Trust

Last updated on July 18th 2026

The model companies work hard to keep their models safe — that's their job. Our job is keeping your life private, from the model companies.

In practice that means three commitments you can check: we tell you exactly which providers touch your conversation and why, we never use your data to train models, and your workspace — the recordings, transcripts, and analyses your meetings produce — belongs to you, until you delete it.

The short version

• A Waterr call is one human and one agent. Every agent introduces itself as an agent — honestly artificial, always.
• We do not use your conversations, transcripts, or recordings to train or fine-tune any model.
• Recording is off unless the meeting enables it, and scenarios can require explicit participant consent before anyone joins.
• Everything a meeting produces lands in your workspace. It stays until you delete it — there is no silent expiration, and deletion is yours to trigger.
• Sensitive fields — participant names, transcripts, memory, consent messages — are encrypted at the column level (AES-256-GCM) in our database, on top of TLS in transit.
• Every provider in the data path is listed on this page, with what flows to it. No unlisted processors touch your conversation.

Where your conversation goes

No vague diagrams — this is the actual path of a live meeting:

• The call itself runs over Daily.co's WebRTC network.
• The agent's realtime intelligence is Google's Gemini Live, running on Vertex AI. Call audio — and camera or screen frames, if you share them — stream to it for the duration of the call. Ori Realtime, our own interaction model running on infrastructure we operate, is available as an opt-in realtime backend.
• The live transcript is produced by Deepgram from the call audio.
• The agent's voice is synthesized by the voice vendor your scenario selects — OpenAI by default; ElevenLabs, Cartesia, Azure, or Deepgram optionally. Only the agent's own lines are sent for synthesis.
• Recordings — only when enabled — are written to our AWS S3 bucket and played back through signed CloudFront URLs.
• Transcripts and analyses are stored in our database with content encrypted at the column level.
• Post-call analysis routes through our model gateway to the model you select (Google, Anthropic, or OpenAI). The gateway passes conversation content through; it does not store it.
• Billing meters token and character counts — numbers, never content.

Subprocessors

The complete list of third parties that process customer data, and what each one receives:

The Waterr desktop app can additionally connect to services you explicitly authorize (email, calendar, and other integrations via OAuth); those connections act on your accounts only with your authorization and are covered in the Transparency & Trust Center.

Retention

Your data persists until you delete it. There is no automatic expiration and no silent cleanup — what your meetings produce stays in your workspace, under your control, for as long as you keep it. Third-party model providers in the data path retain API traffic per their own published API terms, which exclude using customer API data for training.

Deletion

• Meetings, transcripts, and recordings can be deleted directly from your workspace or via the API.
• Uploaded scenario documents can be removed from their scenarios at any time.
• For complete removal of an account and everything attached to it — including storage copies of recordings — email [email protected] from your account address and we will purge it.

Training

We do not use your conversations, transcripts, recordings, or uploaded documents to train, fine-tune, or improve any model — ours or anyone else's. There is no default data-sharing program to opt out of, because there is nothing to opt out of.

Security

• All traffic is encrypted in transit with TLS.
• Sensitive database fields — participant names, transcript content, agent memory, consent messages, webhook and redirect URLs — are encrypted at the column level with AES-256-GCM.
• Recordings and files live in cloud object storage (AWS S3, Azure Blob) with provider-managed encryption at rest.
• API access uses scoped keys; account access uses token-based authentication.
• Account-level actions — key creation and revocation, plan changes, organization changes — are written to an append-only audit log.

Honestly artificial. Always.

Everything built on Waterr identifies itself. Every agent tells you it's an agent — in its first breath, not in the fine print. Not agents that pass as human; agents worth talking to as agents.

The full document

This page is the short, checkable version. The long form — data lifecycle, enterprise controls, vulnerability reporting — lives in our Transparency & Trust Center, alongside our Privacy Policy and Terms of Service.

Questions your security team wants answered directly: [email protected].