01
Verify without decrypt
Your auditor confirms the record is intact and correctly attributed without reading a word of it. Proof for them, privacy for your people, and neither traded for the other.
Audit
An AI audit trail that keeps the whole turn.
Where a log tells you something happened, Verillian writes down the whole interaction: what was asked, any reasoning the model exposed, what it did, and the ruling it met. Signed and sealed under your key.
01
The record
Inside one entry sits the entire turn.
Open a single entry and you're reading the context the model was given, the prompt the person typed, any reasoning the model exposed, the response, the tool call it attempted, and the allow or block that met it. Where a redaction ran, the sealed prompt itself shows the model was handed a stand-in token instead of the real value, which is how you can see it held. So when a records request or a courtroom asks what your AI did last spring, you answer from evidence, not recollection. The audit controls behind that record are aligned to CJIS Security Policy v6.1.
02
Why it's evidence
Proof that survives a hostile reader.
A record only counts as evidence if it holds up in front of someone who doubts you. This one is built for exactly that reader, in three ways.
02
Readers get recorded
Every look at the record is written into its own signed chain, and if the access can't be written down, the record doesn't open. A reviewer's reading becomes evidence, never a gap.
03
Sealed on your servers
The chain lives encrypted on infrastructure you run, under a key only you hold. We never receive the record, so we hold no copy of it to lose.
03
How it works
Click into any row.
Read the whole turn.
Every governed interaction is one signed row. Open it to see the prompt, any reasoning the model exposed, the tool call, and the decision, exactly as it was recorded on the device.
Last 7d
email...
anthropic, openai...
Any
Filters
Hide turns without responses
Clear⋯
9 turns query_run / signed on device / last 7d
Time
User
Device ?
Provider
Model
Decision
Redaction ?
Tool ?