Data firewall

Definition

A filtering boundary that turns prompt-adjacent input into a minimal delivery signal before anything reaches the ad market.

What leaves the app

In the recommended CS-S (S1/P0) setup, the firewall outputs only:

  • Topic category: a coarse label such as travel, coding, finance, health.
  • Language: used for localization and policy enforcement.

What stays inside

The firewall is designed so the ad market does not receive:

  • Raw prompt text
  • Conversation history
  • User identifiers or account data
  • Personal data
  • Model output

Consent and failure mode

Topic-based matching requires explicit consent under Compute Sponsoring. Without consent, delivery falls back to context-free matching (CS-N).

If the config is missing/invalid or the firewall cannot validate the request, the sponsoring path blocks by default. See Safety.

Related

Last updated: 2026-04-03

Next step

Explore the integration path.

Start with the SDK for the production path, or talk to the team if you want rollout, placement, or partnership guidance.