PIPEDA-aligned handling
We collect the minimum needed to answer, book, and dispatch. Call recordings and transcripts belong to you. We do not sell them, and we do not use them to train foundation models.
Security & compliance
Your calls are customer data. We treat them that way.
Right On Call is a boutique receptionist, not a consumer app that trains on whatever you type. Recordings exist so we can dispatch, book, and debug, not so we can sell a dataset.
Callers are told
Ontario is one-party consent for most business calls. We still put a short recording disclosure on the line, and the agent will identify as automated if asked.
Guardrails in code, not just a prompt
The receptionist cannot invent availability, cannot diagnose, and cannot quote a repair price. Gas, smoke, or spark reports take a bounded safety script and a human callback.
You can turn it off
A dashboard kill switch stops the agent from answering. Tokens for third-party tools are encrypted at rest. Database access is row-level restricted per client.
Who else touches a call
Voice AI cannot live entirely inside one Canadian closet. Telephony, speech, and models run with named vendors. We keep that list short, and we do not claim a data-residency story we cannot keep.
| Vendor | Role | Region |
|---|---|---|
| Twilio | Telephony and SMS | US / CA numbers |
| Vapi | Live voice pipeline | US |
| Anthropic | Conversation and summaries | US |
| Supabase | Call log and dashboard data | configurable |
| Render | Application hosting | US / EU |
What the default product is, and is not
- Baseline handling is PIPEDA-aligned: minimum collection, customer-owned recordings, deletion on request. We do not sell call data or use it to train foundation models.
- We are not SOC 2 certified on the default stack. If procurement needs that, it is a scoped build with real audit cost. Say so up front and we will price it.
- We are not a HIPAA product out of the box. Clinics stays a waitlist until someone pays for a covered-entity stack (BAA, tighter retention, extra vendor review). That work is available; it is not included in the $150 plan.
- Foundation-model vendors process audio to produce the conversation. We do not then reuse that audio to train those models.
- Retention follows the client agreement. Default is to keep recordings long enough to audit a dispatch, then drop them.
For the longer, plain-language version of this page, read PIPEDA, Recordings, and US Subprocessors.