---
title: "Use Pioneer AI"
description: "Ask Pioneer AI about the page you are on, follow its work, reuse saved prompts, pop a conversation out, save it to a ticket, and rate replies."
url: "https://support.outpostplatform.com/pioneer-ai/use-pioneer-ai/"
product: "depot"
type: "how-to"
status: "draft"
ai_summary: "Using Pioneer AI: open with the header button or Alt+P, send with Enter, / for saved prompts, page context, tool, Thinking, navigate and launch cards, tables and charts, remote session control, the context meter, Stop and Continue, history (rename, delete, New chat), the pop-out window, Settings > AI > Prompts, Save to ticket with time, Helpful / Not helpful, and the chat's error messages."
availability:
  plans: ["home-lab", "essentials", "enterprise", "msp"]
  permission: "ai:use"
  preview: false
source: "https://gitlab.com/outpostplatform/docs/-/edit/main/src/content/docs/pioneer-ai/use-pioneer-ai.md"
license: "CC BY 4.0"
---

# Use Pioneer AI

Pioneer AI answers in a drawer beside whatever console page you are on, so
you can investigate an endpoint, an alert, or a ticket without leaving it.
This page shows how to hold a conversation, follow what Pioneer AI does, come
back to a conversation later, and turn one into a ticket note.

## Before you begin

- You need the `ai:use` permission. Every role has it by default.
- Pioneer AI must be on for your organization, and on the MSP plan for the
  client you have selected. See [Set up Pioneer AI](/pioneer-ai/set-up/).
- Pioneer AI reads what your permissions allow and sends it to the model
  provider your organization chose. Do not paste passwords, keys, or other
  secrets into a message. See
  [Pioneer AI data and privacy](/pioneer-ai/data-and-privacy/).

## Ask a question

1. In the header, select **Open Pioneer AI**, or press <kbd>Alt</kbd>+<kbd>P</kbd>
   anywhere in the console. The button is shown only on wide screens. The
   shortcut works at any width.
2. Enter your question in the message box. Press <kbd>Enter</kbd> or select
   **Send**. Press <kbd>Shift</kbd>+<kbd>Enter</kbd> to start a new line.
3. Read the reply as it arrives under **Pioneer AI**. **Working...** shows
   until the first part comes in.
4. Ask a follow-up in the same box. Pioneer AI has the whole conversation, so
   a follow-up can refer to an earlier answer.

Each message also tells Pioneer AI which page you are on: its address, its
title, and the endpoint, ticket, alert, site, incident, script, or client it
shows. On an endpoint's page, "why is this one offline?" is enough.

The first reply gives the conversation a title, which replaces **Pioneer AI**
as the drawer's heading.

## Follow what Pioneer AI does

A reply can contain more than text. Each part appears in the order Pioneer AI
produced it:

| Part | What it shows |
| --- | --- |
| A tool card | The tool's name, a chip for its tier (**Read**, **Change**, **High impact**, or **Display**), and **Running...** until it finishes or **Failed** if it failed. Expand it to see the input and the result. |
| **Thinking** | The model's reasoning, when the model provides it. Collapsed by default, and never saved. |
| An approval card | **Pioneer AI wants to take an action**, with **Approve** and **Deny**. See [Approve Pioneer AI actions](/pioneer-ai/approve-actions/). |
| **Opened** and a page address | Pioneer AI opened a console page for you. **Go there** opens it again. |
| **Remote desktop to** an endpoint | A card with **Connect**. Selecting it starts remote desktop through the usual session flow. |
| A shell on an endpoint | A card such as **PowerShell on** an endpoint, with **Open shell**. Selecting it opens a shell through the usual session flow. |
| **Controlling** an endpoint | Pioneer AI is operating the screen of a remote session you have open, with the step, the last action, the time left, and **Stop**. See [Control a remote session](#control-a-remote-session). |
| A table or a chart | Results Pioneer AI arranged for you, with a title. |
| **Plan** | A checklist of the steps Pioneer AI intends to take on a larger task, such as **Plan (2 of 5 done)**, updated as it works. |
| **Sub-agent** | A part of the task Pioneer AI handed to a sub-agent, with its goal and status. See [Sub-agents](#sub-agents). |
| **Sources** | The web pages an answer used, under the answer. See [Pioneer AI web search](/pioneer-ai/web-search/). |
| **Save this as a memory?** | A note Pioneer AI offers to keep for later conversations. See [Manage Pioneer AI memories](/pioneer-ai/memories/). |

Launch cards never connect on their own. Your permissions and your
[remote access policy](/remote-access/policy-settings/) apply when you select one, as if you
had started the session from the endpoint's page.

Under the conversation, the meter shows how much of the model's context the
conversation uses, such as **Context 12% (16k of 131k tokens)**, the tokens
used so far, and the cost so far when there is one. When a conversation
outgrows the model's context, Pioneer AI summarizes the earlier messages and
says **Earlier messages were summarized to keep this conversation within the
model's memory**.

To stop a reply, select **Stop**. The reply keeps what had arrived and shows
**Stopped.** An action already running finishes first.

Pioneer AI can read several things at once, and it keeps large results, such
as a list of a thousand endpoints, inside the conversation so it can filter
them and show you a full table without rereading them. You do not need to do
anything for this.

### Spending limit per reply

An owner sets the most one reply may cost, $0.50 by default. Model calls,
sub-agents, and web searches all count. When a reply reaches it, Pioneer AI
summarizes what it found and what is left, and stops with **Pioneer AI stopped
because this reply reached the spending limit an owner set for one reply.**
Select **Continue** to carry on with a fresh limit. See
[Set a spending limit per reply](/pioneer-ai/set-up/#set-a-spending-limit-per-reply).

## Sub-agents

For a large task, such as checking the same thing on several sites, Pioneer AI
can hand up to 3 parts to sub-agents that work at the same time and report
back. Each appears as a **Sub-agent** card with its goal and a status:
**Working**, **Done**, **Failed**, or **Hit its limit**. Expand it to see the
**Model calls**, **Input tokens**, and **Cost**.

- A sub-agent works only in the areas an owner allows. By default it can read
  Outpost data, research (web search, documentation, and memories), and draft
  scripts. See [Choose what sub-agents may do](/pioneer-ai/set-up/#choose-what-sub-agents-may-do).
- A change a sub-agent wants still needs your approval. Its approval card has
  a **Sub-agent** chip and says **Asked by a sub-agent working on:** followed
  by its goal. If you do not answer within 10 minutes, or you leave, the
  sub-agent is told the action was not approved and carries on without it.
- A sub-agent can never take a high-impact action, cannot talk to you, and
  stops after 8 model calls.

## Background checks

When you start something that takes a while, such as a script batch on many
endpoints, a command, or waiting for an endpoint to come back online, Pioneer
AI can keep an eye on it after the reply ends. It adds a message to the
conversation when the job finishes or the wait runs out, for example
**Background check: script batch** with how many endpoints succeeded and
failed, and a link to the results.

You do not need to keep the conversation open. When a background check
reports, the notifications bell in the header shows **Pioneer AI results**
with a count. Select it to open your conversations. A check waits up to 24
hours, or 2 hours for an endpoint to come online. A conversation can have up
to 5 checks running, and you up to 20.

## Control a remote session

When a task needs the endpoint's screen, such as a setting with no command
line, Pioneer AI can see the screen and click, type, and scroll in a remote
desktop session you have open. Ask from the **Pioneer AI** drawer inside the
remote viewer, or from this chat while the viewer is open in the same browser.

- You approve each task once: the endpoint, the goal, and a budget, 40
  actions and 10 minutes by default. Automatic approval never applies.
- The viewer's status bar reads **Pioneer AI is controlling this session**,
  with **Stop**. A click, key press, or scroll of your own also stops it.
- Each step sends a screenshot of the session to the AI provider. Outpost
  does not store screenshots.
- It needs a model that can read screens, the endpoint's
  `remote_access.ai_control` policy setting, and the `remote:ai_control`
  permission.

See [Let Pioneer AI control a session](/remote-access/pioneer-control/).

## Undo an action

Some changes Pioneer AI made can be reversed exactly, such as removing
endpoints from a device group, creating a tag, or assigning a profile.

1. In the drawer header, select **Actions in this conversation**. The panel
   lists what Pioneer AI ran, newest first.
2. On the action, select **Undo**. The dialog **Undo this action?** says what
   undoing will do, after **This will:**.
3. Select **Undo**.

An undo is a change of its own. A routine one runs when you select **Undo**. A
high-impact one, such as adding endpoints back to a device group, shows
**Undoing this is a change of its own, so it needs approval first.** and an
approval card, which you answer as for any other action. The action then shows
**Undone**.

**Undo** appears only where Outpost knows exactly how to reverse the change.
Actions that cannot be reversed with certainty, such as running a script,
have no **Undo**.

## Come back to a conversation

Conversations are saved, so closing the drawer does not lose one. Opening the
drawer again shows the conversation you had open.

- **See your conversations.** Select **Your conversations** in the drawer
  header. The list shows each title and its last activity. Select one to open
  it. **Show more** loads older ones.
- **Start another.** Select **New chat**.
- **Rename or delete one.** In the list, select **Rename** or **Delete**. A
  deleted conversation cannot be recovered.

If you close the drawer while a reply is being written, Pioneer AI finishes
the action in progress, saves the conversation, and ends the reply with **The
reply was interrupted before it finished.** Select **Continue** to let it carry
on.

On the MSP plan, a conversation belongs to the client that was selected when
it started. Selecting a different client starts a new conversation.

Conversations are deleted automatically after the retention period an owner
sets. See [Pioneer AI data and privacy](/pioneer-ai/data-and-privacy/).

## Open a conversation in its own window

1. Send at least one message, and wait for the reply to finish.
2. In the drawer header, select **Open in its own window**.

The conversation opens in a separate browser window and the drawer closes.
When Pioneer AI opens a page from that window, the page opens in your console
window. If no console window is open, the card offers **Open in console**.

If nothing opens, your browser blocked the window. Allow pop-ups for your
console's address and try again.

## Use a saved prompt

A saved prompt is text you send often, such as a triage checklist, stored
under a short command.

1. In an empty message box, type `/`. The list of saved prompts opens.
2. Keep typing to narrow it, then choose a prompt with the arrow keys and
   <kbd>Enter</kbd>, or select it.
3. The prompt's text fills the message box. Edit it if you need to, then send
   it.

To add one:

1. Go to **Settings** > **AI** > **Prompts** and select **New prompt**.
2. In **Command**, enter the word that follows the `/`, such as `triage`. Use
   1 to 32 lowercase letters, digits, or hyphens.
3. In **Name**, describe it in up to 80 characters.
4. In **Prompt**, enter the text, up to 8,000 characters.
5. Optional: select **Share with everyone in the organization**. Sharing needs
   `ai:manage`.
6. Select **Save**.

Personal prompts are yours alone. Shared prompts are offered to everyone, and
only people with `ai:manage` can edit or delete them. Saved prompts are also
offered to AI apps you connect with MCP. See
[Saved prompts in MCP](/pioneer-ai/mcp-server/#saved-prompts).

## Save a conversation to a ticket

Pioneer AI can write a ticket-ready internal note from the conversation: what
it found, the actions taken and their outcomes, and the next steps.

1. In the drawer header, select **Save to ticket**.
2. In **Ticket**, search by subject words, a ticket reference, or the
   requester's email address, and select the ticket. On an MSP's practice
   console with no client selected, the search covers every client.
3. Read the **Internal note** Pioneer AI wrote, and edit it. It is saved as an
   internal note only, so the requester does not see it.
4. Optional: select **Also log time**, enter **Minutes** (1 to 1440), and
   select **Billable** if the time is billable.
5. Select **Save**.

The dialog confirms with **Saved to** and the ticket reference. If the note
was saved and the time entry was not, the message says so. Log the time on
the ticket.

## Rate a reply

Under each finished reply, select **Helpful** or **Not helpful**. Your rating
is saved with the conversation. Select the other one to change it.

## Verify

After you send a message, a reply appears under **Pioneer AI** with no error
banner beneath it.

## If something goes wrong

| What the chat shows | Cause | Fix |
| --- | --- | --- |
| **Pioneer AI is turned off.** | Chat is off on **Settings** > **AI**, or no source is usable, such as a Remote Outpost with no endpoint saved. | Ask someone with `ai:manage` to turn it on. See [Set up Pioneer AI](/pioneer-ai/set-up/). |
| **Pioneer AI is turned off for this client.** | On the MSP plan, the client policy or the client's override turns it off for the selected client. | Ask someone with `ai:manage` to change the client's override. |
| **A client this conversation used has AI access turned off by your organization's policy.** | The conversation worked with a client that has since been turned off. | Start a new conversation. |
| **You don't have the ai:use permission, so Pioneer AI is not available to you.** | Your role lacks `ai:use`. | Ask an owner to add it to your role. |
| **Pioneer is still working in this conversation, in this window or another.** | A reply is still being written, perhaps in another window. | Wait for that reply to finish. |
| **This month's Pioneer AI budget has been reached. Raise it in Settings to keep using Pioneer AI.** | Outpost Cloud AI reached the monthly budget. | Ask someone with `ai:manage` to raise or clear the budget. |
| **This month's AI budget** (your personal budget or the organization budget) **is used up.** | With your own endpoint, a budget was reached. The message says when it resets. | Ask someone with `ai:manage` to raise it. See [Set budgets for your own endpoint](/pioneer-ai/set-up/#set-budgets-for-your-own-endpoint). |
| **Your plan's free Pioneer AI allowance for this month is used up.** | On Home Lab, this month's free Outpost Cloud AI allowance is used. The message says when it resets. | Wait for the reset, or use your own endpoint. |
| **This trial has reached its Pioneer AI limit. Activate your subscription to keep using it.** | The free trial reached its Outpost Cloud AI limit. | Activate your subscription. See [Free trials](/account/free-trials/). |
| **Too many Pioneer AI requests. Wait a minute and try again.** | You sent more than 20 messages in a minute. | Wait a minute. |
| **The message is too long to send.** | One message is longer than 32,000 characters. | Shorten it, or split it across messages. |
| **Pioneer AI stopped after the most steps one reply may take.** | A reply took 25 model steps. | Ask it to continue. |
| **Pioneer AI stopped because this reply used as much of the model as one reply may.** | A reply reached its token limit. | Ask it to continue. |
| **Pioneer AI stopped because this reply reached the spending limit an owner set for one reply.** | The reply reached the spending limit per reply. | Select **Continue**, or ask someone with `ai:manage` to raise the limit. |
| **This action was already undone.** | Someone already undid the action. | None. |
| **Your console session ended. Sign in again to continue.** | Your console session expired during a reply. | Select **Sign in again**, then open the conversation and continue. |
| **Another window took over this conversation.** | The conversation was continued in another window. | Continue in that window, or reopen the conversation here. |
| **Could not reach Pioneer AI. Check your connection and try again.** | Your browser could not reach your Outpost. | Check your network connection. |

With your own endpoint, a failure can also show one of the messages listed in
[Verify the source](/pioneer-ai/set-up/#verify-the-source).

## Next steps

- [Approve Pioneer AI actions](/pioneer-ai/approve-actions/)
- [Let Pioneer AI control a session](/remote-access/pioneer-control/)
- [Pioneer AI data and privacy](/pioneer-ai/data-and-privacy/)
- [Connect AI tools with MCP](/pioneer-ai/mcp-server/)