Skip to content
Help centre

AI governance

How the on-device AI behaves: the model registry, the guardrail policy and its violation log, the thresholds behind caseload insights, and the drafting and scheduling settings.

In SideClick: Admin → AI governance

AI governance is where the practice decides how the on-device AI behaves. Nothing here sends anything anywhere: the models run on this computer.

Model registry lists every AI model version with when it was deployed, its metrics and its status: pending, under review, approved, rejected or deprecated. The buttons on a row move it to its next status — Review, approved, rejected or deprecated. A rejected model can never be re-approved, and rejected and deprecated models refuse to load.

Voice dictation model installs, or re-activates, the on-device model behind dictation. Dictation stays hidden until one is installed.

Guardrail policy sets what happens when a generation trips a check — Prompt injection, Incomplete fields, Bias markers or Hallucinated terms. Block refuses the generation; Warn and Log let it through. Every violation is written to the audit log whichever you choose, and the Guardrail violation log at the bottom lists them with the action taken.

Outcome-insight thresholds tune how sensitively the caseload scan flags goals at risk, plateaus, stalled plans and overdue outcome measures. The values are kept within safe ranges and apply to the next scan; no model is involved in that scan.

AI drafting & extraction chooses which of the patient's own signals ground a drafted note, the note's character budget, and the confidence below which extraction proposals stay hidden. The patient's consent is still the gate for the previous-note excerpt.

Scheduling decision-support switches on Waitlist standby suggestions and Adaptive reminder timing, both model-free.

Each section has its own Save button. A saved change applies to the next generation, scan or booking, never to one already made.

Related

  • The AI featuresDrafts, summaries, insights and dictation run on the computer itself — decision support, never a diagnosis — and only with the patient's consent.
  • Caseload insightsThe signals SideClick raises about your caseload — goals at risk, plateaus, stalled plans, overdue measures and reports — and the optional one-sentence summary.
  • Model setupDownload and activate the on-device AI model behind note drafting, or check which model is active, on your computer.
  • AdminThe practice's figures for a period — clinical, financial, outcomes and practice — each tile explained when you point at it, and exportable as a PDF.
  • WaitlistThe patients waiting for an appointment, and where a slot freed by a cancellation is confirmed or released.

Still stuck? support@sideclick.io reaches the people who build the product.

Try SideClick free for 30 days

Full capability, no card required. Your data is sealed on our server and we cannot read it.