See what the AI did.
Every action, on a chain, with a head you can check.
Every vault read, every write, every grant you approve or revoke lands on your Token Holder usage chain. The audit view renders them as plain sentences — who, what, when — filterable by kind and date, with the chain head hash and the number of rows chained under it printed at the top. Press Verify integrity and Token Holder recomputes the chain and answers.
Export is a compliance artifact, so it behaves like one: JSON and CSV are available on real rows and disabled on the sample data a signed-out visitor sees, because a downloaded file of sample events is exactly the thing someone would later mistake for evidence.
Chained activity, in plain sentences
Live todayVault reads, writes, grant approvals and revocations, rendered as readable rows and filterable by kind and date.
Chain head + integrity check
Live todayThe head hash and its chained-row count sit at the top of the view; one press recomputes the chain and returns a verdict.
Verify integrity · head hash + row count
Export that refuses to launder samples
Live todayJSON and CSV export on real rows; disabled on the demo seed, so a sample file can never be mistaken for a record.
Live file-by-file panel with diffs
RoadmapA panel that fills as the agent works — time, tool, path, agent identity, and the exact diff one click away, each event Ed25519-signed. Built against the local agent host; not yet mounted in the workspace.
Two different records, and this page keeps them apart. The usage chain above is live today and covers reads as well as writes, but its rows are chained rather than signed one by one. The per-event Ed25519 signatures, the diffs and the exportable receipt belong to the local agent host's file-activity ledger, which covers file writes only and is not wired into the workspace yet.
The verifiability claim is always one click from a way to check it — no account, no trust required.