Home / Records

Every consequential AI action, reconstructable.

Capture what your system did, which model and policy governed it, what evidence it used, and where human oversight intervened. Every record is hash-anchored and exportable.

Open event taxonomy

One schema for the complete behavior trail.

inferencemodel behavior

Inputs by hash or reference, model identity and version, output, timestamp, and linked decision.

human_oversightreview and intervention

Review requests, affirmations, overrides, withdrawals, reviewer identity, and notes.

system_changecontrolled change

Model, policy, prompt, data, and deployment changes with version lineage.

data_subjectrights and notices

Notices, access, erasure, retention, and human-review requests without corrupting history.

incidentforensics

Material events, containment, remediation, and linked activity for reconstruction.

memorywhat the agent knew

Memory writes and revisions share the same authoritative record layer.

Evidence Pack v1

From decision to portable proof in one call.

An Evidence Pack ties a decision to what the agent knew, used, and produced at that moment, with a verifiable package hash.

ReviewEvidence view
EU regulatorArticle 12 system records and oversight trail
Rejected consumerAdverse-action reasons and human-review status
Bank validatorModel and policy versions, changes, and cited evidence
LitigationPoint-in-time production, legal hold, and chain of custody
AuditorContinuous evidence in an open, verifiable format

Make consequential AI behavior reconstructable.

Discuss a workflow →