query without companies to POST /runs. The agent finds the events
that match, decides which columns the question needs, and fills one row per
event.
Writing the request
Thequery is the whole brief. Put in it:
- what happened: deals, recalls, layoffs, lawsuits, product launches;
- who or where: an industry, a country, a type of company;
- when: “since Sep 15”, “in Q3 2026”, “last week”. Event runs read their time window from the query;
- what to capture, if it matters: “with the buyer, the target and the deal value”.
Parameters
string
required
What to find, with its time window.
integer
default:"50"
The most rows to return: 50 or 100. More is refused with
too_many_rows.
A run is charged per 50 rows delivered past the first 50.Choosing a mode
fastfor a first look, or when 1 to 3 minutes matters more than recall.standardwhen every row must be checked against its sources and you want a cited briefing with the table.maxfor the hardest lists, where missing an event costs more than the extra time and price.
What you get
GET /runs/{run_id} returns, once the run has finished:
columnsandrows: the table;window_fromandwindow_to: the time window the agent read from your query;report: the cited briefing thatstandardandmaxruns write.

