Operational features (web terminal, AI chat, file manager)
The cabinet still ships a web terminal, an AI chat panel, and a file manager. They work, and they will keep working. But they are no longer the canonical way to operate a CMDOP machine.
Primary surface is desktop and CLI. The cabinet’s operational widgets remain for browser-only situations and SEO continuity. For everyday work, use ../desktop/ or ../cli/.
Why this changed
Three forces pushed operational work out of the browser:
- Latency. A native PTY and a local relay outperform a WebSocket-bridged shell.
- Reliability. Disconnected network, sleeping laptop, refreshed tab — the desktop session survives all three.
- Trust. Real operators want a trustworthy surface for incidents. Auditors want a clear answer for “where did this command come from”.
Splitting the surfaces lets each one do one job well: the cabinet is for admin and review, the desktop and CLI are for action.
Web terminal — view-only by default
The cabinet’s terminal is now positioned as a read-only inspector for sessions started from desktop, CLI, or agent calls. You can still launch commands from it for compatibility, but those commands are flagged in audit and may be blocked by workspace policy.
- For session inspection: Sessions inspector — open transcripts and metadata.
- To actually run something: open Machines tab in desktop or
cmdop connect <machine>.
AI chat in the cabinet — fallback channel
A web chat panel exists for the moment when desktop is not installed. It uses the same agent loop as everywhere else, but does not benefit from desktop’s per-machine inspector chat or local LLM tools.
- Primary chat: Desktop Chat tab and
../cli/chat. - Per-machine direct pipe: Desktop Inspector Chat.
File manager — light browsing only
The cabinet’s file manager browses files on a registered machine. It is fine for “show me what’s in /var/log” but slow for transfers and lacks the desktop Projects tab features (workspaces, drag-and-drop, search index).
- Real file work: Desktop Projects tab and the agent’s file tools.
- Scripted transfers:
../cli/files.
Schedules — read-only inventory
Schedule authoring is moving out of the cabinet. Existing schedules keep running; the cabinet shows the inventory and the run history.
- Author new schedules: declare them in a skill manifest or use
../cli/trigger. - Inspect past runs: Schedules & runs.
What is not changing
A few things still work the same in the cabinet because they always belonged there:
- Workspace administration — invitations, roles, policies (see Workspaces).
- Billing — payments, subscriptions, usage (see Billing).
- Audit log — every authenticated request, exportable (see Audit log).
- Account — profile, 2FA, tokens, OAuth (see Account).
Workspace policy can block command execution from the cabinet entirely. If you see a “command blocked by policy” message in the web terminal, switch to desktop or CLI — your workspace owner deliberately steered you there.