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

# The Palagus agent

> Ask questions about your project's data and about how to use Palagus, right inside the app.

The agent answers questions in plain language. It reads your project's data for you, and it knows how Palagus works. The agent is in beta.

## Open the agent

* Click **Agent** in the sidebar. The agent opens as a side panel next to the page you are on. Click **Agent** again to close it.
* To give it the whole window, click the expand icon (**Open as page**) in the panel's header. On that page, **Back to side panel** returns to the panel.

On a small screen, the panel fills the screen.

## Ask from the search

Press **Ctrl K** (**⌘K** on a Mac), or click the magnifier next to the project name at the top of the sidebar. A search opens over the page you are on.

* **Go to a page.** Type part of its name, for example `comp`, and press **Enter** to open **Competitors**. The search finds the pages your sidebar shows you.
* **Ask a question.** Type your question and choose **Ask the agent** below the pages: click it, or press **Enter** while it is marked (when no page matches, it already is). The question is sent at once.
* **Change it first.** Press **Tab** instead. The question goes into the agent's question field, and nothing is sent yet: change it if you like, then send it.

Either way, a chart attached to your question comes off. While an answer is still running, or when the agent cannot answer now (no plan, or the allowance used up), **Enter** does what **Tab** does: the question waits in the field, and where the agent cannot answer, the panel says why.

Where the agent is not offered, the search only finds pages.

<video autoPlay loop muted playsInline controls aria-label="Opening Competitors from the search, then asking the agent from it why visibility dropped" poster="https://media.palagus.ai/mcp/recordings/agent-search.poster.242ac2490251.webp">
  <source src="https://media.palagus.ai/mcp/recordings/agent-search.8d6890b88e83.webm" type="video/webm" />

  <source src="https://media.palagus.ai/mcp/recordings/agent-search.8d6890b88e83.mp4" type="video/mp4" />
</video>

Typing `comp` and pressing **Enter** opens **Competitors**. Back on the Overview, **Last 24 hours** is chosen. The question "Why did our visibility drop?" goes to the agent with **Enter**, and the answer covers those 24 hours.

## Who can use it

* The agent is included in every plan you can buy. If your workspace's plan does not include it, **Agent** does not appear in the sidebar.
* Workspace owners can use it, in the projects of their own workspace. A role in an agency does not count in its clients' workspaces: there, the owner of the client's workspace can use it.
* It answers only about projects in your own workspaces. If you are viewing another workspace's project, switch back to your own workspace to ask.
* **Without a plan**, your conversations stay readable, but you cannot ask new questions. The note in place of the question field leads the workspace owner to **Billing**. See [Without a plan](https://docs.palagus.ai/without-a-plan.md).
* **If your results are locked**, because the workspace never had a plan beyond its free trial, its conversations are locked with them. They are kept, and they open again once the workspace has a plan.

## What you can ask

**About your data.** For example:

* "How visible is my brand right now?"
* "What changed last week?"
* "Who are my strongest competitors?"
* "Which pages are cited most often?"
* "Show me an answer where my brand comes off badly."
* "Since when has my visibility been falling?"

**Suggestions.** In a new conversation, the empty question field shows questions that fit the page you are on, one after the other: on the Overview's **Sources** tab, for example, which sources AI answers cite most. Press **Tab** to take the one shown, then send it or change it first. Once the conversation has a question, the field shows none.

The agent reads the same data you see in Palagus: visibility, share of voice, sentiment, position, competitors, prompts, topics, personas, cited sources and the AI engines' answers themselves.

<video autoPlay loop muted playsInline controls aria-label="Taking a suggestion on the Sources tab and asking the agent which sources AI answers cite most" poster="https://media.palagus.ai/mcp/recordings/agent-ask.poster.e72ba64a076f.webp">
  <source src="https://media.palagus.ai/mcp/recordings/agent-ask.b8e03e2f35f5.webm" type="video/webm" />

  <source src="https://media.palagus.ai/mcp/recordings/agent-ask.b8e03e2f35f5.mp4" type="video/mp4" />
</video>

On the Overview's **Sources** tab, **Tab** takes the suggestion "Which sources do AI answers cite most?". The answer names the time window first and then ranks the most-cited domains. Clicking the summary above it shows what the agent read.

**About Palagus.** For example:

* "Where do I find our invoices?"
* "Where do I find the suggested competitors I'm not tracking yet?"
* "How is share of voice calculated?"
* "How do I invite a colleague?"

The agent looks the answer up in this documentation and in the app's own pages. A place in the app comes as a link, so you can go straight there.

<video autoPlay loop muted playsInline controls aria-label="Asking the agent where the suggested competitors are" poster="https://media.palagus.ai/mcp/recordings/agent-help.poster.ea72a15a9bc6.webp">
  <source src="https://media.palagus.ai/mcp/recordings/agent-help.e071fe255518.webm" type="video/webm" />

  <source src="https://media.palagus.ai/mcp/recordings/agent-help.e071fe255518.mp4" type="video/mp4" />
</video>

The link in the answer opens the Suggested tab of Competitors, and the agent stays open beside it.

## Recommendations

A new conversation opens with **How can I help you today?** and up to three recommendations, picked for the project the agent is working in:

* **No prompts or no competitors yet:** a question that asks where to add them. The answer links the page where you do it yourself.
* **AI answers name brands you don't track yet:** a question that lists them.
* **Once the project has prompts:** the skills **Weekly pulse** and, with competitors, **Where you lose**.

A question is sent as soon as you click it. A skill opens its form first, as from **Skills**. A recommendation never adds or changes anything for you.

## Skills

A skill is a ready-made analysis you start with one click instead of typing a question. The agent then works through a fixed checklist: which data to read, in which order, and how to write it up. These are the same analyses as the [reports](https://docs.palagus.ai/reports.md), run in your conversation.

**Start a skill.**

1. Click **Skills**. In a new conversation the button is in the middle of the panel. At any time, you can also type `/` in the question field, or click **+** next to it and choose **Skills** (see [Add to your question](#add-to-your-question)).
2. Choose a skill from the list. The skills are grouped:
   * **Status**: Weekly pulse.
   * **Diagnosis**: Drop diagnosis, Where you lose.
   * **Sources**: Source authority.
   * **Track**: Campaign tracker.
3. A small form opens in the conversation. Fill in what the skill needs, for example the time window, or for Campaign tracker the day your campaign started. Fields marked \* are required.
4. Click **Run skill**.

The skill always runs on the project the agent is working in. Your question appears as the skill's name and what you chose, for example "Weekly pulse · Last 7 days".

**Follow its steps.** Above the answer, the agent shows the skill's steps, and under them one line that says what it is doing right now, such as "Reading the prompts…". A step gets a tick only when the agent has actually read the data for it. Click a step with an arrow to see what was read for it. Each step also says, in words, how it went:

| Step says | Meaning |
| :- | :- |
| (tick) | Done: the data for this step was read. |
| not needed | The skill only needs this step in some cases, and not in this one. |
| skipped | The answer was written without this step. |
| refused | The agent tried to read the data for this step, and the request was refused. |
| not reached | The answer stopped before this step: you clicked **Stop**, the time limit was reached, or an error ended it. |

When the answer is finished, the steps fold up into one line, such as "4 of 5 done · 1 skipped". Steps that did not go as planned stay visible. **All steps** shows the rest.

**Ask on.** After a skill you can ask follow-up questions in the same conversation, like after any answer. **Retry** runs the skill again with the same choices.

**Skill or report?** A skill runs in the agent: it counts against the agent's monthly allowance, and you can ask follow-up questions. A report on the [Reports](https://docs.palagus.ai/reports.md) page is kept as a document of its own. Skills answer in English, unless you have already asked a question in your own language in the same conversation.

If two of your projects have the same name, a skill cannot tell them apart. Palagus then asks you to rename one of them first.

## Add to your question

Next to the question field, **+** opens a short menu: **Skills** and **Charts on this page**. Two signs typed into the question field open the same lists.

**`/` for a skill.** Type `/` at the very start of the question field. The list of skills opens. Type on to narrow it by name: `/week` leaves Weekly pulse. A `/` anywhere else in your question is just a character.

**`@` for a chart.** Type `@` at the start of the question field or after a space. The list shows the charts on the page you are looking at that you can ask the agent about: the ones with an **Ask AI** button. Type on to narrow it: `@topic` leaves Topic Visibility. On a page without such a chart, the list says "Nothing on this page to attach". An `@` inside a word, as in an e-mail address, is just a character.

A chart you choose appears as a small tag above the question field. What you typed stays; only the `@` and what you typed after it are removed. The chart goes with your next question: the agent reads that chart's figures itself, for the same time window, filters and settings, and answers about it. Click the **×** on the tag to take the chart off. One chart goes with a question; choosing another one replaces it. The tag is gone once the question is sent, or when you open another page.

**Ask AI on a chart** asks from the chart's side. It opens a small card next to the chart. Suggestions for that chart type themselves into the card's empty field: press **Tab** to take the one shown, or type your own question. Press **Enter**, or the arrow, to ask. The agent's panel opens and answers there, about that chart, for the same time window, filters and settings; the card only asks. While the agent is still answering, the card cannot send and says so. **Esc** closes the card.

When the agent cannot answer now (no plan, or the allowance used up), the question waits in the agent's question field with the chart attached, and the panel says why. If you take the chart off with **×** before you change that question, the question goes too; once you have changed or added to it, your text stays.

**With the keyboard.** While a list is open:

| Key | What it does |
| :- | :- |
| ↑ ↓ | Move through the list. |
| Enter | Take the marked entry: a skill opens its form, a chart is attached. Nothing is sent yet. |
| Esc | Close the list. What you typed stays, and the next Enter sends it as a question. |

If no skill matches what you typed after `/`, Enter sends it as a question.

## What the agent does not do

* **It changes nothing.** It cannot add or delete prompts, topics, personas or competitors, change settings or start runs. If you ask it to, it says so and tells you where in the app you can do it yourself.
* **It answers only about your Palagus data and about Palagus.** It declines general questions, writing tasks and questions about other companies' data.

## How it answers

**What it is doing.** While the agent answers, one line above the answer says what it is doing right now: "Thinking…", what it is reading, such as "Reading the prompts…", and then "Writing the answer…". When the answer is finished, the line folds up into a short summary of what the agent checked and how long it took, such as "Checked the overview, the timeline and 10 prompts · 6 s". Click it to see each step and what it read. A step that was refused is marked as refused.

**Its plan.** When answering takes more than one look at your data, the agent shows its plan above the answer: the steps it will take, such as "See where the project stands" and "Follow visibility over the week", and one sentence on what it found so far and what it checks now. The plan appears as the agent starts its first look, so you see the steps before they run. Palagus ticks each step once the agent has really read that data; the agent cannot tick a step itself. When a result calls for another look, new steps are added below the earlier ones. A look the plan did not name gets a step of its own from Palagus, named for what was read, such as "Read the overview". If the agent made no plan, Palagus lists the looks itself from the second one on, without a sentence. Looking something up in this help is no look at your data and gets no step. When the answer is finished, the plan folds into one line, such as "Plan done 2/2". Click it to see the steps; click "2 steps" under a sentence to see what the agent read for it. When an answer could not be finished, the line says "Plan unfinished" and why, such as "The answer could not be finished.", and the steps it did not reach are struck through. A plan takes the place of the summary of what the agent checked. A question that one look answers usually has no plan; if the agent planned more, the line counts only the steps it did, such as "Plan done 1/2". A skill shows its own checklist instead.

**Time window.** When an answer contains figures, it starts by saying which time window they cover.

**The page you are on.** The agent uses the project, time range and filters of the page you are looking at, unless your question says otherwise. On a prompt's page, "How is this prompt doing?" is about that prompt. You don't need to say which project you mean.

<video autoPlay loop muted playsInline controls aria-label="Asking the agent, on a prompt's page, for an answer where the brand comes off badly" poster="https://media.palagus.ai/mcp/recordings/agent-why-drop.poster.da9d7328dd43.webp">
  <source src="https://media.palagus.ai/mcp/recordings/agent-why-drop.fad5f6648e68.webm" type="video/webm" />

  <source src="https://media.palagus.ai/mcp/recordings/agent-why-drop.fad5f6648e68.mp4" type="video/mp4" />
</video>

On a prompt's page, "Show me an answer where we come off badly here" is about that prompt, without naming it. The agent quotes an AI engine's answer to it and what that answer says about the brand.

**Language.** The agent answers in the language of your question, even when your prompts or the AI engines' answers are in another language. Names of prompts, topics, personas and brands stay exactly as they are in your project, untranslated.

**Cited sources.** "Which pages are cited?" is answered with a ranking of every citation in the time window, not with a few single answers. If you ask about one topic, persona, country, AI engine or prompt, the ranking counts only the answers for it.

**Next steps.** Asked what to do next, the agent answers with a short numbered list of steps you take yourself in the app, each linked to its page where there is one — it only reads your data and never carries out a step for you.

**Charts.** When you ask for a chart, or ask how something developed over time or how it differs across AI models, personas or topics, the agent can show a chart below its answer.

* A line shows visibility, position or sentiment over time. It shows the first five brands named in the time window, highest first, the same brands as the ranking in the app. Your project is among them only when it is one of those five. Brands never named are left out. "+3 more not shown" counts the named brands left out.
* Bars show your project's visibility per AI model, per persona or per topic.
* Below the chart is the time window it covers.

The agent draws the chart from the data it read, never from figures it writes itself. The chart stays in the conversation when you come back to it. When you open the summary above the answer, its step shows as "Showed a chart". A skill's answer has no chart.

**Questions back.** When your question can mean several things — "Compare us with our competitor" in a project that tracks several — the agent can ask back which one you mean, with two to five options below its question. Click one to send it right away, as if you had typed it, or type your own answer instead. The option you chose stays ticked in the conversation. The agent only asks back when the data cannot settle it: "our strongest competitor" it finds itself. An answer that asks back has no suggestions under it, and a skill never asks back. When you open the summary above the answer, its step shows as "Asked you to choose".

**Ask next.** Under the newest answer, the agent can suggest two or three questions to ask next, under **Ask next**. Click one to send it right away, as if you had typed it. The first one also stands grey in the empty question field: press **Tab** to put it there, edit it if you like, and send it. The suggestions appear once the answer is done, only after an answer that read your project's data, and never after a skill. When you ask the next question, they move to the newest answer. Suggestions count towards the monthly allowance as part of the answer.

**Links.** An answer only links into the Palagus app and into this documentation. A link into the app opens in the same tab, in the project you are working in. A documentation link opens in a new tab. Any other link is shown as plain text.

## Conversations

* Each project has its own conversations. Click the list icon (**Conversations**) in the panel's header to see them. On the full page, the list stands beside the conversation; on a narrow screen, such as a phone, it opens from the same icon above the conversation. You can search them by the questions you asked.
* Click the pen icon (**New conversation**) to start over. On the full page, it is a button with that name; on a phone, the pen icon.
* When you reload the page, your last conversation in that project opens again.
* You can delete a conversation. This cannot be undone. Deleting a conversation does not give back any of the monthly allowance.
* A conversation is deleted automatically 90 days after its last message.

## While it answers

* Click **Stop** to end an answer early.
* **Retry**, under the last answer, asks your last question again, or runs your last skill again.
* If you close the tab or reload while it answers, the agent keeps working. When you come back, you see "The agent is still answering…" and then the answer.
* An answer that was cut short says so: "Stopped: the answer could not be finished" or "Stopped: the time limit was reached".
* If you ask many questions in a short time, the agent asks you to wait a moment before the next one.

## Notifications

Your browser can tell you when an answer is ready, so you can work in another tab while the agent answers.

* Under the first answer of a conversation, the agent offers **Get notified when I reply**. Click **Turn on**, then allow notifications when your browser asks.
* From then on, when an answer is ready while the Palagus tab is in the background, your browser shows a notification. Click it to go back to the answer.
* When the agent asks you to choose between options, the notification says it needs your choice.
* There is no notification while you look at the Palagus tab, and none for an answer you stopped yourself.
* There is no notification if you reloaded or closed the tab while the agent was answering. The answer is there when you come back.
* If you blocked notifications for Palagus, there is none either. To change this, use your browser's site settings for Palagus.
* **Not now** hides the offer in this browser. You can still allow notifications later in your browser's site settings.
* Some browsers cannot show these notifications: Safari on an iPhone or iPad, and browsers on Android. There, the offer does not appear.

## Monthly allowance

Each workspace has a monthly allowance for the agent, counted in agent credits. It is included in every plan you can buy, as a fair-use limit.

| Plan | Agent credits per month |
| - | - |
| The plans you can buy (prompts, AI models, how often) | 150,000 |
| Pro (earlier fixed plan) | 30,000 |
| Max (earlier fixed plan) | 150,000 |

A typical answer uses about 10–20 credits. Questions to ask next and skills use more.

* Once 10 % or less is left, a note above the question field says how many credits are left of the month's total and the day it starts again. Until then the panel says nothing about usage. If you may change the plan, **Choose a plan** in the note opens the plans.
* When the credits are used up, the question field is replaced by a note with the date you can ask again.
* The allowance starts again on the first day of each month (UTC).
* Your conversations stay readable when the allowance is used up.
* **Workspace Settings** → **Billing** shows the credits used this month by day, by project and by member, and earlier months under **Month**. A deleted project or account keeps its line as **Deleted project** or **Deleted account**.

## Use it outside Palagus

The agent reads your data with the same tools as the [Palagus MCP server](https://docs.palagus.ai/the-palagus-mcp-server.md). Connect your own AI assistant to that server, and you can ask the same questions there and build the same analyses into your own work. [Setup](https://docs.palagus.ai/setup.md) shows how to connect it.

* **The same tools for reading.** The agent uses all twelve reading tools of the MCP server, including the two that search and read this documentation. [Tools](https://docs.palagus.ai/tools.md) lists them.
* **More tools in your assistant.** Through the MCP server, your assistant can also add prompts and topics and start runs, if you allow it when you connect. The agent in Palagus has none of these.
* **Your assistant's own rules.** How the agent answers, as described on this page, applies in Palagus. Your assistant decides for itself how it answers.
* **Not part of the allowance.** Questions you ask through your own assistant do not count against the agent's monthly allowance.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.