Skip to main content

Changelog & Upgrades

Updates to the Nabla Core API — including new features, fields, or endpoints — are released under a new-dated version to ensure stability and predictability. This means these changes won’t affect your integration unless you explicitly opt into the newer version (see Targeting a specific version).

That said, some updates — like internal AI models improvements, bug fixes, or strictly non-breaking changes that don’t affect the schema or observable behavior — may be rolled out across all versions.

Below is a list of all available Core API versions, along with their associated changes.

2026-07-01​

  • New Server API endpoints for E/M (Evaluation & Management) CPT code generation:

    • POST /generate-em-codes-async — Submit a transcript and note to generate E/M CPT codes asynchronously. The endpoint infers the visit type (if not provided) and determines the MDM level in parallel, then computes the appropriate CPT codes. The result is kept for two hours.
    • GET /generate-em-codes-async/{id} — Poll the status and results of an in-progress or completed E/M code generation.
    • POST /compute-em-codes — Deterministic, fast recomputation of E/M CPT codes from structured encounter inputs (no transcript or note required). Use to recompute codes after the user changes encounter info such as modality, payer, or total encounter time.

  • New asynchronous normalized data generation endpoints on both Server API and User API:

    • POST /generate-normalized-data-async — start an asynchronous job to extract ICD-10 and LOINC codes from a Nabla-generated clinical note in a FHIR-compliant structure.
    • GET /generate-normalized-data-async/{id} — poll the status and retrieve the results of an asynchronous normalized data generation job.

  • Dictation WebSocket (/dictate-ws) — the server may now emit a new ASYNC_CORRECTION item on both the Server API and User API. This item carries a suffix and replacement pair instructing the client to replace the trailing portion of the dictation that matches suffix with replacement, allowing the server to refine or fix previously streamed DICTATED_TEXT segments after additional processing (e.g. punctuation, formatting, model-based corrections). Clients opt in by setting the new enable_async_corrections boolean (defaults to false) on the initial CONFIG frame.

  • Template-level note customization — optional note_template_custom_instructions (nullable string, same 700-character limit as section custom_instruction) for note-wide guidelines on a template:

  • ChartSync — follow-up note generation. Replaces the free-text draft_assessment_and_plan input with a structured source_note that describes both the provenance of the seed content and the semantic role of each section. See the ChartSync guide for more details.

    • structured_context.draft_assessment_and_plan has been replace by structured_context.source_note in /generate-note and /generate-note-async request bodies on both Server and User APIs.

    • New supported_source_note_section_kinds field on the template detail endpoints, listing the semantic kinds the template accepts as source-note inputs.

    • Migration from draft_assessment_and_plan. The previous field accepted a single free-text string used to seed the Assessment & Plan section. To migrate, move that string into the new structure:

      // Before
      {
        "structured_context": {
          "draft_assessment_and_plan": "1. Type 2 diabetes — continue metformin..."
        }
      }

      // After
      {
        "structured_context": {
          "source_note": {
            "type": "DRAFT",
            "sections": [
              {
                "kind": "ASSESSMENT_AND_PLAN",
                "text": "1. Type 2 diabetes — continue metformin..."
              }
            ]
          }
        }
      }

2026-06-12​

  • Haitian Creole transcription support. Haitian Creole (HAITIAN_HT) can now be used as a speech locale value on the transcription endpoints (/transcribe, /transcribe-async, /transcribe-ws) of both the Server and User APIs. Support is experimental and access is limited — please reach out to get access. Haitian Creole cannot be combined with any other locale: when transcribing Haitian Creole, speech_locales must contain exactly one entry. This change also applies to all prior versions that take a speech_locale or speech_locales too — you can specify HAITIAN_HT on most of the latest API versions.
  • Removed family_history field from normalized data. The family_history field has been removed from the NormalizedData response object on both User and Server APIs (/generate-normalized-data). Family-related conditions are already covered through dedicated ICD-10 codes in the standard conditions field.
  • Custom note templates — templates and sections are now identified by a freetext key (string) instead of a fixed enum. Changes across all existing endpoints:
    • note_template (enum) note generation request bodies has been renamed to note_template_key (string). This only impacts the Server API.
    • template (enum) in note generation responses has been renamed to template_key (string).
    • Section identifiers (key / section_key) are now freetext strings instead of a fixed enum.
  • New Server API endpoints to browse available templates:
    • GET /generate-note/templates?locale={locale} — list all templates available to the organization.
    • GET /generate-note/templates/{template_key}?locale={locale} — get metadata and sections for a single template.
  • New User API endpoints for template customization:
    • GET /note-settings/templates — list all templates available to the authenticated user.
    • GET /note-settings/templates/{template_key} — get metadata and sections for a single template.
    • GET /note-settings/templates/{template_key}/customization — get note template customization.
    • PATCH /note-settings/templates/{template_key}/customization — update note template customization.
    • GET /note-settings/templates/{template_key}/customization and PATCH /note-settings/templates/{template_key}/customization replace the now deprecated GET/PATCH /note-settings/note-sections-customization/{template_key}.

2026-04-24​

  • Introduce an optional provider specialty field. This field lets you set a user's medical specialty. If set, it will for now have impact, but in the near future, this piece of information will be leveraged to improve the dictation, transcription, and note generation.
    • Added optional specialty field to user management endpoints on the Server API (POST /users, PATCH /users/{id}, GET /users/{id}, GET /users).
    • On the other Server API, the same specialty field is available on the request bodies for /transcribe, /transcribe-async, /dictate, /dictate-async, /generate-note, and /generate-note-async.
    • On the User API, transcription, dictation, and note generation use the authenticated user's profile specialty instead (no specialty field on those request bodies).
  • Added optional encounter_date field to note generation endpoints (/generate-note and /generate-note-async) on both User and Server APIs. When provided, this date grounds relative temporal references (e.g., "last year") in the generated note. Defaults to today's date when omitted.

2026-03-16​

  • Add new bulk Dot Phrase management endpoints to the User API, each supporting up to 100 items per request:
  • Added created_at and updated_at timestamp fields to the dot phrase API responses.
  • /generate-patient-instructions now returns an error when no patient instructions can be generated. Previously, this endpoint would silently return a placeholder string ([Insert your patient instructions here]) when the note lacked actionable content. It now returns error code 83023 (NOTE_HAS_NO_PATIENT_INSTRUCTIONS_INFORMATION) with HTTP 400, allowing clients to handle this case explicitly (e.g., display their own placeholder).

2026-02-20​

  • Breaking change: JWT audience us-central1 and eu-west1 are now deprecated.

    • If you are using the us-central1 or eu-west1 regions in the URL of your JWT audiences; use the generic us and eu instead.

  • Diarization support.

    • Diarization support for transcription WebSocket sessions (/transcribe-ws). Starting with this API version, when speaker_type=UNSPECIFIED is set in the configuration of an audio stream, the Nabla transcription engine will perform speaker diarization. For each transcript item, the system will infer and return either speaker_type=DOCTOR or speaker_type=PATIENT.
    • Diarization support for transcription HTTP endpoints (/transcribe and transcribe-async). Starting with this API version, the returned

    If there is not enough context to reliably determine whether the speaker is a healthcare provider or a patient, the speaker_type will remain UNSPECIFIED, effectively behaving as if diarization were not enabled.

  • The locale and template fields are now required properties of the note request schemas for both Server and User APIs. These fields specify the locale and template used to generate the note and must be included in note data structures.

  • The note response schema for both Server and User APIs has been updated to include the locale and template fields, indicating the locale and template used for the generated note.

  • The note_locale and note_template fields have been removed from the request bodies of the following endpoints:

    • Server/User API: edit-note-with-instructions, generate-normalized-data, generate-patient-instructions These fields are replaced by locale and template in the note when making requests to these endpoints.

  • Changes to the response body of /generate-normalized-data:

    • The is_hcc and is_mcc fields within conditions have been moved into the nested coding object.
    • A new alternative_coding field has been added to conditions. This field is an array of ICD-10 codes that are closely related to the primary coding and represent plausible alternative coding for the condition.

2025-11-14​

  • New optional fields in object structured_context for all note generation endpoints:
    • patient_demographics: Demographic information about the patient.
    • draft_assessment_and_plan: Starting text for generating the Assessment & Plan section.
  • New optional structured_context field for the /edit-note-with-instructions endpoint.
  • Renaming for clarity and consistency:
    • Rename patient_context to unstructured_context in note generation and edit-note-with-instructions endpoints.
    • Rename previous_note to current_note in note generation endpoints.
    • Rename active_encounter_diagnoses_coding to encounter_diagnoses_coding in:
      • structured_context object in both 'generate-note' and 'edit-note-with-instructions' endpoints.
      • normalization (/generate-normalized-data).
  • Patched on January 7th: Fixed an issue where the current_note field was not taken into account during asynchronous note generation.

2025-08-05​

  • Add a new optional input field for all note generation endpoints:
    • structured_context: Structured context about the patient and the encounter, which will be taken into account when generating the note. Currently only includes:
      • active_encounter_diagnoses_coding: The ICD-10 coding for the patient's active encounter diagnoses. If the template used for the note generation contains an ASSESSMENT_AND_PLAN section, these diagnoses will be used for the "by problem" structuration of this section if they are discussed during the encounter.
  • Added a new active_encounter_diagnoses_coding parameter to /generate-normalized-data. If related diagnoses are found in the input note, the call will return in priority the codes passed.

2025-07-24​

  • Remove suggested_dot_phrases from response of /generate-note in Server API as it has always been null (Dot Phrases are only applicable for User API).
  • Add new CRUD endpoints to manage Dictation Text Replacements for a user.
  • Extend support for per-section customization options to note generation in French. This change applies to all prior versions too.
  • Add new optional text_field_context property to dictation WS config to improve dictation quality by providing initial text context.

2025-05-21​

  • Add new boolean fields is_hcc and is_mcc to the FHIR condition objects returned by the /generate-normalized-data endpoint.

2025-05-07​

  • Change the dictate WebSocket protocol. The WebSocket dictation protocol evolves to a faster and simpler model, where small units of text —such as individual words, short phrases, or punctuation marks— are sent once and don’t require any further updates. Clients are expected to append each received unit in order, without inserting spaces or applying any formatting. This change introduces the following breaking changes.
    • Evolution of configuration options in CONFIG frame.
      • The dictate_punctuation field has been renamed to punctuation_mode and is now an enum, with currently one possible value: EXPLICIT.
      • The speech_locale field has been renamed to dictate_locale and is reduced to the five following locales: ENGLISH_US, ENGLISH_UK, SPANISH_ES, SPANISH_MX and FRENCH_FR.
      • The enable_audio_chunk_ack option has been removed, since audio acknowledgement is now always activated.
    • Simplification and renaming of DICTATION_ITEM frame:
      • The DICTATION_ITEM frame type has been renamed to DICTATED_TEXT.
      • It has been simplified and now contains only a string text property.
      • Fields id, is_final, start_offset_ms and end_offset_ms properties have been removed since dictated texts are now necessarily final and are returned in chronological order.
  • Add new code, name and trace_id fields in WebSockets error frame JSON body. Also, the message field formatting changed (reminder: do not rely on this formatted string but only on the numeric code).

2025-05-05​

  • Add new corresponding_note_problem in the conditions returned by /generate-normalized-data endpoint. Use the include_corresponding_note_problems flag in the input body to enable this feature.

2025-04-03​

  • Enforce a limit of 700 on the length of the custom instructions per section for note generation endpoints
  • Add new User API endpoints to allow users to configure and persist their note generation preferences:
  • Change in User API /generate-note and /generate-note-async endpoints:
    • The following fields have been removed from the request body:
      • note_locale
      • note_template
      • note_sections_customization
    • These endpoints will now automatically apply the settings stored in the user’s note settings and per-section customization.
  • Change in User API /edit-note-with-instructions endpoint:
    • The optional field note_sections_customization has been removed.
    • The endpoint will now automatically apply the settings stored in the user’s per-section customization, for the locale and note template provided as endpoint parameters.

2025-02-25​

  • Bilingual support has been added to transcription endpoints and Websocket.
    • In request bodies and WS config frame, speech_locale property has been removed and replaced by new speech_locales property, an array that can contain one or two distinct speech locales.

2025-01-07​

2024-12-23​

  • Add a new audio chunk acknowledgement protocol, this change applies to all prior versions too.
  • Require endpoints paths to start with either "/v1/core/server" or "/v1/core/user"; paths like "copilot-api/server," "copilot/server," "copilot-api/user" or "copilot/user" are not supported anymore.

2024-11-08​

  • Fixed the case for server-to-client frames in Transcribe Websocket. It is now SCREAMING_SNAKE_CASE for all the frames, e.g. transcript_item becomes TRANSCRIPT_ITEM.

2024-10-01​

2024-04-22​

  • Add a new endpoint /generate_patient_instructions to generate a patient-friendly summary from an encounter.
  • Add a new endpoint /generate_normalized_data to generate normalized data from a note, available on both the Server API and the User API.
  • Add a new WebSocket endpoint /dictate for transcribing medical dictation.
  • Add a new endpoint /users/find_by_external_id/:external_id to fetch a user with a unique external ID from another system.
  • Add new endpoints to manage Dot Phrases for a user.
  • When set on a user, suggest matching Dot Phrases alongside the generated note in the 'digest' endpoint.
  • Add four new templates: GENERAL_MEDICINE_WCC, SOAP_WCC, GENERAL_MEDICINE_EMERGENCY, SOAP_EMERGENCY.
  • Add a title property in the note generated by server digest, server digest async, user digest, and user digest async endpoints. The title is a short description of the note content.
  • Add a patient_context property in digest and digest_async requests body.
  • Add a PATCH /users/:id endpoint.
  • Add external_id, metadata and created_at on the User object, the external ID and metadata are editable through the patch endpoint.
  • Remove upcoming_appointments_details from the output of the /generate_patient_instructions endpoint. Upcoming appointments details are now included inside instructions starting for this version.
  • Remove note_generation_mode request property from digest and digest async endpoints.
  • Remove upcoming_appointments_details from the output of the /generate_patient_instructions endpoint. Upcoming appointments details are now included inside instructions starting for this version.

2024-02-27​

  • Add new endpoints: activate-user and deactivate user.
  • Add new endpoints to fetch a single user or all users using filters and pagination.
  • Add split_by_problem to digest and listen-from-file endpoints.
  • Add optional client_request_id parameter to /digest and /digest_async endpoints.
  • Deprecate [es-ES, es-MX] as note generation languages in Core API for digest and listen endpoints.
  • Remove note.section.content: string[] field, usage to be replaced by text: string which is equivalent but formatted.
  • Remove the note generation feature from Listen Websocket API (recommend using the dedicated endpoint).

2023-06-26​

First version of the API.