AI Audit Trail Without Storing the Prompt
We do not store what you told the AI. We store that you can prove you said it.
Framework / standard: EU AI Act Art. 12 · ISO/IEC 38507 §6.5
Almost every company uses AI. Almost none can say what it sent to its provider yesterday at 15:42, who requested it and whether it contained personal data.
Linedat logs every model call in a governed ledger, with enough information to audit without storing sensitive content.
Forensics without storing content
Each entry stores provider, model, action type, who and when, timing, detected PII categories (indexed) and deterministic SHA-256 hashes of prompt and response — never the raw text. If something looks off, there is forensics; if you are concerned about privacy, we do not store the prompt.
Chained by trigger
The AI usage ledger is cryptographically chained by database trigger (the app never writes the hash), just like the data audit log. It proactively materialises the high-risk logging requirement of EU AI Act Art. 12.
The limits (what we do not claim)
By design the original prompt cannot be recovered: only hashes and categories are stored. The record is written after the user receives the response (it is not cryptographically bound to the content returned). PII categories depend on the sanitiser having been run for that feature.
How Linedat helps
Linedat gives you tamper-evident AI forensics without becoming a repository of sensitive prompts — evidence that you can prove it, not the content itself.
Related capabilities
Other catalogues store logs. Linedat stores proof — verify it with a button.
PII Sanitisation Before Reaching the ModelBefore a national ID or an IBAN leaves for OpenAI, it is redacted — in the critical path, not as an option.
Bias Detection with p-valueWe do not tell you there is bias: we prove it, with a p-value.