Audit log and undo snapshots
Every action through WECOMMANDER leaves a record, and file edits keep the previous content so they can be undone.
What the audit log records
- Every tool call: who (the user behind the AI client or API key), which device or server, the tool, its risk, and the outcome: success, error, denied, approval pending, approval used, allowlisted or run under full trust.
- Every approval and denial, by whom.
- Every change to AI permissions, allowed commands, autonomous mode and scheduled tasks, with the old and new values where relevant.
- Scheduled task runs, as actor scheduled-task:<id>, with who approved the task.
- Root actions are marked with runAs.
Secrets are masked before anything is stored. For SSH servers the gateway records the tool, server and decision but never command text, file contents or output. Open the log in the dashboard under Administration → Audit log.
Undo snapshots
- fs_replace and edit_block keep the previous content of the file; fs_write and fs_append do so with with_undo (files up to 4 MiB).
- fs_versions lists up to eight snapshots of a file.
- fs_undo restores one only if the file has not changed since (its SHA-256 must match), so a newer change is never lost. The snapshot is used up.
- A file written as root because the service user could not write it has no snapshot.
- You can ask the AI to undo its change, or do it yourself through an MCP client.
Local logs
Local MCP keeps its own audit file on your computer (time, client, tool, server, risk, outcome; never content). The agent's local mode writes one with --audit-log.
- Approvals: web, e-mail and desktop notifications: How a high-risk AI action waits for a person, where approval requests arrive (dashboard, e-mail, Windows and macOS notifications, Linux Management Center) and what one approval allows.
- Secret masking: How WECOMMANDER masks passwords, tokens and keys in command, terminal and log output on the device before the result reaches the AI, and what is not masked.
- MCP tool reference: Every WECOMMANDER MCP tool, grouped: what it does, its risk and approval rule, where it works (agent devices, SSH servers, Linux only) and its key arguments.