Read the audit log
GET /v0/envs/{env}/audit
next is the last event’s id when the page is full, and null when there is nothing more right now. To follow the log, store the last id you processed and call again with after=<that id> on a schedule:
Reading the log is itself audited: every pull that returns events records an
audit.exported event (the format, how many events, and from which id). An empty pull records nothing, so polling an idle environment doesn’t fill the log.The event object
The hash chain the event belongs to (one per tenant) and its position in it. A tenant’s environments share one chain, so
seq has gaps when you read one environment.person, service_account, agent (acting for a run’s caller), system (the platform on its own) or anonymous (for example a failed sign-in).What it was done to. Names are kept for things (agents, guardrails, webhooks…), never for people: a user is named only by id, so erasing a person never means editing the log.
{field: [from, to]} for the fields that changed. Secrets are stored as keyed fingerprints, and long text such as prompts and memory values as a length and a fingerprint, never the text itself.The hash chain:
hash is an HMAC-SHA256 over the event’s fields and prev_hash, the hash of the previous event in the same chain. Checking the hashes needs the installation’s audit key; the panel’s Log intact check does this on every visit to the Audit page.Actions
Reads are not recorded, except exports of the log itself. A prompt deployed from the Prompts tab appears as
version.config_changed followed by version.promoted.
This route returns the events of one environment. Events that belong to no environment — such as sign-ins, or changes to a tenant’s users made on the Users page — are shown to tenant admins in the panel’s Govern → Audit page.

