Security & data residency

Your callers' voices stay in region.

A phone call carries some of the most personal data a business handles. Here is exactly where an AI phone call is processed — Switzerland, the EU or the US — and how every hop is protected.

Start free Sub-processors

● region-pinned
● tenant-isolated
● secrets encrypted
● tools screened

Data path

Where a call is processed, hop by hop

Every hop of a Swiss or EU call, from the phone network to the stored record.

01Phone networkCarrier number (Twilio or Telnyx) in the country of the line
02Media transportLiveKit, EU region for EU/CH calls
03Agent runtimeFrankfurt (EU) worker
04STT · LLM · TTSphi-cloud gateway, pinned to CH / EU / US
05Call recordSupabase, Switzerland (eu-central-2)

Data residency

Pick the region once. The call stays there.

Inference runs in the workspace’s region with no automatic cross-border fail-over. The real-time media layer that carries call audio runs in the EU (Frankfurt) for every call today; the database is in Zurich.

Swiss workspace

Transcribed, reasoned about and voiced by hosts in Switzerland where one exists.

EU workspace

Transcription, the model and the voice run in the EU.

US workspace

Inference runs in the US. Premium US-hosted voices are opt-in for US and rest-of-world agents only.

Building a GDPR-ready AI receptionist for Switzerland or the EU? Residency is set per workspace — nothing to configure per call.

Controls

How call data is protected

Controls as built into the product. This page describes architecture; it does not claim third-party certifications.

Region-pinned inference

Transcription, the model and the voice for a Swiss call run in Switzerland, for an EU call in the EU. A fail-over stays inside the region; a call degrades to “take a message” rather than leaving it.

Tenant isolation in the database

Every table carries a tenant id and default-deny row-level security. A workspace can only ever read its own agents, calls, contacts and bookings.

Encrypted tool credentials

MCP server secrets and calendar OAuth tokens are encrypted at rest (AES-GCM) and decrypted only for the call that needs them. They are never returned to a browser.

Screened tool results

Output from your MCP tools is screened for prompt injection before the model sees it. A tool that tries to steer the agent is dropped and the action is reported as not completed.

Audit trail

Every call keeps its transcript, the tool actions taken, the consent basis it was dialled on and its cost. Workspace changes are logged with the member who made them.

API keys, hashed

API keys are shown once and stored only as a SHA-256 hash with a display prefix. Revoke any key instantly from the dashboard.

For your review

What your DPO and your developers will ask for

A Data Processing Addendum, the sub-processor list, the calling rules we enforce and the API surface your team will integrate.

Processors

Sub-processor list

Who touches call data, for what, and in which region.

Learn more

Compliance

AI calling compliance rules

AI disclosure, consent, calling hours and opt-outs by country.

Learn more

Legal

Privacy policy

How personal data is handled, retained and deleted.

Learn more

Developers

REST API & MCP

Hashed API keys, per-workspace scope, and the MCP voice agent server.

Learn more

FAQ

Security questions

Enterprise needs a dedicated region, SSO or DPA review? See Enterprise pricing.

Do you train models on my calls?

No. Call audio and transcripts are used to run your agent and are never used to train models.

Can health data go through an agent?

Only on an agent configured for the PHI-grade inference path, which routes exclusively to hosts with the required agreements. General agents are not for health data.

Can I get a DPA?

Yes. Our Data Processing Addendum and sub-processor list cover call data; the sub-processors are listed on the compliance page.

MCPDial, resident by default.

Build an agent in the region your customers are in. Test it free in your browser before it touches a phone line.

Start free Sub-processors