No. SleekAI is configured with explicit guardrails to decline every diagnostic, treatment, dosage, or result-interpretation question. The instruction includes a hard refusal: never diagnose, never recommend specific treatments. If a patient asks 'is my cholesterol number bad', the bot routes them to a conversation with their physician rather than offering a guess. Audit-test the bot quarterly with edge-case prompts (lab numbers, medication dosages, symptom panels) to confirm the refusals hold.

HIPAA compliance depends on your full stack and BAAs, not on a single plugin. SleekAI is a WordPress plugin that uses an API key you provide; for US practices handling PHI, you need a HIPAA-eligible OpenAI configuration (or equivalent with your chosen vendor), a BAA in place with that vendor, encryption at rest on the WordPress database, and access controls on who can read conversation logs. The plugin gives you the building blocks, the compliance posture is a stack-wide responsibility.

SleekAI doesn't write directly into your EHR. It links out to your patient-facing booking flow (Athena, Epic MyChart, NextGen, Doctolib, Cliniko, Calendly) so the bot is the front door and your EHR remains the source of truth. Most practices prefer this split: it avoids the dual-write reconciliation risk, and keeps the bot's failure mode harmless (a missed booking, not a corrupted patient record).

The bot is configured to recognise urgency keywords (chest pain, shortness of breath, severe headache, signs of stroke, hypoglycemia symptoms, severe abdominal pain) and immediately direct the patient to call 911 or the local emergency number, rather than wait for chat triage. This is non-negotiable in the system prompt and tested as part of the standard configuration. A visible disclaimer in the widget header reinforces the channel's scope.

It can explain your refill process and point patients to the right form, portal, or pharmacy line. It never authorises a refill or estimates whether one is appropriate. For controlled substances the bot is configured to defer to in-person review per your state's controlled-substance regulations. The pattern is the same as for prescriptions in general: the bot is signage, not authority.

Yes. If your scheduler offers a telemedicine appointment type, the bot can route patients to it just like an in-person slot, with a note about the video platform you use (Doxy, Zoom for Healthcare, MyChart Video). It can also explain which states or licensing jurisdictions you can practice telemedicine in, since US internal medicine practices are often state-licensed and patients sometimes travel across state lines.

Yes, in your WordPress database. Retention is configurable. For PHI we recommend short retention windows (days, not years), encryption at rest on the WordPress database, role-based access on who can read logs, and BAAs across the stack including your hosting provider and the API vendor. Document the chat channel in your Notice of Privacy Practices and your records-retention policy so the bot fits cleanly inside your existing compliance framework.

Yes. Multibot lets you scope assistants per provider or per location if you have a multi-site internal medicine group. A single shared bot with physician-aware routing also works for smaller practices. The system prompt can include each physician's special interests, languages spoken, and weekly schedule patterns so the bot routes intake intelligently rather than dumping every new patient into a generic queue.