Privacy and encryption

Agent Approve receives the information needed to show supported agent activity and make policy or approval decisions. The exact content depends on the integration, event type, mode, and encryption settings you choose.

• Privacy levels control how much command detail remains in event history.
• Data Retention controls how long you keep that history available in the app.
• Encryption keys explains end-to-end encryption, rotation, Key Retention, and missing-key recovery.

Approval content and history

Approval Mode needs enough command or tool information to evaluate your active policy. When end-to-end encryption is enabled, your computer creates a service-encrypted approval copy and a separate device-encrypted copy. The service opens the approval copy to evaluate policy. It cannot use that same copy as your readable event history.

The device-encrypted copy used for event history follows your selected Privacy Level. Other supported activity content is also encrypted for your devices when end-to-end encryption is enabled.

Some metadata remains available so Agent Approve can route a request and show when an agent, event, or decision occurred. End-to-end encryption does not mean that every field is hidden or that policy can be evaluated without processing approval content.

See Encryption keys for key handling and Privacy levels for the difference between tool name, summary, and full-content history.

Policy Rules and Generate New Rules

Policy Rules is included with eligible Solo and Pro plans. Rule Evaluator processing starts only after you choose a rule evaluator, confirm its data sharing, and enable a rule that needs it.

The rule evaluator receives the action being checked: its tool name and arguments, including any file contents or message text in those arguments, plus the working directory, platform, shell, and agent type when supplied. It does not receive earlier commands, tool outputs, or conversation history. Agent Approve does not read files from your computer to check a rule.

If you select Truncate action, an oversized action is sent as a prefix with a note stating how many bytes were sent and omitted. The rest is not checked. This setting does not change the complete action used by your lists or live approval.

Agent Approve Fast uses TypeSafe and included Google Gemini processing. You can instead configure your own provider account, or use one as a backup. See Choose a rule evaluator for the available providers and what leaves your device for data sharing. Choosing None stops rule evaluator requests. Exact-match rules run on Agent Approve's servers without a rule evaluator.

Each rule check keeps an evaluation record with rule identifiers, match probabilities, the decision, timing, and token counts, but not the command. These records follow your Data Retention setting.

Generate New Rules sends your description and follow-ups, draft cards, and the selected policy's rules, list entries, and default behavior to the providers disclosed in the app. It does not send your agents' tool-call history, events, files, encryption keys, rule evaluator credentials, or other policies. Nothing is sent until you submit your description.

Drafts are stored encrypted and expire after 30 minutes without activity, or two hours after the session starts. Scheduled cleanup removes expired drafts within 24 hours; discarding a draft removes it immediately. Saved rules are turned on and can be turned off or removed at any time.

Handling of data on the providers' own systems is governed by their terms. The full statement is in the Privacy Policy.

Do not paste secrets into support messages. When troubleshooting, remove tokens, keys, full device identifiers, private repository paths, command output, and customer data unless Support asks for a specific redacted field.