Safety and quality

The important limits aren't instructions the AI might ignore. They run as code before and after every reply.

#Before a reply

• Emergencies go to people. Every incoming message is checked for a possible emergency before any AI reply. The person is given the emergency number first (112 in India, 911 in the US), and your emergency contact is alerted straight away. Possible self-harm gets the crisis line (Tele-MANAS 14416 in India, 988 in the US).
• Private until confirmed. Personal details stay out of the agent's view until the person proves who they are (Identity and privacy).
• Staff own their conversations. While your team has a conversation, the AI doesn't reply in it.

#After a reply, before it's sent

Each reply is checked before it goes out. When a check fails, the agent is told what was wrong and writes the reply again; it is never sent as it was.

• No medical decisions. Clinic agents never diagnose, suggest medicines or explain results. They book a visit or bring in your team instead.
• No made-up facts. A booking is only confirmed after it's actually booked, and only at a time that was really free. The agent can't claim it checked someone's identity when it didn't.
• Required details are said. If a service has preparation instructions (for example, fasting before a blood test), the confirmation must include them.
• No promises the system can't keep. If the agent tells someone a person will get back to them, a task for your team must exist.

#Who the agents say they are

On calls, the phone assistant says it's an AI assistant and that the call may be recorded and transcribed. Callers can interrupt it at any point, or ask for a person.

#How conversations are tested

Every change to the agents is run against a set of scripted patient conversations: a new patient booking a blood test, a patient switching doctors, someone asking for a time that's just been taken, a question the agent can't answer, a request for a human. Each conversation is graded on what actually happened (was the visit booked at the right time, with the right doctor, were the preparation instructions given, did the team get a task when one was promised), not only on how the replies read. A change ships when every scenario passes.

#Where your data lives

Your business's conversations, records and settings are only visible to your account and to API keys you create (see Webhooks and API for the privacy levels a key can have). Credentials you enter for channels and integrations are stored encrypted and never shown again in full.