Skip to content
Baseloop Help Center home

Add an AI agent column

An AI agent column gives an AI model one task and runs it on every row: describe a company, score how well it fits, sort replies, write a first line. At the end, every row has the agent's answers.

Before you start

You need a table with the columns the agent should read, for example the company's name and domain.

Add the column

  1. Click Add column at the end of the column headers, choose AI agent, then Advanced AI agent.

    The add column menu with Advanced AI agent highlighted

  2. Under Prompt, describe the task for one row. Type / and pick a column to put its value in the prompt.

    A prompt that reads the Company and Domain columns

  3. Open Output. Under Output fields, give each answer you want a name and a type, for example summary (Text) and fit_score (Number). Click Add field for each extra answer.

    Two output fields: summary as Text and fit_score as Number

  4. Optional: open Model to choose another AI model. Each model shows its price per row next to its name.

    The Model section with the provider and the model

  5. Click Save & run 10 rows. Baseloop names the column from your prompt; rename it from the column's menu if you like.

    Save & run 10 rows at the bottom of the panel

How you'll know it worked

The first 10 rows show a check and Response. That means the agent answered; the answers are inside the cell.

The new column with Response in every cell and one failed row

Double-click a cell to read every answer. To sort, filter or reuse one answer in another column, hover it and click Add column: it becomes a column of its own, filled in for every row.

One row's answers, with Add column on Fit score

When the first 10 look right, run the rest from the column's run button.

Settings explained

Enable web search. Lets the agent look things up online. Leave it off when the row already holds what the agent needs: it costs less.

System prompt and Examples (shown while web search is off). The system prompt sets the agent's role and tone. Examples pair a sample input with the output you want, and the agent copies their style.

JSON Schema (under Output). For nested answers, such as a list of people at each company that you then send to another table.

Use my own API key (under Model). Pay OpenAI, Anthropic or Google directly instead of using credits. Add the key under Settings → Integrations first. The most capable models are only listed when this is on.

Rate limit. How many requests the agent may send per time window, 10 per second by default. Lower it if your own key hits its provider's limits.

Quick AI agent: one answer, no setup

When one answer per row is enough, choose Quick AI agent instead of Advanced AI agent in step 1. Name the column and click Create. In the menu that opens, describe what you want under Prompt, check the Type Baseloop suggests, and click Save changes. It can search the web and costs 1 credit per row.

If something goes wrong

A cell says "The model returned no answer matching the output schema". Right-click the cell and choose Run empty & failed. If many rows fail, make the prompt and the output fields simpler.

Some answers are empty. The agent leaves an answer empty when it cannot establish it, instead of guessing. Treat an empty answer as "unknown", not "no", in run conditions and formulas.

I can't find a model, such as Claude Opus or GPT-5.5. Those run only on your own API key. Turn on Use my own API key under Model.

The cell shows "Response", not the answer. That's expected for an agent with output fields. Double-click the cell, or give the answer its own column with Add column.