AI assistant (optional)
The assistant answers questions about one patient's record, using only the clinic's own data and a language model that runs on the clinic's own server. It stays off unless the clinic turns it on.
What it does
- Off by default. The clinic decides whether to turn it on.
- Staff ask about one patient at a time, and answers come only from that patient's record.
- Runs on the clinic's own server through a local OpenAI-compatible endpoint. Ollama is the default, and llama.cpp, vLLM and LM Studio also work.
- Nothing is sent to a third-party AI service.
- bge-m3 embeddings and a Gemma chat model, with vectors stored in PostgreSQL with pgvector and the indexed text encrypted.
- A per-patient opt-out. When a patient opts out, their indexed data is deleted.
- Chat history is deleted after 30 days, and questions are logged as hashes only.
- Meant for documentation and finding information in the record. It does not make diagnoses.
How it fits
The assistant reads what is already written in visits and on the patient record. It helps staff find that information quickly, and the clinical decisions stay with them.
Related
See eKlinika with demo data
Tell us about your clinic: country, specialty and team size. We reply by email to arrange a walkthrough.