Security and data protection at EasyVoice
What happens to the calls your organization receives: where audio, transcripts and summaries are processed, who is involved and how we protect them. For your data protection impact assessment, your records of processing and your IT review.
Last updated:
At a glance
Call processing through EU endpoints
Telephony, speech recognition, AI model and speech output run through endpoints in the EU. The only exception is sending email (see below).
Stored in Germany
The servers and the database with calls, transcripts and summaries run at Hetzner in Germany, in an ISO 27001 certified data center.
Processing under Art. 28 GDPR
Your organization remains the controller; we process on its behalf. Our data processing agreement applies with the contract; a signed copy is available on request.
No training on your calls
EasyVoice does not train AI models on the recordings, transcripts or summaries of your calls.
Consent step at the start
Every call starts with an automated notice and consent step. It is built into the call flow, not left to an instruction to the AI model.
Your data belongs to your organization
Calls can be deleted at any time and are deleted after 90 days by default. After the contract ends you can export your data for three months.
Data residency: EU endpoints
A call reaches EasyVoice through the phone network. It is then processed in real time: speech recognition, AI model, speech output. Afterwards, the recording, transcript and summary are stored for your team. Every step is encrypted and runs through endpoints in the EU:
- EU endpoints for all call processing. Telephony, call orchestration, speech recognition, AI model and speech output are reached only through endpoints in the EU. Several of the providers are headquartered in the USA; we have data processing agreements with all of them.
- Agreed together. Which speech recognition and speech output your assistant uses is agreed with you during setup. All eligible providers are listed below.
- No retention during AI inference. With LiveKit Inference, audio and model inputs and outputs are neither stored nor used for training (zero data retention).
- On-premise available. For organizations with special requirements, EasyVoice can also run on their own infrastructure.
Subprocessors
These companies process EasyVoice data on behalf of emergo technologies GmbH. We have data processing agreements with all of them. The list matches Annex 2 of our data processing agreement (section "KI-Telefonassistent"); we notify you 30 days in advance of any new or replaced subprocessor (DPA section 8.2). For the website, see our privacy policy.
Call processing: telephony, speech and AI EU endpoints
These services process the call itself: audio, transcript and the assistant's replies. All of them are reached through endpoints in the EU. Services marked "Per assistant" are used only if your assistant is configured that way.
| Provider | Purpose | Processing | Company seat | Data | Use |
|---|---|---|---|---|---|
| Hetzner Online GmbH | Data center: servers, database with calls, transcripts and summaries, sign-in service | Germany | Germany | All application data | Always |
| LiveKit, Inc. | Real-time call orchestration (LiveKit Cloud), transfer of the recording to storage, AI inference via LiveKit Inference: language model (Gemma 4) and end-of-turn detection | EU endpoints | USA | Audio, transcripts, phone number | Always |
| Twilio Inc. | Telephony: phone numbers, connection to the phone network (SIP), SMS | Ireland (Twilio region IE1) | USA | Phone number, call metadata, audio, SMS content | Always |
| Microsoft (Azure) | Speech recognition (Azure Speech), post-call analysis: summary, categorization, data extraction (Azure OpenAI), storage of call recordings (Azure Storage) | EU (Germany, Sweden) | USA (DPF) | Audio, recordings, transcripts | Always |
| Cartesia AI, Inc. | Speech output (text-to-speech); speech recognition via LiveKit Inference | EU endpoints | USA | Assistant reply text, audio | Per assistant |
| Speechmatics Ltd | Speech recognition via LiveKit Inference | EU endpoints | United Kingdom | Audio | Per assistant |
| Gladia SAS | Speech recognition | EU | France | Audio | Per assistant |
Notifications and quality assurance No audio
These services never receive audio. Note on email: it is currently sent through servers in the USA. Call summaries are emailed only if your organization turns notifications on.
| Provider | Purpose | Processing | Company seat | Data | Use |
|---|---|---|---|---|---|
| Twilio Inc. (SendGrid) | Sending email: invitations and password links; call summaries, if enabled | USA | USA | Name, email address; with notifications, the call summary and captured details | Always |
| Langfuse GmbH | Quality and error analysis of call processing (tracing), Langfuse Cloud EU | EU | Germany | Transcript excerpts, AI model inputs and outputs, technical metrics | Always |
"Processing" is where the data is processed; "Company seat" is where the company is registered. DPF: certified under the EU-US Data Privacy Framework. "Per assistant": used only if your assistant uses this service.
Technical and organizational measures
A summary of the key measures. The full technical and organizational measures under Art. 32 GDPR are in Annex 1 of the data processing agreement.
- Encrypted transport. All connections between users, our servers and the services we use are encrypted (TLS).
- Tenant separation. Each organization's data is logically separated at application level. Every access happens only in the context of the respective customer account.
- Sign-in and roles. Central sign-in for the dashboard, role-based access rights. New users receive a personal invitation.
- Access only when needed. Staff get access only to the systems their task requires (need-to-know), via single sign-on and hardware security keys (FIDO2). All work devices are encrypted and centrally managed.
- Recovery. Snapshots of the production systems are taken so they can be restored in case of failure.
- Annual review. The measures are evaluated every year and adapted to the state of the art.
AI on the phone
- Notice and consent at the start. Every call starts with an automated step that informs callers and asks for their consent. It is built into the call flow and does not depend on the AI model following an instruction.
- Your rules, your workflows. Your organization decides what the assistant says and is allowed to do. It acts in your systems only through the interfaces agreed upfront.
- Handover to people. Anything the assistant should not resolve itself goes to the right person on your team, with a summary and the captured details.
- Traceable. Every call is filed in the dashboard with recording and transcript, so your team can check results and listen to the original if speech recognition got something wrong.
- Limits stated openly. AI output can be incomplete or wrong. EasyVoice handles organization and communication on the phone and gives no medical advice, diagnoses or treatment recommendations.
Your data, your decision
- GDPR roles. Your organization is the controller and emergo technologies the processor (Terms, section 11.1). Callers contact your organization about their data; we support you with that.
- Automatic deletion. Calls are deleted after 90 days by default; your organization can change the period in the dashboard and archive individual calls. A call deleted in the dashboard is deleted immediately.
- Export and deletion when the contract ends. After the contract ends you can export your data for three months. After that we delete it completely (Terms, section 7.4).
- Turn email notifications on deliberately. Summaries by email are optional and enabled per assistant. If you prefer not to send call content by email, use the dashboard only.
Compliance overview
We distinguish clearly between what we do ourselves and what comes from our service providers.
| Topic | Status |
|---|---|
| GDPR and Austrian Data Protection Act | Data processing agreement under Art. 28 GDPR as part of the terms, applying with the contract. Subprocessors under their own agreements (see above). |
| Health data (Art. 9 GDPR) | Call processing through EU endpoints, storage in Germany and the EU. Your organization, as controller, ensures the legal basis for the processing (DPA section 5.2). |
| ISO 27001 | Certification held by our data center operator, Hetzner Online GmbH. emergo technologies GmbH itself is not ISO 27001 certified. |
| Medical Device Regulation (MDR) | Not a medical device; EasyVoice handles organization and communication on the phone, no diagnosis or treatment recommendation. |
Documents
Public
On request
- A copy of the DPA signed by us
- Support for your data protection impact assessment
Found a security issue?
Please report possible vulnerabilities confidentially to office@emergo.dev with the subject "Security". Describe how to reproduce the problem, do not access other people's data, and give us time to fix it before publishing details. Our security contact is also listed in our security.txt.
Report a vulnerability →