AI Across ArkCloud Health

Intelligence embedded throughout the healthcare journey.

Here is exactly where that is true today, and where it is not. Choose an application to see what each AI feature receives, what it produces, what a clinician reviews and whether you can use it now.

36 available

AI features customers can use today

Rehab 10, RCM 9, Access 4, Care 2, Insights 2, Workforce 1, Across the platform 8. Every one produces a draft or a read-only explanation a person reviews; the admin agents stay off until an administrator enables them.

8 in development

Built and tested, not yet switched on

Review this note (encounter review); Home exercise reports, summarised for the therapist; Patient intake answers, summarised for the reviewing clinician; Outside records, summarised with their sources; Decision support over published guidance; Visit recording and voice intake (speech-to-text); Call notes from recorded phone calls (Voice Intelligence); Next steps after a call (Voice Intelligence)

Private network

Where the model runs

Gemma 3 (4B), on a separate on-premises model server that ArkCloud operates on its own private network, reached over an encrypted (TLS) connection. Not a third-party AI cloud.

Application explorer

AI, application by application.

Each application lists its AI features, the automation that is sometimes mistaken for AI, and the things it does not do. “Rules, not AI” marks 9 live features that run on fixed, written rules - useful and dependable, and not machine learning.

The Clinical Brain answers questions and prepares a handoff summary; Draft SOAP note, progress-report and discharge-summary narratives, outside-record suggestions, measure explanations, patient message replies and missed-visit outreach are AI drafts a clinician reviews. Goals and plans of care are never written by AI.

ArkCloud Rehab module guide

Clinical Brain: questions about an episode

LiveUses AI

A therapist asks a question about the episode or assessment in front of them - the latest score, what is still outstanding, what the documentation status is - and gets an answer built only from that patient's governed record.

What problem it solves: Answering "where are we with this patient?" usually means opening several screens. The Clinical Brain gathers the answer in one place without the therapist leaving the chart.
How AI is used: Rules, not the model, decide what is being asked: the question is matched to one of the approved question types, and a question that matches none is declined. The software then fetches the facts that question type allows. The model only turns those facts into a readable sentence, and a rule-based check compares the answer against the facts and withholds it if it says anything they do not support. It runs on Gemma 3 (4B), on a separate on-premises model server that ArkCloud operates on its own private network, reached over an encrypted (TLS) connection.
What it receives: The therapist's question
Structured facts for that episode chosen by rules: scores, assessment and documentation status, outstanding items
Recent turns of the same conversation
What it produces: A short written answer, shown in the panel
Suggested next steps that link to the screen where the work is done
A clear refusal when the question is outside the approved types or the answer fails the check
What a professional reviews or approves: Read-only. It cannot change the record, sign anything or place an order; acting on an answer is always the therapist's own step in the normal screens.
Availability: Live in ArkCloud Rehab on the episode and assessment pages, for staff permitted to view rehab patients.

How to use it, what it never does and who reviews it

Try itIllustration with synthetic data. No model is called and no real patient is shown.

Episode (synthetic)

Patient: Sample Patient A (synthetic)
Episode: Left knee, post-operative rehab
Visit: 6 of 12 authorised
Clinician: Demo therapist

Ask about this episode

Choose a question to see how the answer is built, step by step.

How ArkCloud AI works

Assistance, not autonomy.

AI in ArkCloud suggests; people decide. Every AI feature follows the same path, from the data it is allowed to see to the record of what happened.

01
Healthcare data
Only the signed-in user's organisation is read, after a role check. Rules decide which facts a feature may send - an episode's scores, a size-capped set of visit notes, one account's minimised summary, a screen's totals - and names, birth dates and member IDs are left out wherever the text does not need them.
02
AI processing
Before any patient information is read, the destination is checked. On the hosted service it is a separate on-premises model server that ArkCloud operates on its own private network, reached over an encrypted (TLS) connection. An external endpoint is refused.
03
Suggested output
The model produces an answer, an explanation or a draft - never a change to a clinical record, a submission or a payment. Every number, date, code and quote is checked against the facts it was given; text that says more is withheld, and the facts or a rule-built version are shown instead.
04
Human review
Every AI panel is labelled AI before it is opened. Answers are advisory; drafts are edited and sent by a person; agent proposals wait for approval. A clinician inserts the draft sections they accept, and signing an AI-assisted note requires attesting to that review.
05
Approved workflow
The work continues through the normal screens, under the user's own permissions: the note is signed, the call is made, the charge is posted - by a person.
06
Audit trail
AI queries and refusals are recorded with who, which record, which model and which prompt version - not the text. Draft generated, draft accepted and AI-assisted note signed are chained into the tamper-evident audit log.

Safeguards built into the software

Role-based access: Each AI feature checks the same permission as the screen it sits on. Only supervisors and administrators run agents, only an administrator changes an agent's approval rules, and administrators can switch the RCM assistant, the RCM drafts and the in-screen assists off for their organisation.
Tenant isolation: Every AI request reads inside the signed-in user's organisation, and record ids are re-checked against it. One organisation's data cannot reach another's prompt.
No third-party AI cloud: It is not a third-party AI cloud. The software sends patient information only to a model on the same machine or to a private-network host an operator has explicitly acknowledged, over TLS; any other endpoint is refused, and no patient information is sent to an external AI provider.
Rules before the model: Rules - not the model - decide what is being asked and which facts are relevant. A question outside the approved types is declined instead of guessed at.
Output review: Answers that fail the fact check are withheld. Drafts are never written into the record by the model, and an AI-assisted note cannot be signed without attestation. Each AI feature has an hourly limit per person and per organisation.
Auditability: Every AI query is an audit event. Agent runs and decisions, and documentation decisions involving AI, are hash-chained so later alteration is detectable.

What AI in ArkCloud never does

Diagnose, order, prescribe, or write a treatment plan, care plan, pathway or goal
Suggest a diagnosis or procedure code, or score a patient's risk
Sign, submit, post or send anything
Change a clinical record, or act beyond the approval rules an administrator set
Answer from data outside the user's organisation or permissions

Limits we state plainly

The hosted service runs one small model (Gemma 3 (4B)) on one model server, shared by every AI feature, which bounds answer quality and how many requests run at once. ArkCloud's AI features have not been clinically validated, and this page makes no accuracy, safety or regulatory claim for them. Clinical decisions remain the clinician's.