Skip to main content
A scribe session can produce one or more medical record documents — one per purpose (e.g. primary for the clinician-facing record, rpa for EMR injection). Each document is the filled output of the session templates.

Listing the documents

Use the same session id you used with the scribe session endpoint — either your consultationInternalId or the session id (ss_...) the API returns:
Response:

Doctor edits

The medicalRecord is the record the doctor has in Scribe, including the text they typed or dictated after the AI generated it. When the doctor edits a finished record, Telepatia sends a scribe_session.updated webhook; fetch the documents again to read the new text, and compare updatedAt to tell that the record changed. The items come in no guaranteed order of purpose. Select the record you need by its purpose (primary is the consultation record), not by its position in the list.

Transcript

The response carries the session transcript next to items: one transcript for the session, the same for every document. transcript.text is the full consultation transcript as plain text: the text Telepatia read to fill the records. transcript.updatedAt is when the stored transcript last changed. transcript is null in these cases. A null transcript does not change the documents: they are still returned.

Selector codes

A template field can limit its answer to a list of options, as a selector in your EMR does. Each option is a { label, value } pair in the field enum (see Enum options): the label is the text a person reads, and the value is your code for the option (for example, the option id in your EMR catalog).
  • Where: the code is the field value in medicalRecord, at medicalRecord.<sectionKey>.<fieldKey>. The keys are the ones you set in the template.
  • Format: the value string exactly as you defined it. A multi-select field (an array whose items have an enum) returns a list of value strings. A field with no answer in the consultation returns null, or its default when you set one.
  • How Telepatia selects it: the model reads transcript.text, gets the list of label and value pairs of the field, and selects the options whose label matches what the doctor and the patient said. The record stores only the value, so you can change or translate a label and your integration still receives the same code.

Diagnosis codes

Each document carries its diagnosisCodes — the diagnosis ICD-10 (CIE-10 / CID-10) codes, as the doctor-curated set (the model’s suggestions plus any codes the doctor added, minus the ones they removed). Each entry has a type of primary or secondary. The list is empty when the session has no diagnosis codes. Past medical conditions carry their own resolved codes in medicalRecordSummaryStructured.pastMedicalConditionsCoded: each entry pairs the condition content with a code object holding the best-match code and scored options. It is null until code resolution has run for the record.
Wait until the session’s status is completed or completedWithErrors (see Session lifecycle) before fetching its documents — they are only produced after the AI pipeline finishes.

CDSS alerts

When clinical decision support is enabled for the session, each document carries its cdssAlerts — corrective alerts that flag omissions, incorrect dosages, or non-pertinent orders against the record. Each alert has a stable opaque id (use it as-is; don’t parse it), the clinical elementId it refers to, an actionCategory (add | remove | replace | modify), a severity (red | orange | yellow), and a status reflecting the doctor’s decision (pending | accepted | rejected). It is null when CDSS is disabled for the session or no alerts were generated, and [] when it ran and produced none. See Clinical decision support for the full alert model.

Filtering by purpose

Pass ?purpose=primary to get only the clinician-facing document. Other valid values include rpa (for EMR injection) and emr.

Field reference

The response also has these fields next to items: