The two modes
MediSina supports cloud AI (OpenAI, Anthropic, others) for clinics with reliable internet, and local LLM (Ollama, llama.cpp) for clinics where data must stay on-premise or where bandwidth is unreliable. Each clinic chooses at install time and can switch later.
Cloud mode
- Lower setup cost — no GPU required on-site.
- Higher quality (frontier models).
- Tradeoff: requires internet and a per-call cost.
Local mode
- Data never leaves the hospital network.
- Works offline.
- Tradeoff: needs a modest GPU server; quality is good but not frontier.
What the AI does
- Triage assistance. Surfaces concerning findings in vitals and free-text notes for a clinician to confirm. Never makes the diagnosis.
- Clinical-note summarization. Compresses long admission notes into a structured handoff summary, with the original preserved.
- Decision-support hints. Flags drug interactions, missing tests, and standards-of-care gaps. Always optional, never blocking.
- Multilingual transcription. Doctor speaks in any language; structured note appears in the patient record.
What the AI does not do
- No autonomous prescriptions, orders, or referrals. Every action is human-confirmed.
- No PII leaves the site unless cloud mode is explicitly enabled.
- No audit gaps. Every AI call is logged: who asked, what was sent, what came back, what was accepted.
Agentic workflows (preview)
In limited contexts, MediSina can run multi-step workflows on the clinician’s behalf — e.g., “if a critical lab value comes back, draft the patient notification and route to the on-call doctor for approval.” All steps are visible, reversible, and audited.
Want to see it?
Schedule a demo — we’ll walk you through both modes with anonymized examples.