Documentation (coming soon)

De-identification

Model: aug-deid · Status: Coming soon · Scope: agents:deid · Schema: extraction/0.1

Finds and removes personal data from clinical text before it is stored, shared, used for research or sent to other services: names, phone numbers, addresses, Aadhaar / ABHA / PAN numbers, MRNs, emails, dates, ages over 89, facility and clinician names.

Request

JSON

{
  "model": "aug-deid",
  "messages": [{"role": "user", "content": "Ramesh Kumar (MRN 448812, ph 98450 12345) seen on 3/11/2026 at Sunrise Clinic, Indiranagar…"}],
  "augsidius": {"instructions": {"mode": "mask", "keep": ["clinical_dates_shifted"]}}
}

`mode`	Result
`mask`	Replace with type tags: `[NAME]`, `[PHONE]`, `[ID]`
`surrogate`	Replace with realistic fake values, consistent within the document
`detect`	Return spans only; text unchanged

keep options: clinical_dates_shifted (dates shifted by a constant per document), age, city.

Response payload

JSON

{
  "text": "[NAME] (MRN [ID], ph [PHONE]) seen on [DATE] at [FACILITY], [LOCATION]…",
  "entities": [
    {"type": "name", "span": {"start": 0, "end": 12}},
    {"type": "id", "subtype": "mrn", "span": {"start": 18, "end": 24}},
    {"type": "phone", "span": {"start": 29, "end": 40}}
  ]
}

Use aug-deid first in pipelines that send text to analytics or external tools. De-identification reduces but does not eliminate re-identification risk; combine it with access controls.

Care plan drafting

Patient interview