Using the Sidebar Agent
Control browsers with natural language commands from the Chrome sidebar
Opening the Sidebar
The AI Agent sidebar is your command center for browser automation. Here's how to access it:
Click the Prompteryx icon in your Chrome toolbar. The sidebar opens in Chrome's side panel, next to the page you're on.
The Sidebar Interface
The main area shows your conversation with the AI agent. You'll see:
- Your commands and requests
- AI responses and status updates
- Each step the agent takes (open a page, click, type, scroll) as a row on a timeline
- The agent's thinking, when "Show agent thinking" is on in Agent settings
- Error messages if something goes wrong
The bar near the top of the sidebar shows "AI Credits" used in the current chat and your account's "Balance". Each command you send uses credits based on the complexity of the task and the AI processing required. Runs are billed to the account the Prompteryx Desktop is signed in to, shown with the site in the line under the header.
Choose where the AI should execute actions:
- New tab: Opens a fresh tab for automation (default)
- Current tab: Runs actions on the page you're currently viewing
Type your commands in the text box at the bottom. You can also:
- Press Enter to send your message (Shift+Enter adds a new line)
- Click the paperclip icon to attach images
- Click the sliders icon to change the model, thinking level or AI vision for the next message only
- Click the pause button to pause and resume a run
- Click the stop button to cancel an ongoing operation
Giving Commands
You can give commands in natural language—just describe what you want to do as if you were explaining it to a human assistant.
"Go to amazon.com and search for wireless headphones"
→ Navigates to Amazon and performs a search
"Click on the first search result"
→ Clicks the first product in the search results
"Fill in the email field with test@example.com"
→ Finds the email input and types the address
"Take a screenshot"
→ Captures the current page state
"Scroll down to the pricing section"
→ Scrolls the page to find the pricing area
You can give complex, multi-step instructions in a single message:
"Go to github.com, navigate to my repositories, and find the project called 'my-app'"
The AI will break this down into individual steps and execute them in sequence, keeping you informed of progress along the way.
Action Recording
Every action the AI performs is automatically recorded with precise element selectors. This means you can:
- Export your session to Visual Studio as a reusable workflow
- Edit and refine the recorded actions
- Schedule the workflow to run automatically
- Share the workflow with team members
Use the "Visual Studio" button in the bar near the top of the sidebar when your session is complete—its badge counts the recorded actions.Learn more about exporting →
Safety Confirmations
When the AI model flags a step as needing your approval, the agent stops and asks before proceeding. This typically covers actions such as:
- Submitting forms with personal information
- Making purchases or payments
- Deleting data
- Posting public content
When a confirmation is required, you'll see a "Confirm this action" dialog with:
- An explanation of what the AI wants to do
- "Continue" button to allow the action
- "Cancel" button to stop and reconsider
Tips for Best Results
Instead of "click the button", say "click the blue Submit button at the bottom of the form". More context helps the AI find the right element.
Let the AI complete each step before giving the next command. Each step row switches from "Clicking" to "Clicked" (and so on) when it is done, and the chat shows "Task completed" at the end.
Describe elements by their visual appearance: "the red delete button", "the search box in the header", "the third item in the list".
For complex tasks, attach a screenshot showing what you want. Say "make it look like this" or "click the element I circled".