Every AI tool a control room needs, in the platform it already runs

Live, in the room, or afterwards. Live captions on the call; AI Scribe with the microphone and the computer's own audio as two separate speakers; any uploaded recording on request, with AI Scribe jobs first in the queue.

Silence stays silent. A speech gate and a hallucination filter make sure a pause, a cough or background music is never turned into words.

Flag a line as it is said. Click any line to flag it; the flag is shared with every console on the call and survives into the transcript.

Certified, subtitled, redactable. A transcript PDF with its MD5 and SHA-256; subtitles on playback, withheld from a redacted copy so muted words stay muted; click a line in the redaction studio to mute it. Part of Digital Evidence Management.

From any source. A recording's transcript, a chat or agent conversation, an AI Scribe session, or notes typed straight in - with an extra-context box for what the transcript cannot know.

Offences with the legislation. The classifier names each offence considered, the statute it falls under and your national recording code, and the principal offence - then the investigation plan builds on it. Catalogued to your jurisdiction.

Scored against your standard. Audits and assessments return a score, a grade and the reasoning beside each answer, so a supervisor reviews the judgement, not just the total.

Reviewed, certified, retained. Edited and approved in the browser, downloaded as .docx, with a certificate naming the requester, the template and the input, and a retention policy like any other file.

An operator is always in charge. An escalation flags the tile, a human message pauses the agent, and the agent's own summary is pinned at the top of the chat.

Pathways can hand over to an agent. A public reporting pathway routes to a person or to an agent, with a cap on how many it runs at once.

The conversation becomes the document. The same AI Documents work on an agent chat as on a recording, headed by the agent's summary.

Every reply is screened. Incoming messages are checked for prompt injection - encoded attempts included - before the agent reads them.

Written by you, not by us. Built on the manage page, enabled per organisation, picked per call. Change a question and the next call uses it.

Every tick is evidenced. "Heard at 01:12" jumps to the transcript line, so a supervisor can see why the box was ticked, not just that it was.

Shared across the room. Every console on the call sees the same ticks and the same summary, updated on every new utterance.

Nothing missing from the document. What the Companion could not hear is still unticked when you hang up - so the gap is asked about before the statement or the handover is drafted.

In the call - live captions on video consultations, each side translated for the other, with the language that was actually spoken shown on every line.

AI Translate - a two-way screen for the front desk, the emergency department or the station: hold to talk, tap for hands-free, or type, and spoken and typed lines land in one bilingual thread. Recorded if your organisation allows it.

Chat, invites and reminders - text messages are translated both ways, and the recipient's language is picked up from their own phone.

92 languages, detected as spoken - from Arabic and Hindi to Ukrainian and Vietnamese, the caller is never asked to pick from a menu.

Text too - chat messages, invites and appointment reminders go out in the recipient's language, picked up from their own phone.

On the Responder app - AI Translate and AI Scribe open from the map for any responder your organisation allows.

Switched on from the call. One control on the console; the caller's phone opens the meter with the instructions on screen, in their language.

Live on the console. Readings stream to the operator as they change, and the session's tile shows when vitals have been received.

An estimate, labelled as one. Blood pressure is shown as an estimate, so the clinician knows what they are looking at.

Kept with the session. The readings stay on the consultation record, beside the video, the transcript and the documents.

Hosted by GoodSAM. A speech model and a language model, run by GoodSAM on dedicated GPUs. The weights sit on our servers, not behind someone else's API.

Transcripts are data, not instructions. Every transcript is fenced as untrusted text before a model reads it, and messages from the public are screened for prompt injection - encoded attempts included.

Reviewed before it leaves. Every document is a draft until a person approves it; creativity is switched off, so the same transcript gives the same draft, with the gaps marked as not stated.

Governed like evidence. A separate permission to generate, templates enabled per team, a retention policy, deletion with a reason, an audit event for every request, and a usage report per organisation.

Translation, precisely scoped. Typed and captioned text is translated by an enterprise machine-translation service and nothing more. Speech and drafting never leave GoodSAM.