Core system tools

28 min

Mark (Optimizely Agent Platform) uses system tools to take action on your behalf. This page lists the built-in system tools provided and maintained by Optimizely. You and Mark can call these tools if they are enabled. Review the tools so you can see what actions are available for you and your workflows.

System tools by functionality

Click a tool's name to expand it and learn when to use it, its required and optional parameters, and example prompts for calling the tool. If you do not provide a required parameter, Mark prompts you for it.

Audio and video

Browse the web

Code execution

Date

Email notifications

File operations

Ideate and reason

Images

Keywords

Search

Teammates

User input

The card – A card holds an optional picture, markdown text, an optional progress bar, and one box. The box holds the question, an optional editable table, an optional CSV file upload, and the buttons. By default, the sections display in this order: picture, text, box, progress bar. Set layout to change the order.

Buttons – A card has one to six buttons. Each label has up to 60 characters, and the labels must differ from each other. When you click a button, its label becomes your reply. Every button uses the primary look unless the agent sets it to default, subtle, or danger.

Table and upload – Set table, csv_upload, or both to true. The table starts with empty rows, and you add and remove rows in the card. The upload accepts one .csv file. When either one is on, the card adds a submit button. When only the upload is on, the submit button stays disabled until the file finishes uploading.

Submitted reply – The submit button sends a message. The first line comes from submit_text, and one line follows for each row that has a filled cell. A separator joins the cells in a row. An empty cell stays as a gap, so no value shifts into another column. The CSV file attaches to the same message.

Limits – The question is one line of up to 200 characters. The text accepts markdown up to 6,000 characters. A card holds up to five images, including image_url. Each image must use an https link on the Content Marketing Platform domain *.cmp.optimizely.com. The text rejects raw Hypertext Markup Language (HTML) tags, so write literal angle brackets in backticks.

Agents – Mark and specialized agents call this tool the same way.

When to use

Guide a user through a multistep flow in chat, where each step shows a message and the next choices.

Keep a message and its choices together in one card.

Collect a short list, such as names and email addresses, or a CSV file with the reply.

Show progress through a flow with a progress bar.

Use ask_user_question instead when you only need answers to structured questions.

Parameters

prompt – The question that displays with the buttons. Use one line of up to 200 characters.

options – The button labels. Provide one to six labels that differ from each other, with up to 60 characters each. The label of the clicked button becomes your reply.

(Optional) option_appearances – The look of each button, in the same order as options: primary, default, subtle, or danger. Omit this value to make every button primary.

(Optional) subtitle – Secondary text that displays under the question, up to 300 characters.

(Optional) body – The message, written in markdown, with up to 6,000 characters. Do not use raw HTML tags.

(Optional) image_url – An https link to a picture that displays at the top of the card.

(Optional) image_alt – A short description of the picture.

(Optional) progress – A whole percentage from 0 to 100. The card displays a progress bar and a label such as 45% complete.

(Optional) layout – The order of the card sections, top to bottom: image, body, progress, and actions. The actions section is the box with the question, table, upload, and buttons. The default order is image, body, actions, progress. The list must include actions, must not repeat a name, and must name every section that has content.

(Optional) table – Set to true to display an editable table.

(Optional) table_columns – The column names, one to six, such as ["First name", "Last name", "Email"]. The default columns are Name and Email.

(Optional) table_rows – The number of empty rows the table starts with, from one to 10. The default is three.

(Optional) table_label – The label for the table. The default is List.

(Optional) add_row_label – The label for adding a row. The default is + Add row.

(Optional) row_separator – The text between the cells of a row in the submitted message, from one to five printable American Standard Code for Information Interchange (ASCII) characters. The default is a vertical bar with a space on each side.

(Optional) csv_upload – Set to true to display an upload for one CSV file.

(Optional) upload_label – The label for the upload. The default is Upload a CSV file.

(Optional) submit_label – The label for the submit button. The default is Submit. Do not repeat it in options.

(Optional) submit_text – The first line of the submitted message. The default is Here is my list:.

(Optional) submit_appearance – The look of the submit button: primary (default), default, subtle, or danger.

Example prompts

Show a card that asks whether I finished installing my agents, with the buttons Done installing and Skip for now.

Show a card with an editable table where I enter the names and email addresses of my team.

Show a card with my plan, a progress bar at 50 percent, and the buttons Continue and Skip for now.

Webpage analysis

Workspace

System tools in beta

Tools labeled Preview are not available in Mark chat by default and are in beta. Contact your Customer Success Manager to add your instance to the beta and then Mark administrators can go to Connectors > Tools and enable them in Mark chat.

Memory

Beta

The memory tools are in beta. Contact your Customer Success Manager for more information.

Video

Beta

The video tools are in beta. Contact your Customer Success Manager for more information.

Note

If you use Opti ID, administrators can turn off generative AI in the Opti ID Admin Center. See Turn generative AI off across Optimizely applications.

Core system tools

System tools by functionality

Audio and video

audio_video_processing – Process audio and video files for transcription, subtitle generation, summarization, and content analysis.

Browse the web

browse_web – Opens and extracts content from multiple webpages concurrently and returns the content in Markdown or HTML format.

browse_web_html – Browses a single webpage, extracts information, and returns content in raw HTML format.

take_webpage_screenshot – Captures a screenshot of a webpage by given URL.

Code execution

code_execute – Execute code in a sandboxed environment supporting Python, Bash, and Node.js.

Date

get_today – Returns the current UTC date and time in a human-readable Markdown format.

Email notifications

send_email – Composes and sends emails as a notification tool. Supports setting email triggers by specifying various parameters. See Email domain configuration﻿.

File operations

convert_to_pdf – Converts a file or webpage to PDF.

edit_file – Replace an exact string in an existing text file.

get_file_metadata – Retrieves metadata for a file, such as dimensions or alt text.

list_files – Lists files in the workspace.

read_file – Reads the contents of a file.

read_file_content – Reads a file from the Mark﻿ backend with configurable encoding.

retrieve_file_from_gcs_uri – Read file content directly from a Google Cloud Storage (GCS) URI.

search_files – Searches for files in the workspace by filename or content.

write_content_to_file – Saves generated content by Mark﻿ as a file in the backend.

write_file – Creates a text file in the workspace or persistent context directory.

Ideate and reason

evaluate_response – Evaluates LLM responses as a quality assurance tool. Takes an original prompt and the LLM's response, then uses another LLM to provide a quality score (0-100) and feedback for improvement.

ideate – Generates a plan and suggests available tools for complex or ambiguous tasks. Use this tool when Mark﻿ is overwhelmed with how to proceed or needs a plan for how to address a given problem or topic.

reasoning_step – Articulates internal intermediate reasoning steps, plans, or observations for complex questions to help Mark﻿ continue multi-step work. This is particularly useful for complex questions that require careful consideration or decomposition before formulating a final answer.

Images

analyze_image_content – Helps Mark﻿ understand and interpret the visual content of images, whether they are screenshots you provide or images Mark﻿ has found through a search. Mark﻿ can answer your questions based on what it "sees" in the image.

generate_or_edit_image – Generate new images from text descriptions or edit existing images. Supports style transfers, object manipulation, zoom, and more.

present_images – Displays generated or edited images to you as a rich visual card in the chat interface. Mark﻿ calls this tool after generate_or_edit_image when images are the final output.

Keywords

figure_out_search_keywords – Generates 8-15 related search keywords for comprehensive information about a topic, presented in a markdown table format.

Search

search_web – Searches the web for a given query using Google.

Teammates

invite_teammates – Displays a card in chat where you enter the email addresses and names of colleagues and invite them to your Mark﻿ instance with the Mark User role. Opti ID sends an invitation to each person who is new to your organization.

User input

ask_user_question – Presents the user with one or more structured questions and selectable options to gather input before proceeding with a task.

show_choice_card – Displays one card in chat that combines a message with buttons. The card optionally includes a picture, a progress bar, an editable table, and a CSV file upload.

Webpage analysis

analyze_pagespeed – Analyze a webpage's performance, SEO, or accessibility using Google PageSpeed Insights.

Workspace

list_workspace_mounts – Lists the canvas spaces mounted into the workspace for the current conversation, along with the files each mount contains.

mount_space_to_workspace – Makes the files in a canvas space available to a code run, so Mark﻿ reads and edits them in place. Mounting reaches images and other binary files that canvas editing cannot change.

unmount_space_from_workspace – Detaches a mounted canvas space and removes its files from the workspace. The canvas space and its files remain in Mark﻿. Save pending changes before you unmount, because Mark﻿ discards unsaved edits.

System tools in beta

Memory

recall – Searches stored memories and returns matching entries with snippets and content.

remember – Stores a memory for future conversations, persisting facts, preferences, or decisions across sessions.

Video

generate_video – Generates videos from text descriptions and optional starting images. It supports various aspect ratios and durations, and includes automatic prompt enhancement and audio generation.

send_email – Composes and sends emails as a notification tool. Supports setting email triggers by specifying various parameters. See Email domain configuration.

read_file_content – Reads a file from the Mark backend with configurable encoding.

write_content_to_file – Saves generated content by Mark as a file in the backend.

ideate – Generates a plan and suggests available tools for complex or ambiguous tasks. Use this tool when Mark is overwhelmed with how to proceed or needs a plan for how to address a given problem or topic.

reasoning_step – Articulates internal intermediate reasoning steps, plans, or observations for complex questions to help Mark continue multi-step work. This is particularly useful for complex questions that require careful consideration or decomposition before formulating a final answer.

analyze_image_content – Helps Mark understand and interpret the visual content of images, whether they are screenshots you provide or images Mark has found through a search. Mark can answer your questions based on what it "sees" in the image.

present_images – Displays generated or edited images to you as a rich visual card in the chat interface. Mark calls this tool after generate_or_edit_image when images are the final output.

invite_teammates – Displays a card in chat where you enter the email addresses and names of colleagues and invite them to your Mark instance with the Mark User role. Opti ID sends an invitation to each person who is new to your organization.

mount_space_to_workspace – Makes the files in a canvas space available to a code run, so Mark reads and edits them in place. Mounting reaches images and other binary files that canvas editing cannot change.

unmount_space_from_workspace – Detaches a mounted canvas space and removes its files from the workspace. The canvas space and its files remain in Mark. Save pending changes before you unmount, because Mark discards unsaved edits.