Skip to main content
An agent is a prompt you save once and run on any table. This guide takes a list of companies and has an agent research each one. 2:01 video, recorded August 2026. The app moves fast, and the steps below stay current.

​
What you’ll learn

  • Saved agents: build an agent once on the Agents page, then use it on any table.
  • Input mapping: tell the agent which of your columns it should read on each row.
  • Fan-out outputs: every field the agent returns arrives as its own column.

​
Before you start

You need a table with a company domain or a LinkedIn URL on each row. The table from the Find people guide works well.

​
Build the agent

1

Open Agents

Click Agents in the sidebar. This is your library of saved agents.
2

Start from a template

Pick a card under Start with a template, or click New agent to write the prompt yourself.
3

Check the inputs and outputs

The Inputs card lists what the agent reads on each row. The Outputs card lists the fields it returns.
4

Set the effort

Effort decides how hard the agent works per row. Turn on Web research if it needs to read the open web.

​
Run it on your table

1

Add the agent to a table

From the agent, choose Add to table and pick your table.
2

Map its inputs

Point each input at the column that holds it. The dropdown ranks your columns by fit.
3

Run a first batch

Run a small batch before the whole table, so you can read the answers before spending more.
One new column arrives per output field, and cells fill in as rows finish.

​
Good to know

Yes. A saved agent is a library entry, so add it to as many tables as you like and map its inputs again each time.
The agent had nothing to work with, or it declined to guess. Open the row’s cell and read the steps it took.
An agent is priced by the work it does per row, so effort and web research both move the number. The cost shows before you run.
No. The Agent tab in the column catalog lists Built-in agents you can use straight away.

​
Next steps

Build an agent

The editor, in full, with every configuration section.

Read an agent run

Follow the steps an agent took on a single row.

Agent columns

Add an agent straight from the column catalog instead.

What agents cost

What moves the price, and how to estimate before a full run.