Browser

Overview of the Browser integration for step-by-step web automation.

Overview

The Browser integration lets you automate web interactions through precise, individual actions — navigating to a page, clicking buttons, filling forms, verifying content, and more.

Each run uses:

• Live view to watch the browser in real time.

• Agent step stream to track what the agent is doing step-by-step.

• Restricted tools so the agent is limited to approved browser actions (not arbitrary code execution).

Use the Browser integration when you need:

• Fine-grained control over each interaction on a web page.

• Secure credential handling where passwords and secrets are injected without being exposed to the agent or logs.

• Verification that specific content or values are present before continuing.

• File downloads and uploads as part of a web workflow.

• Data extraction from pages using AI-powered reading. For visual extraction workflows, connect Intelligent Document Processing (IDP).

How It Works

Browser automation follows this workflow:

1. You describe the task in natural language.

2. The agent explores the page and authors a deterministic automation.

3. The automation is run and a verification pass checks stability.

4. You review progress in live view and the step stream.

Setup

The following integrations need to be connected to your Kognitos workspace:

• Browser

Steps

Follow these steps to connect the integration in Kognitos:

1

Navigate

Using the left navigation menu, go to Integrations → Explore Integrations.

2

Find

Search for Browser and click on it.

3

Connect

Click on Connect to add a connection to the integration.

4

Configure

Add a name for the connection. You'll be prompted for authentication details if needed. Then, click on Connect.

Authentication

The Browser book can be used with or without credentials.

To automate workflows that require login or sensitive input, connect with Browser Credentials.

Credentials are stored as named key-value pairs and injected securely at runtime. Credential values are not exposed to the agent and are not shown in your automation outputs.

Credential key naming rules

Credential key names (the left-hand side of each pair) must contain only:

• Alphanumeric characters (A–Z, a–z, 0–9)

• Hyphens (-)

• Underscores (_)

Characters such as @, ., and spaces are not allowed in key names.

Do not put your actual email address (or any value containing @) in the key name field. The key is just a label you choose to reference the credential — the value is where the actual secret goes.

Example — email-based login:

Key	Value
email_address	[e-mail]
email_password	your_password

You can optionally reference credentials by key name in your prompt — for example, "type the 'email_address' credential in the Email field" — though the agent will also discover and use the right credential automatically.

Getting Started

A typical Browser automation follows this pattern:

1. Provision a browser.

2. Navigate to a website.

3. Interact with the page — click buttons, fill out forms, select options.

4. Verify that the expected content appeared.

5. Extract information if needed.

6. Close the browser.

Describe each step in plain English in your prompt. For example:

Provision a browser. Navigate to https://example.com/login. Type "admin" in the Username field. Type the "password" credential in the Password field. Click the Log In button. Verify that "Welcome" is visible on the page. Then close the browser.

Best Practices

• Provision first. Every browser interaction requires a provisioned browser, so always start by provisioning one.

• Prompt with clear intent. Describe the business outcome first, then the key page steps.

• Be specific about elements. Instead of "click the button," say "click the Submit Order button" or "type in the Email field."

• Use scope when elements share the same name. If a page has two sections with a "Street" field, specify which one: "type '123 Main St' in the Street field in the Billing Address section."

• Use credentials for sensitive data. Reference credentials by name rather than typing secrets directly.

• Add explicit verification steps. Confirm expected state after key transitions (for example, "verify that 'Order confirmed' is visible").

Actions

The following capabilities are available in the Browser integration:

Lifecycle

Action	Description
Provision a browser	Start a new remote browser session. Returns a live view URL you can use to watch in real time.
Close a browser	End the browser session and release resources.
Get downloaded files	Retrieve any files the browser downloaded during the session.

Navigation

Action	Description
Navigate to a URL	Open a web page in the browser.
Go back	Navigate to the previous page in browser history.
Go forward	Navigate to the next page in browser history.
Scroll the page	Scroll up, down, left, or right on the current page.

Clicking and Hovering

Action	Description
Click an element	Click a button, link, or other interactive element on the page.
Hover an element	Move the cursor over an element without clicking.
Drag an element	Drag one element to another location on the page.

Typing and Selection

Action	Description
Type text	Type into an input field. Can optionally press Enter after typing.
Type a credential	Securely type a credential value (e.g., a password) into a field. Requires Browser Credentials.
Select a dropdown option	Choose an option from a dropdown or combobox.
Press a key	Press a keyboard key or combination (e.g., Enter, Tab, Ctrl+A).

Forms and File Upload

Action	Description
Fill a form	Fill multiple fields in a single step.
Fill a form using credentials	Fill a form where some fields use securely stored credential values. Requires Browser Credentials.
Upload a file	Upload a file to a file input on the page.

Dialogs and Tabs

Action	Description
Handle a dialog	Accept or dismiss the next browser dialog (alert, confirm, prompt). Must be called before the action that triggers the dialog.
List tabs	See all open browser tabs with their titles and URLs.
Switch tab	Switch focus to a different browser tab.

Waiting

Action	Description
Await text	Wait until specific text appears on the page (up to a timeout).
Await a specified time	Pause for a set number of seconds.

Verification

Action	Description
Verify text is visible	Confirm that specific text is visible on the page.
Verify an element is visible	Confirm that a specific element is present and visible.
Verify a value	Confirm that a form field contains an expected value.
Verify a list	Confirm that all items in a list of text strings are visible on the page.

Observation and Extraction

Action	Description
Take a snapshot	Get a structured, text-based view of all elements on the page. Useful for inspecting page content.
Take a screenshot	Capture a visual image of the current page.
Extract from a screenshot	Use AI to answer a question about the visual content of the page.
Extract from a snapshot	Use AI to answer a question about the structured content of the page.

Extraction actions use AI and are best suited for one-off reads. Avoid requesting them repeatedly in tight loops.

Concepts

Browser instance

Represents a provisioned browser session.

Field Name	Description	Type
session_id	Unique identifier for the browser session.	text
live_view_url	URL to watch the browser session in real time.	text

Browser action result

The result returned after each browser action.

Field Name	Description	Type
success	Whether the action completed successfully.	boolean
result	The action's output (for example, snapshot content, extracted data, PASS/FAIL).	text
action_type	The type of action that was performed.	text
snapshot	A structured view of the page after the action (when available).	optional[text]