Get started with Optimizely Mark for administrators

Mark (Optimizely Agent Platform) is an agent orchestration platform that helps you work smarter across Optimizely One. Mark is automatically available and works by default, but you can configure it based on your organization's needs. You must add users to Mark before they can use it.

Add and manage users and their permissions for Mark in the Opti ID Admin Center. See manage users for instructions.

Give users one of the following roles:

If the available system roles do not provide the specific access you want to grant users, you can create a custom role in the Opti ID Admin Center. See Custom roles for Mark for what roles are available.

The Mark app is Mark's configuration dashboard, accessible through the product switcher in the global navigation.

Customize Mark to tailor its behavior, responses, and integrations to your organization's goals. Configure skills, agents, and tools to shape how Mark works with your team. This gives you a personalized experience tailored to your needs.

Customizations ensure that Mark understands your brand voice, can access the right data, and performs actions that align with your workflows.

Mark supports several customization options, depending on the level of control you need.

Customizing Mark is optional but provides more control and makes the platform more effective. The following are ways that customizing Mark empowers you:

The Mark app provides a centralized interface where you can access Mark's capabilities in one place. The Mark app UI supports everything you need to configure, customize, and use Mark effectively. Each page helps you focus on outcomes while Mark handles the technical details behind the scenes.

Chat is the conversational interface for Mark. It lets you interact with Mark using natural language to complete tasks, get insights, and automate workflows without needing to know the underlying systems or commands.

Ask Mark questions, make requests, or give instructions. Mark determines the right tools and agents to use to deliver accurate, actionable results. Whether you need to generate content, analyze data, or manage experiments, Mark chat provides a simple, intuitive way to work directly within Mark.

See Access and explore Mark chat for information.

Skills are the foundational context, rules, and behavioral guidelines that shape how Mark creates output and tailors it to your organization's unique needs.

See the skills overview for information.

Agents complete tasks for you in Mark. They interpret your requests, decide which tools to use, and deliver consistent results. The Agents page of the Mark App is where you can browse and install agents from the Agent Directory or create and manage specialized and workflow agents tailored to your organization’s needs. By configuring the right agents, you can streamline workflows and automate common tasks across your connected Optimizely products and external services.

See Agent overview for information.

Tools are the actions Mark uses to get work done. Each tool performs a specific task, such as creating a campaign, generating content, or uploading files. View the tools available to your organization, including system tools built by Optimizely and custom tools created by your team. Connect and configure the right tools to let Mark act on your behalf and integrate with your systems and workflows.

See Tools overview for information.

In Mark, tools, skills, and agents work together to complete tasks.

Understanding how these pieces connect helps you design effective workflows and get the most out of Mark.

Imagine Mark as a sophisticated smart home system that intelligently manages and automates various functions within a house. Each part of Mark can be divided into different parts of this smart home system.

Skills – The smart home's core programming and user preferences.

Skills are the fundamental rules, underlying code, and user-configured preferences that govern the entire smart home system. They dictate how all devices and routines behave, ensuring consistency, security, and personalization across the whole house.

Tools – Device interfaces, communication protocols, and core functions.

The tools are the "things" that smart devices and agents use to get things done. These are the capabilities that let devices perform their tasks and interact within the smart home ecosystem, and with external services.

Specialized agent – An individual smart device.

A specialized agent is like an individual smart device (a smart light bulb, a smart thermostat, a smart coffee maker, or a smart lock). Each device has a specific, well-defined function it performs. It receives a command (input) and performs its action (output).

Workflow agents – A smart home "routine" or "scene".

A workflow agent is like a smart home "routine" or "scene". This is a pre-programmed sequence of actions designed to achieve a larger, multi-step goal. It orchestrates multiple individual device actions in a specific order.

Start customizing Mark by connecting your Optimizely products on the Product Connections page.

See the Product connections section for more information.

When connections are in place, you can complete the following in Mark:

Click Settings to configure your organization's Mark instance.

Go to Settings > Appearance to configure the Font Size and Theme in Mark. Adjust these settings to improve readability or match your visual preferences. Changes apply to your Mark instance only and do not affect other users in your organization.

Adjust the font size

Adjust the theme

Go to Settings > Notifications to configure your preferences. For a description of each setting, see Notification settings.

Go to Settings > Chat to configure Mark chat for all users in your Mark instance.

The default model provider determines which AI model family Mark uses to generate responses in Mark chat. This default model provider is also the default for specialized agents, which can override it with their own provider.

Anthropic Claude is the default model provider for all instances. Instances that previously defaulted to Google Gemini were switched to Anthropic Claude, so select Google Gemini to use it again. To change the LLM provider, click the Default Model Provider drop-down list and select Anthropic Claude, Google Gemini, or OpenAI GPT.

See Model assignments by inference level for the model each provider uses at each inference level.

Advanced URL safety checks is a built-in security feature that screens every URL before Mark browses it, protecting your organization from malicious websites, including malware, phishing, and unwanted software.

Mark browses webpages on your behalf to summarize articles, extract data, or research topics. Before any content loads, the requested URL passes through a safety check. This check runs automatically on every browse request. Your end users take no action.

As a Mark administrator, choose one of two safety levels per instance, depending on the protection your organization requires.

URL Safe Browsing operates at two levels:

The Standard level checks every URL against Google Web Risk, a continuously updated database of known threats. It screens for the following threat categories:

Standard checks complete quickly and add minimal latency. Mark enables this level by default for every instance.

The Advanced level adds a Cloudflare Radar deep scan in addition to the Google Web Risk check. If either check flags a URL as unsafe, Mark blocks it immediately.

The deep scan provides enhanced detection for sophisticated threats that may not yet appear in Google's database. For URLs that Cloudflare has not scanned recently, the deep scan adds up to 30 seconds of latency.

When a URL is flagged as unsafe, Mark does not load or return any content from the page. Instead, the browse request returns a message that explains the block and names the threat category detected, such as MALWARE or SOCIAL_ENGINEERING.

Blocked URLs do not produce partial results. Mark stops the entire request before any page content is rendered.

Advanced URL safety checks also applies to links found in email bodies processed through Mark's email trigger system. Mark validates links in incoming emails before the email event is processed, providing an additional layer of protection against phishing and malware delivered by email.

Configure the safety level for each Mark instance. The setting applies to the entire instance, so all users within the instance share the same safety level.

The inference level controls which AI model Mark uses and how deeply the model reasons before responding. Higher inference levels pair more powerful models with deeper thinking to produce higher-quality output, but take longer and use more credits. Lower levels respond faster and use fewer credits, making them ideal for straightforward tasks. See Inference level.

To change the inference level used in Mark chat for your organization, click the Inference Level drop-down list.

Go to Settings > Email Domain to set the email subdomain prefix Mark uses when it sends emails. Only one email domain is supported per Mark instance.

To configure the email domain, complete the following:

Go to Settings > Product Connections to manage where Mark is activated in your specific Optimizely products and instances. These connections let Mark access the data, tools, and features it needs to take action on your behalf. If a product instance is not connected, it cannot use Mark features. Configure your connections first to ensure you can customize and fully use Mark across your Optimizely organization.

To add product connections, complete the following:

Mark lists the instances of a product in the same order throughout Mark: by nickname or name, ignoring capitalization, then by ID.

Go to Settings > RAG Configuration to control whether Retrieval-augmented generation (RAG) is used in direct messages in Mark chat. Agents can use RAG regardless of this setting.

RAG lets Mark access, understand, and query knowledge sources. With RAG enabled, Mark draws on data such as campaigns, content, and tests to provide more contextual and accurate responses. See Retrieval-Augmented Generation (RAG) overview for more information.

To change whether RAG can be used in Mark chat, toggle Enable RAG in Chat on or off.

If your firewall or security settings block Mark from scanning your URLs, add the following IP addresses to your allowlist:

Mark non-EU instances

Mark EU instances

Mark is enabled by default. To disable Mark, see Turn generative AI off across Optimizely applications.

View Mark usage across your organization in the usage and billing dashboards. See the Mark credits section in the Usage and billing dashboards documentation.

View Mark agent usage by agent ID (@name) in the Mark agent usage dashboard. See the Optimizely Reporting for Markdocumentation.