You talk. It drafts. You sign.

Push-to-talk dictation turns spoken words into structured notes, filled forms and charted vitals, with a clinician approving every word before it reaches the record. There is no always-on microphone and no recording of conversations. The clinician presses to speak, the AI structures what they said, and nothing is saved until they apply it themselves.

See it in motion.

A short, narrated explainer. Captions included.

Three ways to write without typing.

Notes by voice: Dictate a progress note in your own words. The AI structures it into a clear SOAP note, subjective, objective, assessment, plan, shown as a draft for the clinician to edit and apply.
Forms by voice: Dictate against the eForms your own team built in the form builder. Each phrase is mapped to the right field, and the form is only persisted when the clinician confirms and locks it.
Vitals by voice: Read observations out as you take them, blood pressure, pulse, temperature, oxygen, and the values land in the right fields on the observation chart, staged for explicit submit.
Nothing to install, on by default: Dictation uses the browser's own speech recognition, open standards, no proprietary microphone hardware, no extra app, no per-user licences. It ships enabled; works in Chromium-family browsers today.

Not an ambient scribe, and that's the point.

Ambient tools record the whole room and guess who said what. HealthOS voice dictation is built to a stricter spec, because trust in the record matters more than novelty.

Single speaker, push to talk: The clinician dictates into their own microphone, starting and stopping on explicit action. Conversations with patients and families are never captured, and there is no always-on listening mode.
The clinician signs everything: Notes stay drafts until applied. Forms require Confirm & Lock. Vitals are staged for explicit submit. The AI never writes to the record on its own.
Minimal data to the model: Each mapping call receives only the new dictation chunk and the form schema, not the accumulated transcript, and not the values already in the form.
Your schemas, not an ontology: Form filling maps against eForms your team authored, structurally equivalent to FHIR Questionnaires, so the AI completes your form, rather than inventing clinical structure.
You choose the model: Scribing runs through your organisation's AI model routing: pick which model structures notes and which fills forms, independently, from the admin portal. Models are hosted in your own tenancy, Claude, OpenAI's GPT models, or low-cost open models like Llama, Mistral or Qwen, and high-volume dictation can run on a cheap model while safety-critical tasks keep a premium one.
Every call audited: Each scribing request is logged through the same AI audit trail as the rest of the platform, which model, which task, when, so AI use in clinical documentation is reviewable, not invisible.

Dictated once, part of the whole record

A dictated note is not a transcript in a side system. It lands in the Canonical Care Record alongside observations, medications and care plans, so it feeds handover, reporting and the family portal like any other entry.

Notes join the person's longitudinal record
Vitals feed the deterioration engine like typed entries
Forms write structured data, not free-text blobs
Available wherever the record is, desktop or tablet

Solutions

Modules

Features

More

You talk. It drafts. You sign.

See it in motion.

Three ways to write without typing.

Not an ambient scribe, and that's the point.

Dictated once, part of the whole record

See voice dictation in action.