Artificial Intelligence

Use of artificial intelligence at colib

Last updated: 2026-09-18

colib uses an artificial intelligence (AI) component for specific, limited features. This page explains in simple and clear terms what these features do, which technologies are used, and how your information is protected. It complements our privacy policy.

WHAT THE AI DOES

The platform uses AI for the following features, each of which is enabled by the clinic and triggered by the practitioner:

  • Transcription of telehealth sessions (speech-to-text): the audio of the session is converted into a written transcript. Information processed: the audio of the session and the resulting transcript.
  • Generation of clinical note drafts (generative AI): a draft note is written from the transcript of the session. Information processed: the transcript, the first names of the participants and, where the practitioner includes them, the client's previous session notes.
  • Generation of a note draft from an audio recording made during an in-person session, after the practitioner has attested that the client agreed to be recorded. Information processed: the recording, its transcript and the participants' first names.
  • For the most detailed note format on the Canadian model, the draft is produced in two steps (extraction of the session's elements, then writing); both steps use the same model in the same region.
  • Rewriting assistant for notes (generative AI): at the practitioner's request, the text of a note being written is reformulated, for example structured, corrected or shortened. Information processed: the text of that note.
  • Extraction of a form's structure from an existing document (generative AI): when a clinic imports one of its own questionnaires, the text of the document is analysed to rebuild its sections and questions in colib. Information processed: the text of the document supplied by the clinic, normally a blank template rather than a completed form.
  • Voice dictation (speech-to-text): a practitioner can dictate a note instead of typing it. Information processed: the short audio recording of the dictation, which is converted into text.

Each clinic can turn the rewriting assistant, form extraction and voice dictation on or off in its settings, and Colib can suspend any AI feature centrally. For the generative features, every clinic uses the same model, Claude Sonnet 4.6 (Anthropic), as described in the Technologies section. A client who has refused the use of AI on their record is excluded from every feature that processes their record. A patient can refuse from the Artificial intelligence tab of the patient portal, or by asking the clinic.

The platform does not use AI to make automated decisions about patients: no diagnosis, no triage and no treatment decision is made by the AI.

TECHNOLOGIES USED

  • Speech-to-text transcription of telehealth sessions is performed first by a model running within colib's own AWS infrastructure in Canada. If that transcription fails, and for voice dictation, the audio is transcribed by the Amazon Transcribe service in the AWS Canada (Montreal) region (ca-central-1); the audio file is deleted from our storage as soon as the transcription is returned, and it is not used to train AI models. That audio is placed in private storage, never publicly accessible.
  • Clinical note drafts, rewriting suggestions and form extractions are generated by the Claude Sonnet 4.6 model (Anthropic), accessed from Canada through the Amazon Bedrock managed service, for every clinic, including clinics located in Quebec. The request may transit through, and be processed in, the United States. The model name and the generation date are written at the end of every AI-generated note draft.

THE PRACTITIONER STAYS IN CONTROL

  • AI features are enabled by the clinic and triggered by the practitioner, who can choose not to use them at any time.
  • The generated note is a draft: the practitioner reviews it, can edit or discard it, and remains fully responsible for the content of the record.
  • Notes produced with AI assistance are marked as AI-generated in the record. The original AI draft is kept unchanged next to the note so that the practitioner's changes can be compared, and a practitioner can report an inadequate AI result directly from the note. Every decision on an AI draft (accepted and signed, edited, deleted, reported) is written to the clinic's audit log.
  • If you have questions about a note produced with AI assistance, or wish to have it reviewed, contact your clinic; colib assists the clinic in responding.

HOW YOUR INFORMATION IS PROTECTED

  • All information (audio, transcripts, notes, records) is stored at rest exclusively in Canada, in the AWS Canada (Montreal) region.
  • The AI processing itself runs in memory only, over encrypted (TLS) connections: the request may transit through, and be processed in, Canada or the United States, for every clinic, without anything being stored, logged or kept there. Your information itself is always stored in Canada.
  • In all cases, Amazon Bedrock does not store or log the content submitted to the model, does not use it to train AI models, and does not share it with the model provider: Anthropic never sees your information.
  • By default, the audio is deleted as soon as the transcription ends and the transcript once the note has been generated. A clinic may choose, in its settings, to keep the recording as an encrypted document in the record and to keep the transcript as a clinical working document; both then remain for as long as the clinic keeps the record. Technical processing records, which contain no transcript, are deleted after 60 days.
  • Each call to the AI is logged (feature, model, region, number of tokens, cost, duration) without the content submitted or produced.

LIMITATIONS

Like any AI system, transcription, note generation, rewriting and form extraction can contain errors or omissions. This is why the output is always presented as a draft to be verified, and never becomes part of the record without the practitioner's review.

CONSENT

Your clinic remains responsible for informing you and, where required, obtaining your consent before using these features as part of your care. For a telehealth session with a clinic located in Quebec, you are asked to tick a consent box before joining when an AI-assisted note is planned; you may refuse on that page, and withdraw your consent at any time from the Artificial intelligence tab of the client portal, which stops AI note generation on all your records. For an in-person recording, the practitioner attests before recording that you have agreed.

QUESTIONS

For any question about the use of AI on the platform, contact our Privacy Officer:

Thibault Bréboin
Privacy Officer, Colib Technology Inc.
thibault@colib.io

Previous versions