Skip to main content
A monitor is an analyst request on a schedule. You write it once, the same request you would type into chat, and it runs when you said it should. Open Monitors in the left navigation. You can have up to 20; the list header shows how many you have used.

Skill

A process

The same checks on demand

Monitor

A request on a clock

The same question, later, without you there

Subagent

A specialist

Not this. A run is a normal conversation.

Monitors list with each monitor's schedule and next run

Writing a monitor

A monitor has two parts: a title, which is how it appears in the Monitors list and how you’ll recognize its runs later, and the prompt that defines what it does. The prompt is a full analyst request, so everything that works in chat works here. Point at a company with $, or a note or folder with @, and the analyst reads them before it starts. A monitor that checks a thesis actually opens the thesis.
Summarize overnight news for and flag anything that touches a thesis.
Common uses: a morning brief every weekday at 7am, a competitive check on one company every Friday, a one-off read of an earnings release the morning after it drops.

Scheduling

The schedule works the way a calendar event does. You pick a first run in your timezone, and the repeat menu offers the common rules that follow from it: run once, every day, every weekday, every week or every second week on that day, monthly on that date or on that weekday, yearly. Custom covers everything else a calendar can say: every N days, weeks, months or years; only in certain months; on chosen weekdays, on the 1st, 2nd, 3rd, 4th or last of those weekdays each month, or on specific days of the month including the last. The fourth Thursday of November, the 1st and 15th of every quarter, Fridays from June to August. Advanced takes a bare iCalendar recurrence rule such as FREQ=WEEKLY;BYDAY=MO,WE;COUNT=8, for rules with an end date or a fixed number of runs. The RRULE generator at icalendar.org builds one from a form. The rule is checked before it is saved. Whatever the rule, the time of day is the first run’s time, and a monitor runs at most once a day.

Every run is a real conversation

When a monitor fires, it starts a fresh chat: a full analyst session with its reasoning and data calls visible, exactly as if you had asked yourself. If a run surfaces something odd, you are already in a conversation about it. Ask the follow-up, pull the filing, update the note. Runs collect under the monitor that produced them, and a dot marks the runs you have not read yet, so you can catch up from a list after time away.

The analyst can set them up too

You can also create a monitor mid-conversation: say “check this again every Monday and tell me if anything changed,” and the analyst drafts it as a proposal you approve. A monitor stays editable afterward: change the wording, move it to a different day, pause it for earnings season, or delete it when you close the thesis.

What’s next

Reactive triggers (price crosses a level, volume spikes) are not built yet. The plan is for the two to work together, with the threshold as the trigger and the analyst as the response, so a price alert arrives as a run that has already reread your thesis. An analyst on a schedule explains the thinking.