Hosting / pinecall monitors
pinecall monitors
The numbers the org watches over a window, the line each must not cross, and when each fired.
$ pinecall monitors add "slow answers" --metric e2e_median_s --above 2 --days 7
mon_3f9a1c2b4d5e · slow answers · e2e_median_s above 2 over 7 days · every agent
$ pinecall monitors add "judges slipping" --metric held_rate --below 0.9 --agent front-desk
mon_8b1d2e3f4a5c · judges slipping · held_rate below 0.9 over 7 days · front-desk
$ pinecall monitors
mon_3f9a1c2b4d5e slow answers e2e_median_s above 2 over 7 days every agent fired 2026-10-08 at 2.41
mon_8b1d2e3f4a5c judges slipping held_rate below 0.9 over 7 days front-desk never fired
$ pinecall monitors rm mon_8b1d2e3f4a5c
mon_8b1d2e3f4a5c forgottenThe numbers this org watches, and the line each must not cross. A monitor reads one number of
the observability series — the same the console's Observability screen draws — over a window of
1, 7 or 30 days (--days, 7 when left out), for every agent or one (--agent), and fires the
first time a call's seal finds it on the wrong side of the line: once a day, as monitor.fired
on the agent's log, where the console's Monitors screen and sessions show it with the day and
the value that crossed. --above fires over the line, --below under it.
| metric | what it is |
|---|---|
e2e_median_s |
the caller's wait from their last word to the agent's first, median, in seconds |
llm_median_s |
the model's time to its first token, median, in seconds |
held_rate |
the share of the judges' verdicts that held, 0 to 1 — a floor, --below |
escalated_rate |
the share of calls a person took over, 0 to 1 |
tool_failure_rate |
the share of tool calls that failed, 0 to 1 |
spend_usd |
what the calls cost, in dollars |
calls |
how many calls there were |
rm <id> forgets one. The list is the world's: --prod for production's monitors.