Four AI features, and the edge of each one.
Nothing here summarises your workspace, drafts your announcements or decides when to send them, because none of that is built. What is built is narrow, and this page says where each piece stops.
Every AI call goes to Google's EU regional endpoints. Vertex AI is pinned to europe-west1, enforced in code.
Please note that the hand-hygiene protocol has been updated and it is important that everyone reads it and follows the new steps when they are on the ward.
The hand-hygiene protocol has been updated. Read it and follow the new steps on the ward.
Two actions, polish and shorten. Nothing replaces what you wrote until you say so.
What each one does, and where it stops.
A boundary is checkable and a superlative is not. Each of these is one you would otherwise find in the demo.
AI PolishCommand and HQ
Rewrites or shortens what you already wrote, with a before-and-after you approve.
Two actions only. No tone selector, no brand voice, no drafting from a prompt.
The AI CopilotCommand and HQ
Answers from your Knowledge Hub and names the articles it drew on.
Keyword overlap across your top five matches, not meaning. It does not watch the feed.
AI TranslateEvery plan
A post or an article in any of ten languages, on demand.
Google Cloud Translation, not a generative model. A translated article loses its headings.
Automatic moderationEvery plan
Post text and images are checked against Google safety classifiers.
The check runs after publishing, and it fails open if the classifier is unavailable.
All four draw on one monthly token allowance per workspace, and it blocks rather than bills.
Three people, and what each of them gets.
- Dr Tessa PalmerHead of clinical governance
Writes the announcement, shortens it with Polish, and reads the diff before letting it replace her text.
Comms Admin - Simon NaylorRegistered nurse
Asks the Copilot where the escalation threshold sits and gets the answer with the two articles it came from, so he can open them.
- Zoe MitchellPorter
Reads the same announcement in her own language, on the plan with the lowest price, because translation is not a Command feature.
Five things, and the plan each one needs.
- AI Polish, on your own words
Rewrite or shorten a passage you wrote, with a before-and-after you approve. It never replaces text without being told to.
Command and HQ - The Copilot, on your own articles
A question in, an answer out, and the articles it drew from named beside it so you can check the source rather than trust the answer.
Command and HQ - Ten languages, on demand
A post or an article translated for the person reading it, without a separate translated copy to keep in step with the original.
Every plan - Automatic moderation, on every plan
Post text and images are checked against Google's safety classifiers and anything flagged goes to an admin queue. It is not a paid add-on and it is not a tier: the smallest workspace on the cheapest plan gets the same check as the largest.
Every plan - Quizzes from an article
An admin turns a Knowledge Hub article or an announcement into a three-question multiple-choice quiz in one click, then edits it before publishing.
Command and HQ, admins only
What the AI does not do.
Every line here traces to something in the code, including the parts it would be easier to leave out.
- It will not write the post for you
AI Polish works on text you have already written. There is no prompt box that produces an announcement, no tone selector and no brand-voice model, and the system prompt tells it not to invent facts.
- The Copilot matches words, not meaning
It scores your articles by keyword overlap and reads the top five. Ask a question phrased differently from the article that answers it and it can miss. The retrieval is a word-overlap score, and this page will not dress it up as something cleverer.
- Moderation runs after publish
The check fires immediately after a post appears, not before, and if the classifier is unavailable the post is treated as approved. It is a queue for an admin to act on, not a gate that holds anything back.
- Translation is not a model
It is Google Cloud Translation. It will not preserve tone or fix an ambiguous sentence, and a translated article loses its headings, lists and embedded blocks.
- There is a monthly allowance
Each workspace gets a token allowance shared across all four features, resetting on the first of the month. At the cap the features stop until it resets. Nothing is billed by surprise because nothing is billed by usage at all.
- Two assistants, not one
The AI Copilot answers from your Knowledge Hub. The Help Assistant, on the in-app help centre, answers from Kayden Connect documentation instead. They share a plan gate and an allowance, and they read different things.
What the AI is pointed at.
- Knowledge HubThe Copilot answers from the Hub, so what is written there is the whole of what it can answer.
- AnnouncementsPolish lives in the composer, on the draft, in the minute before you publish it.
- Admin and governanceA flagged post lands in an admin queue, and what the admin then does is in the audit log.
Try the four things it does, and find the edges yourself.
Kayden Connect is invitation-only before launch. Tell us about your workforce and we will show you the demonstration workspace.
Request early access- Four featuresPolish, the Copilot, translation, moderation.
- In the EUGoogle's EU regional endpoints, enforced in code.
- Two on SparkTranslation and moderation are not a Command upsell.