Floe engages the third parties listed below to help deliver the Service. This page is Annex III to the Data Processing Addendum and constitutes the general written authorisation described in Section 5 of that Addendum.
We will update this page at least 30 days before a new sub-processor begins processing personal data on your behalf. To be notified of changes by email, write to privacy@floe.so and ask to join the sub-processor notification list. You may object to a new sub-processor on reasonable data-protection grounds as described in Section 5 of the Addendum.
| Sub-processor | Purpose | Data processed | Location |
|---|---|---|---|
| Google Cloud Platform | Application hosting (Cloud Run), primary database (Cloud SQL), cache (MemoryStore) | All Customer Data and Demo Interaction Data at rest | United States (us-central1) |
| Cloudflare (R2) | Object storage for session recordings, transcripts, and uploaded assets | Recordings, transcripts, brand assets, uploaded documents | United States |
| Daily | WebRTC transport for real-time voice sessions | Audio and video streams in transit during a session | United States |
| Pipecat Cloud | Hosting for the voice agent runtime | Session audio and conversation state during a session, including a visitor's page-activity summary where the site has that on | United States |
| Sub-processor | Purpose | Data processed | Location |
|---|---|---|---|
| Google (Gemini) | Primary large language model for agent reasoning and generation; text embeddings for search | Conversation content, page context, a visitor's page-activity summary where the site has that on, knowledge-source excerpts, search-index text (ingested content, reference guides, plan content) and query text | United States |
| Anthropic | Fallback large language model | Conversation content, page context, a visitor's page-activity summary where the site has that on, knowledge-source excerpts | United States |
| OpenAI | Audio transcription for uploaded recordings and video | Uploaded audio and video sent for transcription | United States |
| Deepgram | Speech-to-text transcription | Session audio | United States |
| Cartesia | Text-to-speech synthesis | Agent response text | United States |
| Anam | Video avatar rendering, where a customer enables it | Agent response audio and video output | United States |
| Sub-processor | Purpose | Data processed | Location |
|---|---|---|---|
| PostHog | Product analytics and session replay on Floe's own web properties and dashboard | Usage events, page interactions, session replays of Floe surfaces | United States |
| Sentry | Error monitoring and performance tracing | Error traces, stack traces, limited request metadata | United States |
| Langfuse | AI diagnostics and response-quality investigation, where enabled | AI trace identifiers, model usage and timing; model instructions, messages and responses when content diagnostics are enabled (visitor sessions also require recording permission) | United States |
| Grafana Cloud | Operational monitoring, where enabled | Technical identifiers, event categories, timing and failure metadata; no conversation content | United States |
| Resend | Transactional email delivery | Recipient email address, message content | United States |
| MailSlurp | Disposable inboxes used to complete magic-link sign-in to a customer's demo account | Sign-in emails sent to the demo inbox | United States |
Floe contracts with its AI model providers on terms that exclude use of submitted content for training their general-purpose models. Floe does not use Customer Data, Demo Credentials, or Demo Interaction Data to train or fine-tune AI models without your prior written consent, as stated in our Privacy & Data-Use Disclosure.
Last updated