# Now Live: Agent Tables, a Spreadsheet Where Every Cell Can Run an Agent

- Canonical: https://limelit.co/blog/now-live-agent-tables-a-spreadsheet-where-every-cell-can-run-an-agent
- Published: 2026-09-05
- Publisher: Limelit (https://limelit.co)

> Agent tables are live in Limelit: a grid of pages, prompts or competitors where a column runs an agent over every row, with the credit cost estimated first and CSV export after.

Agent tables are live at /agents/tables. A table is a grid you already understand: rows are the things you care about (pages, prompts, competitors, sources), and a column can be bound to an agent that runs over every row. Fill a column, run a cell, run a whole column, or chain columns so one feeds the next. The credit cost is estimated before anything runs, and the result exports to CSV.

## Why a table

Most AI-search work is the same question asked across a list. Is this page cited? What does Claude say about each of these competitors? Which of these fifty prompts name us, and which name someone else? A dashboard answers the aggregate. A table answers it per row, and lets you sort, filter and act on the rows that matter.

The agent builder made single agents cheap to create. Tables are how you point one at a hundred things at once without running it a hundred times by hand.

## What it does

- **Three ways to start.** A blank grid, the batch planner (pick an agent and a segment, and the rows are filled for you), or a CSV import.
- **Agent columns.** Bind a column to any agent that consumes rows. The agent writes its output into the cell.
- **Per-cell, per-column and chained runs.** Run one cell to try it, a column to fill it, or a chain so a second column reads the first.
- **Estimates first.** The cost of a column or a chain is estimated in credits before it runs, and the spend cap applies.
- **Templates.** A gallery of ready-made tables sits under the library: competitor matrices, page audits, prompt sweeps.
- **Export.** Every table exports to CSV, so the result goes into the spreadsheet your team already uses.
- **Segments.** Hand a segment (a saved set of prompts or pages) to the planner and the rows are the segment.

## How to use it

1. Open /agents/tables and choose New table. Import a CSV of pages, or let the planner fill rows from a segment.
2. Add a column and bind it to an agent, for example the Page readiness audit.
3. Run one cell. If the output is what you want, run the column. Export when it is done.

The same operations are exposed through the free MCP server: create a table, add rows and columns, estimate a column or a chain, run a cell, a column or a chain, and export. An agent in Claude or Cursor can build and run a table for you and hand back the CSV.

## What it costs

Creating tables, adding rows and exporting are free. Running an agent column is metered from the model usage it causes, estimated before the run, and stopped by the spend cap. A column that only reads data Limelit already holds costs close to nothing; a column that drafts content costs what drafts cost.

## FAQ

**What can be a row?**
Anything with an identifier the agent understands: a page URL, a tracked prompt, a competitor, a source domain. The planner fills rows from a segment; the CSV import takes your own list.

**Can two columns depend on each other?**
Yes. A chain runs columns in order so a later column can read an earlier one. Estimate the chain before running it; the estimate covers every column.

**Is a table the same as a dashboard?**
No. Dashboards show aggregates. A table holds one row per thing and lets you act on each row. Use dashboards to see the trend and tables to work the list.