audit & evidence
how do I get an audit trail of what my ai agent did?
Run the agent behind gate.cat and use the built-in gate.cat report: a local record of every irreversible action the agent tried and whether it was blocked — counts and rules, no command text, generated on your machine. It's free and part of the core.
If you need a document for an auditor's file, the optional Evidence Pack turns that log into a coverage report. It's evidence supporting your controls over access and audit logging — not a certification, and it doesn't make anyone “compliant.”
“What did the agent actually do?” is a question you want answered before an incident review or an audit — not during one. gate.cat produces that record as a side effect of guarding the agent.
the free layer: a local, tamper-evident log
Every time gate.cat evaluates an action, it records the decision. gate.cat report summarizes it on your machine: how many irreversible actions were attempted, which rules fired, how often the gate intervened. The default report carries counts and rule names, not the raw command text, so the record itself isn't a new leak of sensitive data.
- Local by default. The pip package has zero telemetry; the report is generated where the agent runs.
- Tamper-evident. The point of an out-of-band gate is that the agent can't quietly rewrite its own history — the log sits outside the model's control.
- Free, in the core. This is not a paid feature; it ships with
pip install gate-cat.
the paid layer: an Evidence Pack for the auditor's file
When you need to hand someone a document, the Evidence Pack replays a battery of attack vectors against your configured rules and produces a coverage report — a single evidence element you put in your own audit file. It speaks to universal control concepts that recognized frameworks share: access control, audit logging, tamper-evidence, and change management (the kind of controls described by SOC 2 CC criteria, ISO 27001, GDPR Art. 32 and PCI-DSS Req. 10).
the durable, off-machine copy (optional Cloud)
An attacker — or a confused agent — that can reach the local log can try to erase it. gate.cat Cloud is an opt-in, off-by-default copy of the veto log kept off the machine, so a durable record survives even if the box is wiped. It is designed so the provider can't read your entries: the reference design encrypts each event on your machine with a key we never receive. Cloud checkout stays in announce mode until that end-to-end-encrypted path ships — a security tool that centralizes readable logs would itself be a leak, so we don't ship it that way.
frequently asked
Does gate.cat make me SOC 2 or PCI compliant?
No. gate.cat produces evidence that supports controls those frameworks describe — access control, audit logging, tamper-evidence, change management. It does not make anyone compliant or certified, is not a guarantee, and does not replace your auditor or your own testing. Compliance is owned by your organization.
What is in the free gate.cat report versus the paid Evidence Pack?
The free report is a local summary of attempted irreversible actions and which rules fired — counts and rule names, no command text. The paid $49 Evidence Pack replays attack vectors against your configured rules and produces a coverage document to place in your own audit file.
Does the audit log include the actual commands?
By default no. The report carries counts and rule names, not raw command text, so the record isn't a new copy of sensitive data. That default is deliberate.
Where is the log stored — does anything leave my machine?
It's local by default with zero telemetry. An off-machine copy exists only through opt-in Cloud, which is off by default and designed so the provider cannot read your entries (each event is encrypted on your machine). Cloud checkout stays in announce mode until that encrypted path ships.
Can the AI agent tamper with its own audit trail?
The gate runs out-of-band from the model, so the agent can't quietly rewrite the decision log. For a copy that survives the whole machine being wiped, the opt-in Cloud keeps an encrypted off-machine record.
put a leash on your agent — 60 seconds
The gate, all 38 default policies, the Claude Code hook and the proxy are free forever (Apache 2.0). No account, no daemon, no telemetry.
Running agents across a team? One rogue agent is the whole team’s blast radius. See the team plan — €149/mo flat →
One email when there's a catch worth showing: a real irreversible command an agent tried, and how the wall stopped it. Opt-in, no spam, no telemetry. The gate itself is free.
✓ you're on the list — first catch lands soon.