Built-in Tools Reference

The Agent has a set of built-in tools that the AI can call automatically during a conversation. These tools are available by default in persistent conversations, and script developers usually don't need to call them directly — the AI automatically picks the right tool based on the user's intent.

Understanding what these tools can do helps you write better system prompts and custom tools.

Web Data Fetching

web_fetch

Fetches the content of a URL, supporting HTML-to-text extraction and LLM summarization.

Parameter	Type	Required	Description
url	string	Yes	Target URL (http/https only)
prompt	string	No	Summarization prompt (when provided, the LLM is used to distill the content)
max_length	number	No	Maximum character count for the content

Behavior details:

• 30-second request timeout
• HTML content automatically extracts the main body (removing navigation, sidebars, etc.)
• JSON responses are automatically parsed
• Plain text is returned as-is
• When prompt is provided, the fetched content is sent to the LLM for summarization

Return value:

{
  "url": "https://example.com",
  "content_type": "text/html",
  "content": "The extracted body content...",
  "truncated": false,
  "final_url": "https://example.com/redirected"
}

web_search

Queries a search engine and returns structured search results.

Parameter	Type	Required	Description
query	string	Yes	The search keywords
max_results	number	No	Maximum number of results (defaults to 5, capped at 10)

Supported search engines:

Engine	Description	Configuration required
DuckDuckGo	The default engine	No configuration needed
Bing	Microsoft Bing Search	Requires an API Key
Baidu	Baidu Search	No API Key needed
Google Custom Search	Google Custom Search	Requires an API Key + CSE ID

Search engines are configured in the dashboard under Agent → Settings.

Return value:

[
  {
    "title": "Search result title",
    "url": "https://example.com/result",
    "snippet": "Result snippet text..."
  }
]

get_tab_content

Reads the rendered page content of a given tab, converting it into structured Markdown with CSS-selector annotations.

Parameter	Type	Required	Description
tab_id	number	Yes	Tab ID
selector	string	No	A CSS selector; only extracts the matching part
prompt	string	No	Summarization prompt
max_length	number	No	Maximum character count for the content

Difference from web_fetch: get_tab_content reads the page as already rendered by the browser (including dynamic JS content), while web_fetch makes a new HTTP request.

Return value:

{
  "tab_id": 123,
  "url": "https://example.com",
  "title": "Page title",
  "content": "Structured content...",
  "truncated": false,
  "used_selector": "main"
}

Tab Management

list_tabs

Queries open tabs, supporting several filter conditions.

Parameter	Type	Required	Description
url_pattern	string	No	A regex to match the URL
title_pattern	string	No	A regex to match the title
active	boolean	No	Only return the active tab
window_id	number	No	A specific window
audible	boolean	No	Only return tabs currently playing audio

open_tab

Opens a new tab or navigates an existing one.

Parameter	Type	Required	Description
url	string	Yes	Target URL
tab_id	number	No	An existing tab ID (when provided, navigates that tab; otherwise a new tab is opened)
active	boolean	No	Whether to activate it (defaults to true)
window_id	number	No	A specific window
wait_until_loaded	boolean	No	Whether to wait for the page to finish loading (defaults to true)

close_tab

Closes a tab.

Parameter	Type	Required	Description
tab_id	number	Yes	Tab ID

activate_tab

Activates a tab and focuses its window.

Parameter	Type	Required	Description
tab_id	number	Yes	Tab ID

File System (OPFS)

opfs_write

Writes a file to the workspace.

Parameter	Type	Required	Description
path	string	Yes	File path
content	string	Yes	File content (supports binary via a data URL)

opfs_read

Reads a file from the workspace. By default it auto-detects the file type: text files return their content, binary files return a blob URL.

Parameter	Type	Required	Description
path	string	Yes	File path
mode	string	No	"text" / "blob" / "auto" (default), forces the return mode
offset	number	No	Starting line number (1-based), text mode only
limit	number	No	Number of lines to read, text mode only (pagination is required once text exceeds 200 lines)

opfs_list

Lists the contents of a directory.

Parameter	Type	Required	Description
path	string	No	Directory path (defaults to the root directory)

opfs_delete

Deletes a file or directory.

Parameter	Type	Required	Description
path	string	Yes	File/directory path

User Interaction

ask_user

Asks the user a question, supporting either free-form input or a structured choice.

Parameter	Type	Required	Description
question	string	Yes	The question content
options	string[]	No	A list of choices (when provided, this becomes a multiple-choice question)
multiple	boolean	No	Whether multiple selection is allowed (defaults to false)

Timeout: returns { answer: null, reason: "timeout" } after 5 minutes without a response.

Return value:

{ "answer": "The user's answer text" }

execute_script

Executes JavaScript code on the page or in the sandbox.

Parameter	Type	Required	Description
code	string	Yes	The JavaScript code
target	string	Yes	"page" or "sandbox"
tab_id	number	No	The target tab when target is page (defaults to the current active tab)
world	string	No	"MAIN" or "ISOLATED" (default), page mode only

Execution environment comparison:

Environment	DOM	Page JS	Extension blob URLs	Use case
page + ISOLATED	Yes	No	Yes	Reading the DOM, extracting content
page + MAIN	Yes	Yes	No	Calling page functions
sandbox	No	No	No	Pure computation

Sub-agents

agent

Spawns an independent sub-agent to handle a complex sub-task.

Parameter	Type	Required	Description
prompt	string	Yes	The sub-task description
description	string	No	A short label (3-5 words, used in the UI)
type	string	No	The sub-agent type (see below); defaults to "general"
tab_id	number	No	A tab ID passed to the sub-agent, which will operate on that tab

Sub-agent types:

type	Description	Available tools
researcher	Information retrieval (read-only)	web_search, web_fetch, reading page content
page_operator	Browser automation	Tab management, DOM operations, page interaction
general	General-purpose (default)	All tools

Characteristics:

• A sub-agent has its own independent conversation context
• It cannot use ask_user or agent (to prevent recursion)
• The sub-agent's events are passed to the parent conversation via sub_agent_event

Task Management

This group of tools manages a temporary task list within a conversation (in-memory, not persisted).

create_task

Parameter	Type	Required	Description
subject	string	Yes	Task title
description	string	No	Detailed description

get_task

Parameter	Type	Required
task_id	string	Yes

update_task

Parameter	Type	Required	Description
task_id	string	Yes	Task ID
status	string	No	"pending" / "in_progress" / "completed"
subject	string	No	New title
description	string	No	New description

list_tasks

Takes no parameters; returns a brief list of all tasks.

delete_task

Parameter	Type	Required
task_id	string	Yes

The task management tools mainly let the AI track its own progress while working through complex, multi-step tasks; task data is not persisted.