Using the Co-Pilot
How to effectively use Vibe to automate browser tasks.
Modes
Vibe has two operating modes, selectable via the icon next to the input box.
Agent Mode (Default)
Icon: Robot
The agent can control your browser: navigate, click, type, scroll, and interact with pages.
Use for:
- Filling forms
- Clicking buttons
- Navigating between pages
- Multi-step workflows
- Data extraction with interaction
Example:
"Search Amazon for wireless headphones and add the first result to cart"
Ask Mode
Icon: Eye
Read-only mode. The agent can see the current page but cannot interact with it.
Use for:
- Asking questions about page content
- Summarizing what's on screen
- Quick lookups without navigation
- Safer exploration of sensitive pages
Example:
"What's the price of the product on this page?"
Prompting Tips
Be Specific
Good: "Find the cheapest MacBook Air M3 on Amazon and Best Buy, compare prices"
Bad: "Find laptop prices"
Break Down Complex Tasks
Instead of:
"Book me a flight from NYC to LA next week, find a hotel near downtown, and rent a car"
Try:
"Find flights from NYC to LA for December 28-30"
Then follow up with hotel and car separately.
Use Natural Language
Vibe understands natural requests:
- "Go to Gmail and check my unread messages"
- "Search for Python tutorials on YouTube"
- "Add this item to my cart"
Specify Context When Needed
If the agent might be confused, add context:
- "On this page, click the Buy Now button"
- "In the search results, click the first link"
Browser Tools
The agent uses these tools to control your browser:
Navigation
| Tool | Description |
|---|---|
| Navigate to URL | Opens a specific URL |
| Create New Tab | Opens a new browser tab |
| Switch to Tab | Switches between open tabs |
| List Tabs | Shows all open tabs |
Interaction
| Tool | Description |
|---|---|
| Click | Clicks on elements (buttons, links) |
| Type | Types text character by character |
| Fill | Fills form fields |
| Scroll | Scrolls up, down, or to elements |
| Hover | Hovers over elements |
Advanced
| Tool | Description |
|---|---|
| Keyboard Shortcut | Presses key combinations |
| Drag and Drop | Drags elements |
| Wait | Pauses for page loading |
| Take Screenshot | Captures the visible page |
Memory
| Tool | Description |
|---|---|
| Store Memory | Saves information for later |
| Search Memories | Retrieves stored information |
Stopping the Agent
Click the Stop button to halt execution at any time. The agent will complete its current action and stop.
Parallel Execution
Vibe can spawn sub-agents for parallel tasks:
"Compare prices for MacBook Air on Amazon, Best Buy, and Walmart"
The agent may open multiple tabs and search simultaneously, then combine results.
Error Handling
If something goes wrong:
- Agent repeats actions: Rephrase your request more specifically
- Wrong element clicked: Use more precise descriptions ("the blue Add to Cart button")
- Page not loading: The agent will wait automatically; if timeout occurs, try again
- Task incomplete: Check if the page requires login or has CAPTCHAs
Best Practices
- Start simple - Test with basic navigation before complex workflows
- Watch the first run - Observe how the agent interprets your request
- Use Ask mode for questions - Faster and safer than Agent mode
- New conversation for unrelated tasks - Keeps context clean
- Enable "Show Thoughts" - Understand agent reasoning during debugging