Privacy and Data Ownership
Telemus AI provides controls for the information and actions available to your assistant. Review the active settings before connecting accounts, enabling tools, or allowing unattended work.
Data handling
- Local processing: Speech, file extraction, and AI processing can run locally where the configured provider and platform support it. Connected providers may receive the information needed for their requested work.
- Usage telemetry: Telemetry is off until you opt in. It does not record prompts, file contents, or tool arguments. Raw events are retained for 90 days; aggregate daily counts may be retained for 400 days. Advertising use is outside the product's telemetry scope.
- Access controls: Choose which folders, accounts, and network locations are available. Revoke access when it is no longer needed.
- Private artifacts: Mark supported workspace artifacts
privateto exclude them from supported export and support-bundle paths. Review any material before sharing it; this is not a guarantee that every external destination enforces the same privacy controls.
Credentials and sharing
Android uses platform-backed credential storage. Desktop uses the operating system's secret service where available, with restricted local storage as a fallback. Keep device profiles and backups private.
Logs and support exports redact recognized sensitive information. Automated redaction cannot identify every secret, so inspect a support bundle before sending it. Do not paste passwords, API keys, account-linking codes, or pairing URLs into a public issue or support conversation.
On Android, opening a supported file in another application grants that application temporary access to a materialized copy. Confirm the destination application and the information you are sharing.
Execution and approval settings
The current default permits agent tool execution without a confirmation for every
command: VRAI_EXEC_ASK=off and VRAI_EXEC_SECURITY=full. This can include scripts,
remote commands, file edits, and schedule changes. Enable tools only for work and
resources you intend the assistant to access.
Use stricter settings when you need more review:
| Setting | Effect |
|---|---|
VRAI_EXEC_ASK=on-miss | Requests approval when a command does not qualify for automatic execution under the active policy. |
VRAI_EXEC_ASK=always | Requests approval for execution calls. |
VRAI_EXEC_SECURITY=allowlist | Restricts execution to the configured allowlist and applicable approval rules. |
VRAI_EXEC_SECURITY=deny | Disables command execution. |
A call can request a stricter policy, but cannot relax the configured policy. Unsupported policy values are rejected. Unrecoverable destructive commands are refused regardless of the approval setting. Some actions, including protected sending, payment, paid remote sandbox use, and reviewed plugin actions, have additional confirmation requirements.
Read-only memory recall is automatic by default, including configured external
memory providers. Set VRAI_ACTION_POLICY_MEMORY_READ_ASK=on to request approval
for memory reads. Memory writes and history changes retain their applicable
confirmation requirements.
Channel work
With VRAI_EXEC_ASK=on-miss, ordinary channel work can proceed automatically under
the channel's configured policy. Destructive, credential, sending, billing,
restore, and other protected actions still require the applicable review.
Operators can use VRAI_ACTION_POLICY_PROFILE=default and
VRAI_CHANNEL_EXEC_AUTO_APPROVE=0 to disable channel automatic approval under
that policy. See Skill Library.
An approval applies to the action shown. A later protected step can require a new confirmation. Channel approval prompts normally expire after 30 minutes; your operator can configure this. Confirming an expired prompt does not run the expired command and may restart review of the original request. Repeated confirmation of an already consumed approval does not authorize another action.
See the active policy
In the Command Center, open Workspace Inbox and inspect Approval policy. The card shows the active profile, automatic approval settings, and whether each value comes from a saved setting, the server environment, or the default. Saved settings can be changed with confirmation. Environment-controlled values are read-only there and require the operator to change them.
Use /policy in the terminal. The Web Logs panel and /logs show recent
activity where available. Risky and permissioned actions are recorded according
to the active audit policy, with sensitive details redacted.
Additional protection when ask is off
Set action_policy.exec_ask_off_hardline=on or
VRAI_ACTION_POLICY_EXEC_ASK_OFF_HARDLINE=on to retain confirmations for
destructive, external, and costly action classes even in ask-off/full mode.
Ordinary review actions remain automatic. A saved setting takes precedence over
the environment. Use explicit off to disable this option; an empty or
unrecognized configured value enables the stricter behavior.
This setting applies to supported local and channel work. Existing permissions, session approvals, and the separate memory-read setting still apply. Inspect the effective policy rather than assuming a setting affects every tool identically.