What can the AI assistant help with?
Chat with an AI agent inside diagnos — from an exam you're writing, or from the workspace in general — or reach it from an AI tool you already use, like ChatGPT or Claude. It's not part of the encrypted vault, so never type in a patient's name, email, or phone number; treat it like any AI, since it can make mistakes.
- Updated on
- 2 min read
On this page
The AI assistant is a chat you can open from inside diagnos, or reach from an AI tool you already use outside it. Inside the app it's tied to whatever you're doing — an exam you're writing, or the workspace in general — and each ongoing conversation has its own name, made of a colour and a number (like Red 1), so you can tell several agents apart.
You can also reach an agent without opening diagnos at all: connect an AI tool you already use — ChatGPT, Claude, or anything that speaks MCP — and send a message, a note, or a file addressed to one of your agents. It shows up back inside diagnos next time you open it; there's no live back-and-forth through that connection, only sending.
The reassuring part: what an outside connection can and can't do#
An AI tool connected this way can only write: send a message or a file to one of your agents. It can also read diagnos's public help articles, the same ones you're reading right now. It has no tool to read your conversations, your patients, or your exams — that capability simply doesn't exist on that connection, by design.
Before you type anything#
Every AI conversation — inside diagnos or through an outside connection — carries the same rule: never share sensitive data such as names, emails, or phone numbers. It's an AI, and it can make mistakes.
This also means AI conversations work differently from the rest of diagnos: patients, exams, files, and templates are end-to-end encrypted, so not even the server can read them. AI conversations are not — the model needs to read what you actually wrote in order to answer, so treat this chat the way you'd treat any other third-party AI tool, not like the encrypted vault.
One model inside an exam, "diagnos" everywhere else#
Inside an exam, the assistant always answers with the same model, so clinical answers stay consistent. At the workspace level, it's presented to you simply as diagnos — you don't need to pick or know which model is behind it.