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

# AI agent columns

> Add a column that runs a prompt on every row, reads other columns, and writes back one or more fields.

An AI agent column runs a prompt against each row and writes the result into one or more fields. Use it for what a formula or an enrichment column can't do: research a company from its domain, classify a lead, or write a line on why it's a fit.

## Add an agent column

<Steps>
  <Step title="Open the Agent tab">
    In a table, open the column catalog and select the **Agent** tab. It lists **Your agents** first, starting with **Use AI**, followed by any agents you've already saved, then **Built-in agents** that Oneprofile ships ready to run. Click **Use AI** to build a new prompt from scratch.
  </Step>

  <Step title="Describe the column, or write it by hand">
    The column opens on an **AI Assistant** card with a box asking "What should this column find or write?". Describe what you want, for example "Find the company's HQ city and country", and click **Generate** to draft a prompt, effort, and outputs for you to review. Click **Set up manually** to skip straight to the empty fields instead.
  </Step>

  <Step title="Write the prompt">
    In the **Prompt** card, type `/` to reference another column's value, for example "Write a one-line summary of /Company". Arrow keys navigate the picker, and the right arrow drills into a column that holds several fields or list items. Click a pill to mark that reference required, so a row is skipped when the value is missing, or press Backspace to remove it. The **Add column** button above the prompt inserts a reference the same way, without typing `/`.
  </Step>

  <Step title="Choose effort">
    The **Effort** card sets which model the column runs on. Each option shows its credit cost per row, and **Compare effort levels** opens the full trade-off. Turn on **Web research** to let the agent search the web and read pages instead of answering only from the row and what it already knows, and turn on **Use business context** to add your company profile to every run. The **Advanced** disclosure lets you set a **Starting URL column** the agent's research begins from, and optionally keep it on that domain.
  </Step>

  <Step title="Define the outputs">
    In the **Outputs** card, pick **Fields** to list what comes back, each with its own name and type, or switch to **JSON Schema** to write a raw schema instead. Click **Add output** for another field and **Add description** to explain one to the model. Each field becomes its own column on the table once the agent runs.
  </Step>

  <Step title="Preview, then create and run">
    Click **Preview on 3 rows** to run the column on the table's first three rows for free before committing anything. When you're ready, use the split button to **Create & run 10 rows**, **Create & run 50 rows**, or **Create and don't run**. Each option shows its credit cost before you commit to it.
  </Step>
</Steps>

<Note>
  A column deployed from a saved agent stays linked to it: the prompt, effort, output, and tools are read-only on the column, and its banner tells you to open the agent to change them. You can still remap which table columns feed its inputs from the column itself. Click **Stop using this agent** on the banner to keep a copy of its settings on the column instead.
</Note>

## Saved and built-in agents

The **Agents** page is where you build, save, and browse agents beyond a single table. It opens on **Build an agent**, the same **AI Assistant** builder, with starter templates and a **Browse agents** list of everything you've saved. Built-in agents there are ready to deploy to any table without writing a prompt yourself.

## Next steps

* [Enrichment columns](/tables/enrichment-columns): Add columns that fill themselves in from multiple sources
* [Running and views](/tables/running-and-views): Control what runs and when on a table
* [Credits and billing](/platform/credits-and-billing): See where credits go and how to buy more