Performance issues
“Cmdop is slow” is four problems with four fixes, so identify the layer first. A reply slow to start points at the inference service; one that streams fine but works slowly points at the machine or the task; lag on every screen points at the connection to the relay; and one slow machine among many points at that host.
“Cmdop is slow” is almost always one of four different problems, and they have different fixes. Identify the layer first.
Which layer is it?
| What you notice | Likely layer | Check it |
|---|---|---|
| The reply takes a long time to start, then streams normally | The inference service | Try a different model for the turn; the key that reaches it is set in Settings → AI Providers on the relay. |
| The reply streams fine but the work itself takes forever | The machine, or the task | Read the machine’s health — see Machine health. |
| Everything is laggy, including simple screens | The connection to the relay | cmdop status — see Connection issues. |
| One machine is slow, the others are fine | That host | cmdop care status on it. |
Start here on the affected machine:
cmdop status
cmdop logs -fLook at the machine before you look at Cmdop
Machine Care collects bounded facts about a host — what it observed, what it concluded, and how much of the host it managed to cover — so you can answer “is this box under pressure?” without logging in and running your own scan:
cmdop care statusThe same facts are in the web console under Machines, for every joined machine at once. Care reports coverage rather than assuming it, so a fact it could not collect shows as not collected — not as a healthy zero.
Slow file transfers
Large attachments are cached on the relay, and the relay caps that store. If uploads start reporting storage full, raise the budget rather than hunting for a throughput setting — see the storage section of Self-hosted relay.
What Cmdop does not promise
Cmdop publishes no fixed latency targets or throughput guarantees, and the console does not invent a quality score for a value the relay cannot measure. If a number is not shown for a machine, that is because it was not measured.
Related
Common questions
Why is Cmdop slow?
Identify the layer first. Slow reply start points at inference, slow work points at the machine or task, lag across simple screens points at relay connection, and one slow host among many points at that host.
How do I check whether the machine is the problem?
Run cmdop care status on the affected host or read Machine Care in the web
console. Care reports what it observed and what it could not collect instead of
inventing healthy values.
Does Cmdop publish latency guarantees?
No. Cmdop does not publish fixed latency or throughput guarantees, and the console does not invent quality scores for values the relay cannot measure.