# Prompt on pages

> Ask your own model one question about many pages of the open crawl. Each page's answer becomes a value in a column of the Custom tab and is saved with the crawl. One model call per page, and nothing runs until you press the button.

Describes GLLmolfar 0.9.35, checked 2026-11-08. Canonical page: https://www.getlooploop.com/gllmolfar/docs/prompt-column

- **Where:** Agent · test › Prompt on pages
- **Mode:** Simple and Pro
- **Built from:** The open crawl's pages and a model you connected in Settings › AI models
- **Changes:** Only when you press the button

## On the screen

A form: the column's name, the question, which pages, how many at most, and the number of model calls the run would make. The result of the last run shows below it. In the picture no model is connected yet, so the notice above the form says where to connect one, and nothing has run.

![Agent · test › Prompt on pages in GLLmolfar, filled in before any model is called, on a crawl of getlooploop.com](https://www.getlooploop.com/images/gllmolfar/docs/prompt-column.webp)

Numbered parts of the screenshot:

1. **Sections of the agent.** **Prompt on pages** is one of five: Analysis and tasks, Content, Social posts, Prompt on pages, Risks.
2. **No model connected yet.** The agent works on the person's own models. **Open AI models** goes to Settings › AI models to connect one.
3. **Column name.** Here **Audience**. It becomes the column header in the Custom tab, with **· AI** after it.
4. **Question for each page.** Asked about each page with the page's text; {url}, {title}, {h1} and {meta} are filled in.
5. **Pages and At most.** **All indexable HTML pages**, or the pages of one check; at most 50 in this run.
6. **Skip, Language.** Leave out pages that already have an answer in this column; the language the answers are written in.
7. **Calls before the start.** **50 pages, one model call each.** 450 more are left for the next run. The count is shown before anything runs.
8. **The button.** **Ask about 50 pages** starts the run. Nothing is sent to a model before it is pressed.

## Fields

| Field | What it does |
| --- | --- |
| Column name | Up to 60 characters. The column is shown in the Custom tab as **{name} · AI**. Questions you ran before are listed above the form under **Your columns:**; a click fills the form again. |
| Question for each page | The page's text always goes with the question. `{url}`, `{title}`, `{h1}` and `{meta}` in the question are filled in. Answers are cut at 300 characters. |
| Pages | **All indexable HTML pages**, or the pages of one check from the fix list, with its count. |
| At most | Pages in this run: 50 by default, 500 at most. The rest are counted: **N more pages are left for the next run.** |
| Skip pages that already have an answer in this column | On by default: a second run fills only the pages still empty, and says **N already answered, skipped.** |
| Language | The language the answers are written in. |

## What a run does

- One call per page, to the model chosen for writing in Settings › AI models: Ollama or LM Studio on this computer, or a provider you connected with your own key. Molfar has no model of its own here.
- The page's text is read from the crawl's saved HTML (after JavaScript when the page was rendered).
- A page the model cannot answer, or that cannot be read, is recorded with its reason, and the run goes on.
- The answers go into the open crawl's pages and are saved with it. **Open in the Custom tab** shows them as a column you can sort, filter (**{name}: found**, **{name}: not found**) and export.
- A cloud provider bills your own key. The window shows which models are local and which are in the cloud; Molfar shows no prices.

> **From an agent.** The MCP tool `run_prompt_column` takes the same name, question, check, page limit and language. Called without `confirm`, it runs nothing and answers with the number of model calls; it runs only when called again with `confirm: true`, after the person agreed. See [MCP tools](https://www.getlooploop.com/gllmolfar/docs/mcp-tools).

## States and messages

- **No crawl** — the app says: "Open a finished crawl first: the plan is built from its fix list." The form needs a finished crawl to count pages.
- **No model** — the app says: "No model connected yet" Connect a model in Settings › AI models; **Open AI models** goes there.
- **Result** — the app says: "{ok} answered · {bad} failed" Shown above the first 50 answers of the last run; **N more answers are in the Custom tab.**

## Related pages

- [Agent · test](https://www.getlooploop.com/gllmolfar/docs/agent-test.md): The Agent · test screen: your own models plan tasks from the fix list, rewrite pages, write social posts and review risky pages; GLLmolfar measures the results.
- [Custom extraction and search](https://www.getlooploop.com/gllmolfar/docs/custom-extraction.md): Crawl › Custom extraction & search: up to ten rules (CSS, XPath subset, regex, HTML contains) that each add a column to the Custom tab.
- [Settings › AI models](https://www.getlooploop.com/gllmolfar/docs/settings-ai-models.md): Settings › AI models: connect your own language models, on this computer, on your network or in a cloud, and choose which model does which analysis.
- [MCP tools for agents](https://www.getlooploop.com/gllmolfar/docs/mcp-tools.md): The MCP server inside GLLmolfar: how to connect an AI agent, and every tool it can call to crawl, read fixes, audit pages, search by meaning and save reports.