# Now Live: The Agent Builder, Describe It, Review It, Approve What Ships

- Canonical: https://limelit.co/blog/now-live-the-agent-builder-describe-it-review-it-approve-what-ships
- Published: 2026-09-05
- Publisher: Limelit (https://limelit.co)

> The Limelit agent builder is live: describe an agent in a sentence, get it assembled from tested parts, see the credit cost before each run, and approve everything it publishes.

The agent builder is live at /agents for every Limelit account. You describe the agent you want in a sentence. Limelit assembles it from tested parts, shows you the draft to edit, and runs it on a schedule or on demand. Every run shows its credit cost before it starts, and nothing an agent produces is published without a person approving it.

## Why a builder and not a chat box

A chat agent answers a question. A marketing team needs the same question answered every Monday, with the result filed where the team looks and the follow-up drafted. That is an agent with a schedule, inputs, outputs and an approval step, and it is easier to build from parts that already work than to prompt from scratch each week.

So the builder starts from a library of capabilities Limelit already runs in production, and the describe-to-build line maps your sentence onto them. What you get is editable, not a black box.

## What it does

- **Describe to build.** Type what you want ("watch our category in Claude and draft a fix when a competitor is cited instead of us"). The builder proposes an agent from the library, and you review the proposal before it exists.
- **Templates.** Five ready to run: Weekly KPI pulse, Competitor matrix check, Perception pulse, Citation gap sweep, and Page readiness audit. Each opens as an editable draft, never as a live agent you did not read.
- **New agent recipes.** One capability at a time, constrained on purpose, so an agent does one job well.
- **Schedules and runs.** Run once or on a schedule. Every run keeps its outputs, and the credit cost is shown before it starts.
- **Approvals.** Work that would publish, such as a draft or a page change, lands in Approvals for a person to decide.
- **Workflows.** Curated prompt templates for driving Limelit from Claude, Cursor or any MCP client, absorbed from the old agent hub.
- **Agent tables.** Run any row-consuming agent over a spreadsheet of rows: a list of pages, prompts or competitors. Create a blank grid, plan a batch, or import a CSV.

## How to use it

1. Open /agents and type the agent you want in the build line, or pick a template.
2. Read the draft: its inputs, what it will produce, and where the output goes. Edit anything.
3. Run it once. Check the output and the credits it used. Then schedule it.

If you already drive Limelit from your own agent, the same parts are exposed through the free MCP server: create an agent, run it, read its ports and outputs, decide an approval.

## What it costs

Building and editing agents is free. Running one is metered from the model usage it causes, shown as an estimate before each run, and a spend cap stops Limelit before it is exceeded. The Weekly KPI pulse reads data Limelit already has, so it is close to free; a citation gap sweep that drafts a fix costs what the draft costs, roughly 550 credits.

## FAQ

**Can an agent publish on its own?**
No. Anything that would change a public page or send a message goes to Approvals. A person decides.

**Can I request an agent the builder cannot make?**
Yes. "Request a bespoke agent" on the /agents page sends the description to us, and we build it or tell you why not.

**Do I need to know how the parts work?**
No. The draft explains each part in plain language, and the templates are the same parts pre-assembled. Read the draft before the first run; that is the whole safety model.