Browser Tools

Overview

Tool Name

browser_tools

Purpose

The browser_tools group lets Genesis interact with live web UIs. Use it when a workflow needs to open a page, inspect the current layout, click controls, type into fields, capture screenshots, or step through a multi-action browser flow on an external site or an internal Genesis page.

Use browser_tools for interactive UI work. For simple page reads, API calls, or file downloads, use Web Access Tools instead.

If Chromium is not included in the Databricks App deployment, browser_open returns an unavailable error. Related web actions and non-browser alternatives are described in Web Access Tools. On Genesis Desktop, the first browser_open or app screenshot/runtime inspection starts a one-time browser-tools download when needed. This also applies to CAPTURE or INSPECT_RUNTIME_STATE, including captures inside REVIEW or START. The download is about 140–160 MB and can take up to 150 seconds. If a response reports that installation is still in progress, relay the status and repeat the action in about a minute. If installation fails, relay the reason; the user can retry with Install in Settings → Advanced → Browser tools.

Key Features

Open live browser sessions

Start a persistent browser session against a full external URL such as https://example.com or an internal Genesis path such as /apps/my-app.

Inspect accessible page snapshots

Return a compact text snapshot with headings, landmarks, and interactive elements labeled with el_* IDs for follow-up actions.

Click and type into live pages

Interact with buttons, links, inputs, selects, and text areas by current element ID or matching text.

Capture screenshots when needed

Add include_screenshot=True to capture the current viewport during browser_snapshot.

Close sessions explicitly

End the browser session when the task is complete to release resources.

Input Parameters for Each Function

`browser_open`

Name	Type	Default	Definition
`url`	String	none	Required. URL to open. Use a full URL for external sites such as `https://example.com` or a relative path for internal Genesis pages such as `/apps/my-app`.
`viewport_width`	Integer	`1280`	Browser viewport width in pixels.
`viewport_height`	Integer	`720`	Browser viewport height in pixels.

`browser_snapshot`

Name	Type	Default	Definition
`session_id`	String	none	Required. Session ID returned by `browser_open`.
`include_screenshot`	Boolean	`False`	Also capture a visual screenshot of the current viewport.

`browser_click`

Name	Type	Default	Definition
`session_id`	String	none	Required. Session ID returned by `browser_open`.
`element`	String	none	Required. Element to click. Use an element ID from the current snapshot such as `el_4` or visible text content to match.

`browser_type`

Name	Type	Default	Definition
`session_id`	String	none	Required. Session ID returned by `browser_open`.
`element`	String	none	Required. Element to type into. Use an element ID from the current snapshot such as `el_5` or visible text or placeholder to match.
`text`	String	none	Required. Text to type into the element.
`press_enter`	Boolean	`False`	Press Enter after typing.
`clear_first`	Boolean	`True`	Clear existing text before typing. Set to `False` to append instead.

`browser_close`

Name	Type	Default	Definition
`session_id`	String	none	Required. Session ID returned by `browser_open`.

thread_id and bot_id are supplied implicitly by Genesis. After browser_open, reuse the returned session_id for browser_snapshot, browser_click, browser_type, and browser_close.

Workflow/How It Works

Call browser_open with an external URL or an internal Genesis path.
Review the returned snapshot and identify the current el_* IDs for interactive elements.
Use browser_click or browser_type to take the next action.
Call browser_snapshot whenever you want a refreshed page view or a screenshot.
Finish with browser_close to release the session.

Element IDs are generated from the current page snapshot. After a navigation or UI change, use the latest returned snapshot before taking the next action.

Examples

# Open an internal Genesis page
result = browser_open(url="/apps/customer-health")
session_id = result["session_id"]

# Click a button from the current snapshot
browser_click(session_id=session_id, element="el_3")

# Type into a search box and submit
browser_type(
    session_id=session_id,
    element="Search",
    text="Q1 pipeline health",
    press_enter=True,
)

# Refresh the snapshot and capture a screenshot
browser_snapshot(session_id=session_id, include_screenshot=True)

# Close the session when finished
browser_close(session_id=session_id)

Output

browser_open returns success, session_id, url, snapshot, element_count, and usage tips. On Genesis Desktop, if the one-time browser-tools download is still running, it instead returns progress without a session_id; relay the status and try again shortly.
browser_snapshot returns success, session_id, title, url, snapshot, and element_count. When include_screenshot=True, it also returns screenshot_path and screenshot_reminder.
browser_click and browser_type return success, session_id, url, snapshot, and element_count after the interaction.
browser_close returns success, session_id, and status set to closed.

Configuration Details

Use full URLs for external sites.
Use relative paths such as /apps/my-app or /api/apps/<app_id>/proxy/ for internal Genesis pages.
When browser_open targets the local Genesis backend on localhost or 127.0.0.1 using the active FastAPI port, Genesis automatically attaches the internal service token as the X-Genesis-Internal-Token header.
That header is scoped to the local Genesis origin only and is not sent to external sites or other localhost ports.
The default viewport is 1280x720.
browser_type clears existing text by default; set clear_first=False to append instead.

Limitations or Notes

These tools operate on a live browser session, so page state can change between actions.
browser_click and browser_type work best when you use the current snapshot’s el_* IDs instead of guessing text matches.
include_screenshot=True captures the current viewport rather than a stitched full-page image.
Always call browser_close when you are done to free server resources.

Getting Started

Getting to Know Genesis

Creating A Mission

User Guide

Blueprint

Genesis Data Agents

Genesis Prover

Genesis Apps

Genesis REST API

Genesis Agent Security Controls

Genesis Data Agent's Toolkit

Setup

Slack and Teams

Data Connections

Deployment Options

Overview

Tool Name

Purpose

Key Features

Input Parameters for Each Function

`browser_open`

`browser_snapshot`

`browser_click`

`browser_type`

`browser_close`

Workflow/How It Works

Examples

Output

Configuration Details

Limitations or Notes

​Overview

​Tool Name

​Purpose

​Key Features

​Input Parameters for Each Function

​browser_open

​browser_snapshot

​browser_click

​browser_type

​browser_close

​Workflow/How It Works

​Examples

​Output

​Configuration Details

​Limitations or Notes

Overview

Tool Name

Purpose

Key Features

Input Parameters for Each Function

`browser_open`

`browser_snapshot`

`browser_click`

`browser_type`

`browser_close`

Workflow/How It Works

Examples

Output

Configuration Details

Limitations or Notes