TrustDeck exists because security questionnaires are about trust — so here is exactly how it treats yours.

• Your Trust Corpus stays in your environment. TrustDeck reads the documents you give it inside your own workspace; nothing in this repo, and nothing in the product, ships your corpus anywhere else.
• The Answer Library is yours. It's a file in your workspace, built from your approved answers. Export it, inspect it, delete it at will.

Upload: policies, standards, executive summaries, attestation summaries (e.g., "SOC 2 Type II, issued 2026-03, no exceptions"), subprocessor lists, your completed onboarding interview.

Never upload:

• Credentials, API keys, tokens, or secrets of any kind
• Detailed/unredacted vulnerability findings or pen-test technical reports (upload the executive summary instead)
• Customer data or anything covered by your own confidentiality obligations

TrustDeck answers questionnaires from descriptions of your controls, not from the sensitive material itself. If a document would hurt you leaked, it doesn't belong in the corpus — and it isn't needed there.

Everything under demo/ and benchmark/ is entirely fictional (Acme Analytics, Inc. does not exist). Demo outputs are watermarked "DEMO — Acme sample data," and the agents are instructed to never cite the demo corpus on a real run. See rule 4 in agents/README.md.

Found a problem — a fabricated answer, a demo-data leak, a broken rule? Open a GitHub issue on this repo. Fabrication reports are treated as release blockers.