Trust and data use

What we collect

A workspace stores the company domain, its question set and question roles, public answers collected from ChatGPT, Gemini, and Google Search, source links returned with those answers, analysis outputs, audit records, and provider cost receipts.

Results describe the captured research sample. They do not establish market share, product truth, or how every user will see a company.

Where it goes

Bright Data collects answers from the configured public surfaces. We use OpenAI's GPT-6 Luna model through OpenRouter to code answers, and Anthropic models through OpenRouter for other research steps. Free-text company and buyer descriptions may be included in model requests. Provider data collection is set to deny, but zero data retention is not enabled for every request.

Application records are stored in the workspace database. We will publish production data regions after those environments are configured.

Access and sharing

Every application record is scoped to a company workspace. Membership is checked on every API request, and database row-level policies provide an additional boundary. Provider keys remain on the server.

Private-link sharing is off. Workspace access is controlled by membership.

Retention

Research answers, analyses, audit history and receipts are currently retained so a result can be reproduced and reviewed. A standard deletion period has not been approved yet. We will state the period here before production use rather than imply one.

Zero data retention is not enabled for every OpenRouter request. We ask users to keep personal and confidential information out of free-text fields.

Failures and review

Transport failures retry without consuming an evidence attempt. Evidence is checked deterministically before model audit. Rejected matrix rows are visibly withheld, while accepted rows remain available. A whole scope requires human judgment only when fewer than half of its rows would pass.

Collection and analysis jobs use durable identities, leases, reservations, receipts, and restart-safe checkpoints to prevent duplicate paid work.

What we do not claim

LatentFrame has not been certified under a security or privacy framework. Public AI outputs can be incomplete, inconsistent, or wrong. We preserve complete originals and exact supporting quotes so customers can inspect the basis of each measurement.