Browser Agent

Core capabilities

The browser agent gives agent mode four categories of browser actions:

• Open and navigate web pages

  • Open a specific web page using a provided URL.

  • Follow navigation or pagination links to reach new pages or tabs within the same site.

  • Handle multi-step navigation flows such as "Open page A → click menu B → go to details page C."

• Read and extract information

  • Read visible text on the current page — titles, paragraphs, lists, and tables.

  • Extract key information and return summaries or comparisons in natural language.

  • Find relevant content based on your instructions, such as "Find all price-related content on this page."

• Interact with pages

  • Click buttons and links, switch tabs, or expand and collapse content.

  • Fill in form fields such as input and search boxes, and submit forms.

  • Scroll the page to view more content and avoid missing key information.

• Visual feedback and status awareness

  • Capture the current page state during complex steps, as needed, for analysis.

  • Detect page status (load completion, successful form submission, navigation to a new page) to decide the next action.

Use cases

• Information retrieval and comparison

  • Visit product sites, documentation pages, or blogs to extract key information and generate summaries. Try: /browser open https://example.com/pricing and summarize the plan tiers

  • Compare details across multiple pages — prices, features, or configuration options. Try: /browser compare the storage limits across these three pages and put the results in a table

• Online operations and workflow walkthroughs

  • Walk through a web-based workflow — such as registering an account or submitting a ticket — when permissions are sufficient and risks are controlled. Try: /browser walk through the account registration flow on https://example.com, read-only, do not submit any forms

  • Document the steps for a web-based admin system and draft the instructions. Try: /browser navigate the admin dashboard at https://example.com/admin and generate a step-by-step usage guide

• Development and testing assistance

  • Pull relevant sections from online documentation or an API reference for the code you are writing. Try: /browser open https://example.com/api-reference and extract the authentication section

  • Review a web application's page structure, copy, or interaction logic and suggest improvements. Try: /browser check the checkout flow on https://example.com/shop and identify usability issues

Include your goal and constraints in the task description — for example, "read-only, do not submit any forms" or "only access public documentation pages." This helps the agent work more safely and reliably.

Browser types

The browser agent supports two browser types. Switch between them in the browser agent settings.

• Built-in browser — a lightweight browser panel integrated directly into the IDE. No extra configuration needed. Best for quick previews and simple page interactions.

• Chrome — uses your local Chrome browser. Handles more complex web applications and pages that require specific browser features or extensions.

Use the browser agent in agent mode

The browser agent is built into agent mode — no separate configuration required. Invoke it in two ways:

• Automatically — agent mode detects when a task needs browser access and invokes the browser agent on its own.

• Explicitly — use the /browser command to trigger the browser agent directly.

Best practices

• State the goal, not just a single step

  • Describe the desired outcome in one sentence rather than describing just one step.

  • For security- or permission-sensitive operations, include explicit constraints: "do not submit, pay, or delete anything."

• Provide a stable entry URL

  • Give a direct page URL rather than a search term to avoid unnecessary navigation steps and reduce failure risk.

  • For multi-page tasks, list the key URLs or paths in your prompt.

• Break complex tasks into stages

  • For long workflows — such as a multi-step configuration wizard — split the task into smaller goals and verify each stage before moving on.

  • Adjust your next prompt based on what the browser agent returned in the previous stage.

Security and limitations

• Permissions and privacy

  • Never enter sensitive information through the browser agent — passwords, access tokens, or personal data.

  • For account login, payment, or data modification, complete those steps manually first, then let the agent handle read-only validation or instruction generation.

• Page compatibility and stability

  • Pages that depend heavily on front-end frameworks or complex interactions may load slowly or have elements the agent cannot locate.

  • If a page's structure or content changes frequently, some steps may fail. Use a more specific description or a more stable entry URL.

• Result credibility

  • The browser agent reads real-time web content, but that content may not be authoritative. Verify critical information before acting on it.

  • For legal, compliance, or high-risk business decisions, do not rely solely on automated browser results.