancellation are restricted to the creating account; encrypted job data expires 24 hours after submission.

Offset modes include UTF-16 code units, Unicode code points and Unicode 8 text elements. The actual model is ShinrAI 1.3; dated Azure model identifiers are rejected. The current model does not cover every Azure PII category or modality. Unsupported category/options requests fail explicitly. This is a text PII adapter, with additional compatibility detail in the release matrix; it is not a claim of Azure-wide feature or prediction parity.

See record examples →