Privacy and Session Data
FieldMouthAI stores data needed to run governed FieldObjects, preserve audit trails, and support approved user context.
The system may store
- –submitted input payloads
- –generated result payloads
- –provider and model metadata
- –runtime session records
- –CICP approval records
- –server-computed hashes
- –Vault field values approved or entered by the user
- –registry and manifest metadata
- –guided interview turns for signed-in users
FieldMouthAI uses this data to
- –execute FieldObjects under policy
- –preserve version fidelity
- –create audit records
- –prevent duplicate approvals
- –support Butler Vault prefill
- –show audit history to authorized users
- –verify manifest and registry integrity
Butler Vault
The Butler Vault stores field-level values for reuse. AI output may enter Vault only after CICP approval, only for version-backed sessions, only for authenticated approvals, only for exact field_key matches, and only when the frozen field contract marks the field vault_eligible: true.
AI output is never ground truth by default.
CICP records
CICP records preserve approval evidence. They include server-computed hashes of input, result, approval, provider, model, and version context. CICP records are intended to be immutable audit records.
Guided sessions
Signed-in users may have in-progress guided interview turns saved so they can resume later. Signed-out users use in-memory guided state only.
Admin access
Builder, registry mutation, verification, re-attestation, and audit views are admin-gated. Admin access uses a server-issued signed token.
Training
FieldMouthAI application data is not used to train public AI models by default. Provider-specific processing depends on the configured AI provider and deployment environment.
Important limits
- —do not enter information you are not authorized to submit
- —high-liability use requires human review
- —life-safety protocols cannot auto-execute
- —exported registry packages are read-only evidence packages, not legal certifications
FieldMouthAI stores session data to govern execution, preserve auditability, and support approved context — not to make AI output automatically authoritative.