Lens is an AI-native product. This page describes how we use AI on customer data, where customer data goes when AI is involved, what is and isn't retained, and the choices we make when configuring our AI providers.

The AI providers Lens uses

Lens generates intelligence using Anthropic's Claude family of language models. Voice transcription, when used, is performed via OpenAI's Whisper API as a one-shot service.

NomoCoda does not host, fine-tune, or train its own foundation models. NomoCoda does not send customer data to any AI provider other than those named on this page and on the Sub-processors page.

What gets sent to AI providers

When Lens generates intelligence for a user, the data sent to Anthropic is bounded by what that user already has access to in their connected systems. The synthesis layer reads only signals tagged to that user's identifier and workspace, projects them into a structured summary, and sends that summary to the model along with the user's chosen archetype prompt and recent context.

Lens does not transmit:

• OAuth tokens, refresh tokens, or any credential material.
• Raw payloads returned by connector APIs.
• Another user's data on any prompt, even another user in the same workspace.
• Billing data, account passwords, or personal identifiers beyond the user's first name and chosen archetype.

For voice transcription, the audio sample is forwarded to Whisper for one-shot transcription. The transcript is returned to Lens; the audio is not retained by Lens or by OpenAI under the API terms in effect when this page was last updated.

Model training and data retention

Anthropic does not train its models on inputs to or outputs from the Claude API under the commercial terms in effect when this page was last updated. OpenAI does not train its models on inputs to or outputs from the Whisper API under the commercial terms in effect when this page was last updated. NomoCoda has not opted in to any training program or data-sharing arrangement with either provider.

NomoCoda does not retain raw model inputs beyond the duration of the request that generates them, except where the resulting output (a card body, a chat reply, a transcript) is stored as part of the customer's intelligence record.

Where AI outputs are stored

Generated intelligence — cards, chat replies, summaries — is persisted in the customer's workspace and scoped to the originating user's identifier. Stored AI output follows the same isolation architecture as raw input data:

• Reads are filtered by both workspace and user at the application and database layers.
• Database row-level security policies enforce per-user filtering independently of application code.
• When a user's source-system access narrows, related outputs are removed from the active feed in the same reconciliation pass that drops the underlying signals.