MeshCore Console
Loading box identity…
MeshCore
Private AI business server
System health
One row per subsystem, with the rule that produced its colour beside the value. A grey row is not a fault — it means the subsystem is not provisioned here, cannot apply to this host, or could not be measured. Rows are always drawn rather than omitted, because a hidden row is a blind spot.
Usage metering
Agents
Each card shows the gateway process and what the agent has been doing. The heatmap is message volume by hour over the last two weeks. Channel bindings show the platform and whether it is connected — never a chat identity, a username, or a thread.
Activity
Vault commits, scheduled runs and agent sessions, newest first. Action classes with no entries stay visible at zero so the vocabulary is legible. Scheduled runs record that output was produced and how large it was; the output itself is a model transcript and is never read.
Scheduled jobs
Job starts by minute of the hour
Declared jobs are reconciled against what actually runs — cron, systemd timers, and the agent runtime's own internal schedule. A job declared with no runner never happens; a runner nobody declared is unaccounted work. Both are shown as drift rather than quietly dropped. Cron records no outcome, so a result of "unknown" means exactly that, not success.
Prophet 21 connector
Approval queue
What this console does not show
- Secrets
- No credentials, keys, tokens or connection strings — not masked, not partial.
- Business data
- It shows that a report ran, never what the report said.
- Transcripts
- Message counts and tool names are telemetry; message contents are never read.
- Controls
- Read-only by construction. There is no start, stop, restart, or trigger.
The collector assembles this page from an allowlist and aborts rather than publish anything credential-shaped. The console is a window, not a lever: the moment it could act, it would become an attack surface on a machine that holds ERP credentials.