> ## Documentation Index
> Fetch the complete documentation index at: https://docs.audimate.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Arro Workspace Agent
> Ask the workspace-wide assistant to navigate, find things, and build workflows, task boards, and playbooks from plain-language requests.
Arro Workspace Agent is the AI assistant for your whole workspace. It moves you around the app,
finds and answers questions about your workflows, task boards, and pages, and creates new ones.
Inside a workflow it also turns a raw recording into a runnable workflow by drafting playbooks,
cleaning up steps, and scaffolding task boards.
It is available on every page, and so is everything it can do. The agent does not gain or lose
abilities as you navigate, so you can ask about a task board while looking at a workflow.
## Where to find it
The **Ask Arro** button sits near the top of the workspace sidebar, below **New**. Click
it to open the agent panel alongside whatever page you are on. Your conversation stays with you as
you move between pages.
Arro knows which workflow, recording, playbook, task board, or page you have open when you send a message.
You can say "add notes to this recording" without supplying an identifier.
To work on another recording, name its workflow so Arro can find it.
This panel uses your current Audimate page as context. It does not inspect other applications on your screen.
Inside a workflow, the **Create** button in the header opens the **Create with AI** dialog, a
structured form for the most common build requests. It is the same agent as the chat panel.
In a page or a playbook, highlighting text and clicking **Ask Arro** opens the panel with your
question and the highlighted text already attached. See
[Ask Arro about what you highlighted](/concepts/pages#ask-arro-about-what-you-highlighted).
## What it can do
Everything below works from any page. Where something is not open in front of you, name it.
| Ask for | What it does |
| -------------------------------- | --------------------------------------------------------------------------- |
| "Take me to my task boards" | Opens any page, workflow, or task board |
| "How many workflows do I have?" | Searches, lists, counts, and reads details |
| "Make me a task board for leads" | Creates a workflow, task board, or playbook |
| "Rename this workflow" | Changes a workflow's title and description |
| "Clean up this recording" | Rewrites, annotates, and drops steps |
| "Draft a playbook from this" | Generates or refines a playbook |
| "Add these three contacts" | Adds, updates, and deletes tasks |
| "Add a Priority field" | Adds, renames, and removes task board fields |
| "What is our escalation policy?" | Reads [knowledge base](/concepts/knowledge) and answers with page citations |
| "Create a page from this PDF" | Writes pages from an attached document |
If you only have view access to something, the agent reads it and declines edits.
### Creating something new
The agent checks the name with you before creating anything, and you can ask for it in your private
library or the shared team library. A new workflow or task board opens automatically once it is made. A
task board created and attached to a workflow in one step inherits that workflow's visibility, so the
private-or-shared choice applies only to a standalone task board.
### Changing tasks in bulk
Large requests are split into batches, so "add these 60 contacts" arrives as several writes by
design. If someone edits a task while the agent is working on it, that one write is skipped rather
than overwriting them, and the result card says how many.
## What it will not do
The agent is deliberately bounded. It cannot:
* **Delete a workflow, task board, folder, playbook, or page.** Only tasks, fields, and recording steps.
* **Add or reorder recording steps.** Rewriting, deleting, and annotating are all it can do.
* **Change a field's type or its options** once the field exists. That means removing and re-adding.
* **Rename, move, or re-share anything** beyond a workflow's title and description and a task board's
field names.
* **Change billing, plans, or organization settings.** It will not even navigate you there.
* **Record for you.** Capturing a demonstration happens in the desktop app.
* **Create folders**, or attach an existing task board to a workflow.
A task board can be attached at the moment it is created. After that it is a manual step on the
workflow's **Data** tab.
## Two ways to use it
### The agent panel
The panel opens next to the sidebar with a chat history and a message input, and stays open as you
navigate. It is **context-aware**: on the **Recording** tab it knows which recording you mean by
*"this"*, on **Playbook** a request targets the open playbook, on a task board *"this task board"* means
the one on screen, and on a page *"this page"* is the open page. When you have highlighted text,
*"this"* means the text you highlighted. You can always name something else instead, and it does not
have to be open.
The message input stays below the scrolling chat history, so it never covers a message.
The paperclip (**Attach files**) adds files to a message, and Arro reads them in the conversation.
You can send a file on its own without typing a message, or include text with your attachment.
The supported types and size limits are under
[Attach a document](/concepts/knowledge#attach-a-document).
If a file cannot be read, your draft stays in place. Try sending again, or remove the attachment and select it again.
Above the chat, a row of **kickoff buttons** suggests common actions for the active tab, such as
*Clean up this recording* and *Add section headings*.
### The Create with AI dialog
| Tab | What it does |
| ---------- | ---------------------------------------------------------------- |
| **Create** | Scaffolds a new playbook or template task board from a recording |
| **Edit** | Updates an existing playbook or task board from your description |
| **Manual** | Adds an empty recording, playbook, or task board by hand |
Use **Edit** when you want to change one part of an existing playbook or task board. It applies the
change without rewriting unrelated content.
## Approvals
The agent pauses for your approval on the actions worth a second look. These raise an approval
card:
* Deleting tasks.
* Removing a field.
* Replacing a playbook's whole body.
* Rewriting a page or one of its sections.
* Editing [text you highlighted](/concepts/pages#ask-arro-about-what-you-highlighted).
* Deleting recording steps.
* Any batch of more than five task changes.
* Any batch of more than five page changes.
The card says exactly what changes, and nothing happens until you approve it. Smaller edits, like
creating something, renaming a workflow, adding a field, or changing a few tasks, apply without a
pause.
## How changes are applied
The agent takes you to whatever it changes, so the page opens on what it changed. This
works the same in the browser and in the desktop app.
Every edit is saved immediately. There is no draft and no save step, so the change is written the
moment the agent makes it. A playbook edit goes through the same live document the
editor uses, so a teammate with the playbook open sees the change as it happens.
The approval card is the point of no return for tasks. **Ctrl+Z** (**⌘+Z** on a Mac) on the task board undoes deletions
you made in the grid, not the agent's. Take a snapshot before letting the agent rewrite a
workflow's steps, because a snapshot is your undo. See [Workflows](/concepts/workflows).
The agent only ever edits a workflow's **working copy**. Ask it to change a published snapshot and
it declines and tells you to switch back, because the snapshot is deliberately frozen.
## Handing off a run
The agent builds and refines a workflow. It does not run one. **Start workflow → Run with AI** in
the workflow header starts the run in the desktop app, and it appears in your **Tasks** list. If a
run finishes with errors, open the workflow and ask the agent to triage what went wrong, then run it
again once it is fixed.
## Related
Open one to unlock the agent's editing tools.
The tasks and fields the agent reads and writes.