Draft, under legal review. Last updated 2026-11-02.
These are the third parties that process personal data on our behalf to provide Larynth. Each is bound by data protection terms. Our data processing addendum explains how we choose them, and how we tell customers about changes.
"Per customer region" means data is processed in the region chosen for the customer's account. "Global" means the provider may process data in more than one country. [counsel: confirm each provider's processing locations and transfer mechanism before publication]
Infrastructure and business services
| Name | Purpose | Data categories | Region |
|---|---|---|---|
| Infrastructure hosting provider | Servers for the API, call engine, database, support chat and backups | All service data, including call data | Per customer region; support chat and operations in the EU |
| Cloudflare | Website hosting, CDN and DNS, inbound email routing, object storage, TURN relay for browser calls, web analytics | Web traffic and IP addresses, support emails, stored objects, encrypted browser call media, page view counts | Global edge network; stored data per customer region |
| Telnyx | Telephony: phone numbers, call routing and spam screening | Phone numbers, call audio in transit, call metadata | Per customer region |
| WorkOS | Sign-in and user authentication | Name, email, sign-in events, IP address | US |
| Stripe | Payments and invoicing; backup identity verification | Billing contact, payment details, transaction history; ID and selfie only if used for verification | Global |
| Didit | Identity and business verification before outbound calling | ID document, selfie and liveness check, biometric face match, business registry details | EU |
| Resend | Sending service and billing email | Email address, email content | EU |
| Sentry | Error monitoring | Error details and technical data, with personal data and IP addresses scrubbed | EU |
| PostHog | Product analytics | Internal user and account ids, product events; no names, emails or phone numbers | EU |
| Better Stack | Uptime monitoring and status page | Service health data; email of status page subscribers | Global |
| Mintlify | Hosting the documentation site | Docs page visits and technical data | Global |
AI providers
Each agent uses the speech-to-text, language model and voice providers it is set up with, including any fallback model, or a speech-to-speech model. Only those providers receive data from its calls. We choose regional endpoints where a provider offers them. We do not allow these providers to use call data to train their models. [counsel: verify zero-retention and no-training terms with each AI provider]
| Name | Purpose | Data categories | Region |
|---|---|---|---|
| Deepgram | Speech to text | Call audio | Global |
| Anthropic | Language model | Transcript text, agent instructions, knowledge snippets, collected details | Global |
| Google (Gemini) | Language model, speech-to-speech model, knowledge base embeddings | Transcript text, call audio for speech-to-speech agents, knowledge base content | Global |
| Groq | Language model | Transcript text, agent instructions, knowledge snippets, collected details | Global |
| Cartesia | Text to speech | Text of the agent's replies | Global |
| ElevenLabs | Text to speech | Text of the agent's replies | Global |
| Gradium | Text to speech | Text of the agent's replies | Global |
Not subprocessors
- Support chat runs on software we host ourselves on our infrastructure hosting provider. No third-party chat company receives your messages.
- Integrations you connect, such as calendars, CRMs and automation tools, receive data on your instructions under their own terms with you.
Changes
We update this page before a new subprocessor starts handling customer call data, and give customers notice as set out in the data processing addendum. To be told of changes by email, or to object, write to privacy@larynth.com.