What we use AI for
Pivotical uses AI in four places. Vision models read photographs of medication labels, prescriptions, and lab reports so you don’t have to type them in. Language models generate the daily brief, your 30-day plan, and answers to questions you ask the assistant. A search index lets the assistant cite the right document when you ask about Pivotical itself. Small classifier models route your questions to the right specialist agent.
Each of these is an assistant. None is the doctor.
Which models we run
Today we use Mistral models for most language and vision tasks — Pixtral-large for image understanding, Mistral-large for chat and protocol generation. We may use additional providers in the future. The current sub-processor list lives in the Sub-processors document and is updated within 30 days of any change.
We never use your health data to train public models.
What AI cannot do here
Pivotical’s AI does not diagnose conditions, prescribe medications, or take any action that requires clinical authority. It can suggest a possible pattern, but the conclusion is always yours and your clinician’s.
When the assistant is asked something outside that boundary — for example, “Is this safe to take?” — it answers with what it knows and refers you to a qualified human.
Where it can be wrong
AI models can be confidently wrong. They can misread a label, misinterpret a symptom log, or generate a plausible-sounding plan that doesn’t fit your case. We engineer guardrails — cross-checks, confidence thresholds, human review for high-impact recommendations — but we cannot eliminate error.
If something Pivotical says doesn’t feel right, trust the feeling and check with a clinician.
Human in the loop
For anything material — a flagged drug interaction, a recommended phase change, a flagged anomaly — we surface the AI’s reasoning and ask you to confirm before acting. You can always read why the agent suggested what it did. You can always undo.
If you tell Pivotical you disagree, it learns that for your account. It does not propagate your data to other users or to model training.
What we log
When an agent makes a decision — rewrites your plan, flags a possible interaction, surfaces a brief — we log what data it read, what it produced, and which model version it called. You can read this log in the app under your settings.
We keep these logs for the lifetime of your account so you and your clinician can audit any recommendation.
Your choice
If you don’t want a particular feature to use AI — for example, you’d rather type prescriptions than photograph them — you can turn that feature off in settings. Most agents have a manual mode.
Some core features depend on AI; we mark those clearly so you can choose whether Pivotical is the right product for you.
Reporting issues
If an AI response was harmful, dangerously wrong, or felt like it crossed the medical line, mail care@pivotical.in. We read every report.