Skip to content
Build Perch
Browse guides

How to keep AI prompts private (zero data retention)

Short answer: Zero Data Retention is on by default: every request Build Perch sends to OpenRouter carries a zero-data-retention flag, so OpenRouter routes only to endpoints that attest they keep nothing you send (an endpoint that keeps nothing has nothing to train on). Check or change it under the app menu (☰) → Integrations → OpenRouter → “Zero Data Retention.”

Steps

  1. Open the app menu (☰ in the title bar) and choose Integrations.
  2. Open the OpenRouter entry.
  3. Under “Privacy & Security,” confirm Zero Data Retention (recommended) is checked; it’s on by default, so there’s usually nothing to change.
  4. Leave it on to keep every OpenRouter request (chat, image, speech, music, video, and background summarization) flagged zero-data-retention.
  5. Optional: to keep chat private but let image, speech, music, and video generation use any media model, turn off “Apply to media generation” directly beneath it. It’s on by default and only matters while Zero Data Retention is on.

Notes & gotchas

  • OpenRouter routes only to endpoints that state they keep nothing you send (an endpoint that keeps nothing has nothing to train on); Build Perch can’t verify that end-to-end.
  • Zero Data Retention doesn’t cover OpenRouter’s own storage, which two separate settings in your OpenRouter account control (private input/output logging, and a usage discount in exchange for letting OpenRouter use your prompts and replies), or usage metadata (token counts, latency, model, cost), which is recorded either way. See Zero Data Retention (ZDR) for what it does and doesn’t cover.
  • Turning it off widens which models are available (many free models don’t offer a zero-retention option and reappear once it’s off) at the cost of that assurance.
  • Far fewer media models offer a zero-retention endpoint than chat models, so with both toggles on some image, speech, music, and video picks are refused before anything is sent. Nothing is generated, and Build Perch tells you which model it was, so you can choose another one or turn the media toggle off. A media request carries only the prompt, any text being read aloud, and any reference images you attach.
  • It only affects OpenRouter. Local Ollama models never leave your machine at all, and Ollama Cloud isn’t governed by this toggle.
  • The AI’s own tools can’t reach this switch, so a prompt injection in a file or web page can’t turn it off.
  • Nothing in a workspace can lower this global setting.