> ## Documentation Index
> Fetch the complete documentation index at: /docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Agent overview

> Research events or companies in the background, once or on a schedule.

The agent researches the news for you and returns a table. You describe what
to find, it runs in the background, and you get one event or story per row,
with its sources.

## Two things to research, two ways to run

| | Run once | Keep updated |
| - | - | - |
| **Events**: a description of what happened | [Event research](/docs/agent/event-research.md) | Event monitor |
| **Companies**: a list of company domains | [Company research](/docs/agent/company-research.md) | Company monitor |

* **Events**: "Bank M\&A deals announced in the US since Sep 15". The agent
  finds the events and fills the columns your question needs.
* **Companies**: `stripe.com`, `adyen.com`, … The agent finds each company's
  news stories, optionally only about a topic such as "funding and
  acquisitions".
* **Keep updated**: once a run succeeds, the agent turns it into a
  [monitor](/docs/agent/monitors.md) that delivers only what is new, hourly, daily or
  weekly.

## One endpoint

Every run starts with `POST /runs`. Send `query` for events, or `companies`
for companies:

```json Events theme={null}
{
  "query": "Bank M&A deals announced in the US since Sep 15",
  "mode": "standard"
}
```

```json Companies theme={null}
{
  "companies": { "items": [{ "domain": "stripe.com" }, { "domain": "adyen.com" }] },
  "query": "funding and acquisitions",
  "lookback_days": 90
}
```

```json Keep updated theme={null}
{
  "query": "Product recalls announced by US food companies",
  "mode": "fast",
  "keep_updated": { "schedule": { "every": "day", "at": "09:00", "tz": "America/New_York" } }
}
```

## Modes

Event research has three modes. Company research has one.

| Mode | How it works | Time |
| - | - | - |
| `fast` (default for events) | One agent. | About 1 to 3 minutes |
| `standard` | Deep research: planned slices, every web page read, each row verified, a cited briefing. | About 10 to 30 minutes |
| `max` | Deep research with stronger models: higher recall and cost. | Depends on the request |

Company runs always use `standard`.

## What you get

Both kinds return the same table, so display, export and monitor delivery work
the same way:

* an events run has the columns its question needed, one event per row;
* a companies run has one story per row, company first, plus one coverage row
  per company in `companies`.

See [Results and exports](/docs/agent/results-and-exports.md).


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.