Sub-processors
Last updated: 6 June 2026
About this page
dareena.ai engages a small number of third-party service providers — called sub-processors — to deliver the Service. This page lists each current sub-processor, where they are located, what they are used for, and the categories of data they process.
Caller data at rest — call recordings, transcripts, invoices and customer records — is stored in New Zealand. During live calls, some data is processed in real time by the AI providers listed below; how each provider handles that data is described in the table and summarised under AI processing and training.
Current sub-processors
| Provider | Location | Purpose | Data categories |
|---|---|---|---|
| Voyager Internet Ltd | Auckland, New Zealand | PSTN carrier (call termination, New Zealand DID ranges) and cloud server hosting (production database, application servers, SIP infrastructure) | Call signalling metadata; production database contents (tenant configuration, call records, transcripts, customer service data); application server runtime state |
| Catalyst Cloud | Hamilton, New Zealand | Object storage (call recordings, invoices, agent skills, customer service artefacts, encrypted database backups) | Call recordings (audio); invoices; agent skills configuration; encrypted database backup archives |
| Anthropic (Claude) | United States | AI conversation engine and call analysis | Call transcript text, knowledge base content, conversation context. Not used to train Anthropic's models under Anthropic's commercial API terms. Retained by Anthropic for a limited period for trust-and-safety purposes, then deleted. |
| Deepgram | United States | Speech-to-text transcription | Live call audio and generated transcripts. All dareena.ai requests are opted out of Deepgram's Model Improvement Program: call data is not used to train Deepgram's models and is retained only for the duration necessary to process the request. No call audio is stored by Deepgram. |
| ElevenLabs | United States / European Union | Default text-to-speech voice generation | Agent response text only — no caller audio is sent. Response text may include details a caller has provided (for example, a name or booking confirmation read back to them). Handled under ElevenLabs' standard commercial terms and retained in accordance with their privacy policy. |
| Inworld AI | United States | Alternative text-to-speech voice generation | Agent response text (no caller audio sent) |
| Microsoft Azure | Australia East region | Alternative text-to-speech voice generation (NZ-accent voices) | Agent response text (no caller audio sent) |
| OpenAI | United States | Text embeddings for knowledge base search | Knowledge base content, search query text |
| United States (with regional processing) | Calendar and Sheets APIs, accessed via tenant-authorised OAuth connections. See Google user data handling in the Privacy Policy for details. | Calendar event data, spreadsheet data, OAuth tokens | |
| Nango | Self-hosted by dareena.ai (New Zealand) | Secure OAuth connection broker for third-party integrations | OAuth tokens, integration metadata |
| Clerk | United States | Authentication and user account management | Account holder name, email, authentication credentials, session tokens |
| Resend | United States | Transactional email delivery | Recipient email address, message content |
| MessageMedia | Australia | SMS notification delivery | Recipient mobile number, message content |
| Cloudflare | Global edge (United States headquartered) | Content delivery network for non-personal branding assets, static page hosting (documentation site, marketing site, embedded widget), inbound support email worker, and DDoS mitigation | Non-personal branding assets; inbound support email content; standard CDN request logs |
| Vercel | United States | Tenant portal hosting (advance-portal) | Portal session data, page request metadata. The portal interfaces with dareena.ai's New Zealand-hosted API; tenant data does not pass through Vercel. |
| Linear | United States | Support ticket management (escalations from in-product chatbot) | Ticket content, account holder name and email |
| Instatus | European Union | Public status page hosting (status.dareena.ai) | Operational status information only; no personal data |
AI processing and training
Three AI providers process call content in real time during a live call: Deepgram transcribes what the caller says, Anthropic generates the agent's response from the transcript, and a text-to-speech provider (ElevenLabs by default) converts that response into speech. The caller's voice audio is only ever sent to the speech-to-text provider; text-to-speech providers receive the agent's response text, never caller audio.
Where a provider offers an option to exclude our customers' data from model training, we use it. Deepgram requests are opted out of Deepgram's Model Improvement Program, and Anthropic does not train on commercial API data under its standard terms. No provider in our stack uses caller voice characteristics to identify or profile individuals — dareena.ai performs no biometric processing.
Questions
Email support@dareena.ai.