2026 Optimizely Agent Platform release notes

80 min

Subscribe to updates on this page using the notification and RSS feed options.

October 13

Agent Platform

Released Brand Packs in Optimizely Agent Platform, which build a comprehensive brand skill from your existing brand guidelines, templates, and portals so your brand team approves every text and image output from Mark. See Brand Pack overviewBrand Pack overview.

Added variants to the image editor, which generate versions of an image in different sizes for specific channels from a brand-approved template. See Image editorImage editor.

Released the skills directory, a single place to find and use skills that Optimizely built. See Skills directory overviewSkills directory overview.

Released the unified directory, the location to find and install agents, skills, and Virtual Teammates for your entire Mark instance. See Directory overviewDirectory overview.

Virtual Teammates

Released the Web Manager Virtual Teammate which keeps your website compliant. See Web ManagerWeb Manager.

Released the Content Marketer Virtual Teammate which ties your editorial plan to real audience signals. See Content MarketerContent Marketer.

Agents

Released the following agents to the Agent Directory:

Eye Tracking AuditorEye Tracking Auditor – Analyzes predictive eye-tracking maps from Optimizely Campaign to identify visual hierarchy issues in email designs, so you can improve layouts before you send.

Experiment Value EstimatorExperiment Value Estimator – Generates a stakeholder-ready report on the projected annualized impact of rolling out a winning experiment variation, which helps you communicate the business value of experimentation in dollar terms.

List CreationList Creation – Guides you through creating a global list or wishlist in Configured Commerce, which helps you configure targeted product lists without working across multiple admin screens.

Product Catalog AuditorProduct Catalog Auditor – Scans products and variants in Commerce Connect for data inconsistencies, missing attributes, and publishing gaps, which helps you fix catalog defects before products reach live storefronts.

Promotion CreationPromotion Creation – Guides you through creating promotions in Configured Commerce, including rules, results, and website assignments, which helps you configure promotions through conversation instead of across four admin tabs.

Smart Campaign DesignSmart Campaign Design – Reads a campaign brief and generates an Optimizely Campaign flow architecture, including node sequences, A/B test variants, and a master send schedule, which helps you move from brief to build faster.

WCAG Compliance AuditorWCAG Compliance Auditor – Audits Content Management System (CMS) (SaaS) content against Web Content Accessibility Guidelines (WCAG) standards and applies approved accessibility fixes, which helps you reduce accessibility risk before you publish.

Connectors

Added tools to the LinkedIn Ads connector that create and update campaign groups, campaigns, and creatives. They also upload media and create sponsored posts, so you can run LinkedIn campaigns without leaving Agent Platform. See LinkedIn Ads connector toolsLinkedIn Ads connector tools.

October 6

Connectors

Added Google Tag Manager connector tools, which connect Agent Platform to your Google Tag Manager account so agents can audit and manage tags, triggers, and variables, and publish container versions without leaving Agent Platform.

October 5

Released the Private executions setting for specialized agents, which limits the logs, outputs, and details of each run to the user who triggered it. This lets you use shared agents on sensitive work, such as unreleased campaigns or pricing, without exposing results to others. See Enable private executionsEnable private executions in the Create a specialized agent documentation for information.

September 30

Updated the underlying models for Agent Platform's inference levels, improving response quality. See Inference levelInference level for the full model mapping.

September 21

Released personal files, which give you a persistent document store in Agent Platform that agents and Virtual Teammates access automatically during task execution. See Personal files overviewPersonal files overview and Manage personal filesManage personal files for information.

September 18

Connectors

Added OneNote and Microsoft sensitivity label tools to the Microsoft 365 connectorMicrosoft 365 connector, which let agents manage OneNote notebooks and pages, and apply or check sensitivity labels on OneDrive or SharePoint files.

Added LinkedIn Ads connector toolsLinkedIn Ads connector tools, which connect Agent Platform to your LinkedIn Ads account so agents can plan campaigns, report on performance, and analyze ad engagement without leaving Agent Platform.

Added OpenAI Ads connector toolsOpenAI Ads connector tools, which connect Agent Platform to your OpenAI Ads account so agents can manage campaigns, ad groups, and creatives, and report on performance.

Added Global and European Union (EU) regional endpoints to the Contentful remote Model Context Protocol (MCP) connectorContentful remote Model Context Protocol (MCP) connector. Agent Platform administrators now choose between Contentful (Global) and Contentful (EU) to meet regional data residency requirements.

MCP connectors

Braze remote MCP connector – Added the Braze remote Model Context Protocol (MCP) connectorBraze remote Model Context Protocol (MCP) connector, which connects Agent Platform to your Braze account so agents can explore campaign analytics, segments, and custom attributes, and generate content such as email templates.

Fullstory remote MCP connector – Added the Fullstory remote MCP connectorFullstory remote MCP connector, which connects Agent Platform to your Fullstory account so agents can analyze sessions, funnels, and user journeys to identify friction and diagnose product issues.

September 17

Agent Platform platform

Released sub-agents, which Mark assigns automatically by task complexity so harder questions get more reasoning. Agent Platform manages the Mini, Fast, and Deep tiers on every instance automatically. See Sub-agents overviewSub-agents overview for information.

Released retrieval-augmented generation (RAG) for artifacts, which lets Mark search your canvasescanvases so Mark chat answers draw on your own work. Agent Platform respects artifact permissions and includes artifacts shared with you. See RAG for artifactsRAG for artifacts for information.

Workflow agents

Released the Code step for workflow agents, which runs Python, JavaScript, or Bash without an external service. See Workflow agent code stepsWorkflow agent code steps and Workflow agent code step examplesWorkflow agent code step examples for information.

Added a Run iterations in parallel toggle to the workflow agent Loop step, so independent iterations finish faster. See Set the loop execution modeSet the loop execution mode in the Workflow agent loops documentation for information.

Added a Priority setting to workflow agent triggers, so interactive runs are not delayed by automated batch runs. The options are Standard and Background. See Trigger priorityTrigger priority in the Workflow agent triggers documentation for information.

Optimizely Product Recommendations tools

Released the pr_delete_custom_page tool in Agent Platform to help you delete an existing custom page from your website. See Page and position managementPage and position management for information.

September 13

Updated the Standard, Balanced, and Complex inference levels to Gemini 3.8 Flash, which improves response quality and speed without any change to how you configure Agent Platform or your agents. See Model assignments by inference levelModel assignments by inference level for information.

September 2

Released the following Feature Experimentation agents to the Agent DirectoryAgent Directory in Optimizely Agent Platform:

Feature Experimentation Governance agentFeature Experimentation Governance agent – Analyzes an Optimizely Feature Experimentation project and generates a read-only HTML governance report on a canvas.

Feature Flag Implementation agentFeature Flag Implementation agent – Generates a portable SKILL.md file to implement an Optimizely Feature Experimentation flag in a codebase.

August 27

Released the following agents to the Agent DirectoryAgent Directory:

Restriction Group Creation agentRestriction Group Creation agent – Creates product restriction groups in Optimizely Configured Commerce through a step-by-step wizard. This helps you define customer and product restrictions accurately without hand-building API requests.

Page Builder (PaaS) agent (CMS 13)Page Builder (PaaS) agent (CMS 13) – Analyzes a document, image, or URL and creates a content item of a specified content type in Optimizely Content Management System (CMS) 13 (PaaS). The agent helps you move from ideation to publish-ready CMS content faster and keep content structured according to Optimizely CMS best practices.

Content Analysis agent (CMS 13)Content Analysis agent (CMS 13) – Analyzes existing content of a specified CMS 13 (PaaS) content type and produces a markdown skill for creating content of that type. The agent helps you create consistent, on-pattern CMS (PaaS) content and reduce rework caused by inconsistent content structures.

August 25

Released Virtual Teammates, AI agents that work as first-class Optimizely users with their own Opti ID, persistent memory, and the ability to act across products, which lets you delegate ongoing work instead of re-prompting an agent each time. A Virtual Teammate acts as itself in Optimizely products and on behalf of its principal user in other products. See Virtual Teammates overviewVirtual Teammates overview for more information.

Released the Chief of Staff Virtual Teammate, which prepares you for meetings, delivers daily briefings, and helps you prioritize your work. It pulls together your calendar, email, and messages so you start each day with the highlights that matter. See Chief of Staff Virtual TeammateChief of Staff Virtual Teammate for more information.

Released the Marketing Analyst Virtual Teammate, which turns raw Google Analytics 4 (GA4) data into short, decision-ready reports on a set schedule, so you no longer log into GA4 to pull numbers by hand. See Marketing Analyst Virtual TeammateMarketing Analyst Virtual Teammate for more information.

Released the SEO & AI Search Analyst Virtual Teammate, which helps you stay visible across both human and AI search traffic. It reasons across Google Search Console, Google Analytics 4, Conductor, and Optimizely AI Visibility Analytics to explain why your performance changed and recommend a fix. See SEO & AI Search Analyst Virtual TeammateSEO & AI Search Analyst Virtual Teammate for more information.

Released the CRO Manager Virtual Teammate, a conversion rate optimization assistant that accelerates experimentation and personalization by monitoring user behavior, analyzing competitor UX patterns, and staging experiment configurations directly in Optimizely. See CRO Manager Virtual TeammateCRO Manager Virtual Teammate for more information.

Released the Personalization Strategist Virtual Teammate, which builds and runs personalization campaigns end-to-end and reports on their impact, so you can create and scale campaigns with less manual effort. See Personalization Strategist Virtual TeammatePersonalization Strategist Virtual Teammate for more information.

Released Team Messages, a real-time messaging layer in Agent Platform where you direct, monitor, and work alongside Agent Platform agents and human teammates across products. You can start group chats and channels, share files, agents, and artifacts, and use mentions to reach the right people. See Team Messages overviewTeam Messages overview for more information.

Released Brands, which makes every Mark-generated artifact, including decks, documents, and social images, come out on-brand by default, so you can share what Mark produces without restyling each output. An administrator configures the organization's brand once, and a marketer can bootstrap one in the moment when no brand exists yet. See Brands overviewBrands overview for more information.

Added auto-update for agents installed from the Agent Directory. Installed agents now update automatically when Optimizely publishes a new version, so administrators no longer need to push updates manually. Auto-update is on by default and applies to both specialized and workflow agents. See Agent Directory overviewAgent Directory overview for information.

August 14

Agent Platform platform

Upgraded the Complex and Code inference levels to run on claude-opus-5. Agent Platform maps each inference level to a model for you, so no action is needed. See Inference levelInference level for the full level-to-model mapping.

Optimizely Graph tools

Released the following tools in Optimizely Agent Platform to help you manage your Optimizely Graph content in Optimizely Content Management System (SaaS), CMS 12 (PaaS), and CMS 13 (PaaS).

graph_pinned_result – Manages pinned results and pinned result collections in Optimizely Graph.

graph_synonyms – Manages synonyms in Optimizely Graph. Synonyms help you find content using alternative terms.

July 29

Agent Platform platform

Released Agent BuilderAgent Builder and Skill BuilderSkill Builder, which let you configure, build, and iterate on Agent Platform agents and skills directly in Mark chat without switching tools.

Added Safe URL BrowsingSafe URL Browsing, which screens every URL before Mark acts on it to block malware, phishing, and unwanted software. Screening uses always-on Google Web Risk protection, with an optional Cloudflare Radar deep scan.

Added Version HistoryVersion History, which tracks every change made to an agent workflow so you can see what changed, when, and by whom.

Improved execution status reporting with a new Execution Status TaxonomyExecution Status Taxonomy, so you can see exactly what is running, what failed, and why without digging through logs.

Added the ability to automatically and manually Compact a conversationCompact a conversation in Mark chat. This lets Mark chat automatically save checkpoints so it can automatically get context from earlier in the conversation.

Released input guardrailsinput guardrails, which give you control over what users ask Mark. You define rules in plain language, and Agent Platform trains a classifier to enforce them automatically. This helps keep Mark responses compliant and on topic.

Workflow agent updates

Added Nested WorkflowsNested Workflows, which let you nest a workflow agent inside another to build layered, multi-step automations.

Added static and predicted parametersstatic and predicted parameters, which let you set each workflow variable automatically with the large language model (LLM) or lock it to a static value, so you control how each step runs.

Added an Agent Update WarningAgent Update Warning that alerts you before you edit or delete an agent belonging to a workflow, so you can review the impact first.

UX improvements

Added the ability to edit sent messages in Mark chatedit sent messages in Mark chat, so you can fix a typo or refine a prompt without starting the conversation over.

Added a rotating tooltiprotating tooltip above the chat box that surfaces best practices and helpful features as you work.

Agents

Added the Google Ads Account Auditor AgentGoogle Ads Account Auditor Agent, which flags wasted spend and broad-match issues, then returns prioritized actions sorted by timeframe and impact.

Connectors

Added support for remote MCP serversremote MCP servers, which let you bring your own Model Context Protocol (MCP) servers into Agent Platform to extend it with external tools and data.

Added three connectors that bring external data into your agent workflows:

Google Workspace connector toolsGoogle Workspace connector tools – Reads, drafts, and sends email from inside Agent Platform.

HubSpot remote MCP connectorHubSpot remote MCP connector – Connects your customer relationship management (CRM) data so agents can act on your CRM records.

ZoomInfo remote MCP connectorZoomInfo remote MCP connector – Brings B2B intelligence into your projecting and outreach agents.

July 22

Agent Platform platform

Updated the IP addresses Agent Platform administrators can allowlist if Agent Platform is blocked by your firewall or security settings. See Allowlist Agent PlatformAllowlist Agent Platform.

Tools

Released the Google Search Console connector toolsGoogle Search Console connector tools which let you access and analyze your site's Google Search performance data directly within Agent Platform, without switching between platforms.

Released additional System tools for Product RecommendationsSystem tools for Product Recommendations to retrieve algorithm performance reports for a date range to compare which recommendation algorithms drive the most revenue.

June 30

Opal Agent Usage dashboard

Released a new version of the Opal Agent Usage dashboard, which gives organizations a centralized dashboard for Agent Platform agent usage so administrators can track executions, monitor credit consumption, and see which agents and users are most active. The updated dashboard displays summary cards, an execution-type breakdown, agent and top-user tables, and weekly active-user trends over your selected time period. See Optimizely Reporting for Agent PlatformOptimizely Reporting for Agent Platform for more information.

Agent Platform platform

Released the Agent LibraryAgent Library, a browsable catalog on the Mark chat homepage where you can find and launch more than 45 pre-built agents in one click, so discovering the right agent no longer means knowing where to look.

Released Personalize MarkPersonalize Mark, a pinned onboarding agent that walks you through connecting your tools and building skills in a guided, role-based session, so you get value from Chat in your first few minutes.

Added agent sharing and visibility controls, which let you keep an agent private or share it with named people or your whole organization, with permission levels that govern who can view, run, edit, or delete each agent. New agents default to private on instances where the feature is enabled (not available on all tiers). See Share agents and manage visibilityShare agents and manage visibility.

Released an image editor in Mark chat, which lets you refine a generated image without starting over by detecting and selecting visual elements, transforming them with brush controls, queuing batch edits, and applying smart aspect ratios, so you generate and fine-tune images in one place. See Image generationImage generation.

Released the Skills builderSkills builder for general availability, which lets you create and manage reusable skills that agents draw on, so you tailor Agent Platform to your own workflows.

Added the ability to export workflow agent execution logs to comma-separated values (CSV) or Excel with date-range filtering, and added the same date-range selection to specialized agent log exports, so you can pull an audit trail on demand for performance tracking and compliance. See Workflow agent logsWorkflow agent logs and Specialized agent logsSpecialized agent logs.

Agents

Released the following agents:

Meeting Notes Summary agentMeeting Notes Summary agent – Turns raw notes or a transcript into a shareable one-page summary of decisions, owner-assigned action items, and open questions, so nothing from a meeting gets lost.

Meeting Follow-up Email agentMeeting Follow-up Email agent – Turns meeting notes into a ready-to-send email covering decisions, action items with named owners, and next steps, so you send a clear recap in minutes.

Meeting Preparation agentMeeting Preparation agent – Turns your agenda, prior notes, and attendees into a one-page brief you can read in under two minutes, covering the goal, key context, talking points, and decisions to make, so you walk in prepared.

Newsletter Body Editor agentNewsletter Body Editor agent – Drafts engaging email body copy from a topic, audience, and brand voice, so you go from brief to on-brand draft in one step. See

Email Ideation agentEmail Ideation agent – Brainstorms five to seven newsletter themes and content angles tailored to your niche and goals, so you start from ideas instead of a blank brief.

Email Subject and Preview Text Generation agentEmail Subject and Preview Text Generation agent – Generates paired subject lines and preview text optimized for specific psychological triggers, so you have ready-to-test copy that can lift open rates.

Email Performance Comparison agentEmail Performance Comparison agent – Compares past and current email campaigns across key performance indicators (KPIs) to surface what drove results, so you can repeat what works in your next send.

LinkedIn Ad Brief Generation agentLinkedIn Ad Brief Generation agent – Builds a complete LinkedIn ad brief from a product name and URL, with copy variants, ranked call-to-action (CTA) options, and an optional competitor benchmark that all fit LinkedIn's character limits, so copy is ready to paste into Campaign Manager.

Presentation Outline Generation agentPresentation Outline Generation agent – Turns a topic, audience, and goal into a slide-by-slide outline with key messages, suggested content, and speaker notes, so you build a deck from structure instead of a blank page.

Webinar Outline agentWebinar Outline agent – Turns a topic and optional notes into a ready-to-run outline with timed sections, talking points, engagement moments, and pre- and post-event checklists tuned to your goal and duration, so you can plan a webinar end to end.

Social Media Image Generation agentSocial Media Image Generation agent – Generates correctly sized, platform-ready images for LinkedIn, Meta, X, display ads, and email from a text prompt or a reference image, so visuals stay on-brand without resizing guesswork.

Image Resizer agentImage Resizer agent – Resizes an uploaded image to exact specifications for LinkedIn, Instagram, Meta, X, Google Display, email headers, and web banners from preset names or custom dimensions, so every format comes back ready to use in one place.

Image Text Remover agentImage Text Remover agent – Removes overlaid text such as captions or burnt-in subtitles from an uploaded image and fills the cleared area with the surrounding background, so you can reuse an image without its original text.

Image Metadata Ideation agentImage Metadata Ideation agent – Analyzes email content from Optimizely Campaign to suggest accessible alt text and titles for images, so your emails stay accessible and compliant. Requires Optimizely Campaign.

Case Study Generation agentCase Study Generation agent – Produces a publication-ready B2B SaaS case study from a company name and background material such as interview notes or a transcript, complete with cited metrics, verbatim quotes, and before-and-after results, so you turn raw inputs into a polished study fast.

Image Alt Text Generation agentImage Alt Text Generation agent – Audits up to 20 images on any page URL, scores existing alt text against Web Content Accessibility Guidelines (WCAG) 2.1 AA, and returns ready-to-use copy with a prioritized fix list, so you meet accessibility standards and strengthen search engine optimization (SEO) without checking images by hand.

Agent Visibility Analytics Insights agentAgent Visibility Analytics Insights agent – Examines your Agent Visibility Analytics dashboard to surface trends, anomalies, and recommended actions based on a chosen analysis framework, so you find insights without digging through the data yourself.

Page Builder agentPage Builder agent – analyzes a document, image, or URL and creates a CMS (SaaS) content item of a chosen content type following best practices, so editors move from idea to finished content quickly.

Content Analysis agentContent Analysis agent – Analyzes a content type to find structural patterns and best practices and returns a set you can feed into the Page Builder agent, so generated content matches your CMS (SaaS) conventions.

Content Model Creation agentContent Model Creation agent. – Analyzes URLs or images to build best-practice content models and references existing models to prevent duplicates, so your content architecture stays consistent in CMS 13 (PaaS).

GEO Schema Optimization agentGEO Schema Optimization agent – Analyzes page content and recommends or generates schema types and JSON-LD markup, so your content in CMS 13 (PaaS) is more discoverable across search engines and large language models (LLMs).

SEO Metadata Optimization agentSEO Metadata Optimization agent – Analyzes webpages and generates optimized meta titles, descriptions, and OpenGraph tags, then applies them as drafts in CMS 13 (PaaS), so you improve search rankings without manual edits

Conductor AEO Gap Finder agentConductor AEO Gap Finder agent – Finds high-priority Answer Engine Optimization (AEO) content gaps by analyzing your website against tracked competitors.

Conductor AI Brand Visibility Report agentConductor AI Brand Visibility Report agent – Delivers a snapshot of your brand's visibility across ChatGPT, Gemini, and Perplexity.

Conductor Competitive AI Share of Voice agentConductor Competitive AI Share of Voice agent – Compares your brand against tracked competitors across ChatGPT, Gemini, and Perplexity. Share of voice measures how often AI engines mention your brand compared to competitors.

Tools

Added Zoom MeetingsZoom Meetings connector tools, which connects Agent Platform to your Zoom meetings so agents can work with meeting content.

Added additional system tools for Optimizely Data Platform (ODP)system tools for Optimizely Data Platform (ODP).

CMS 13 Tools

Released tools to help you work with display templates and upload media. See Display template toolsDisplay template tools and Media toolMedia tool.

paas_cms_list_display_templates – Lists all display templates. Takes no parameters.

paas_cms_get_display_template – Retrieves one or more display templates by key.

paas_cms_create_display_template – Creates a display template. Scope the template to exactly one of NodeType, BaseType, or ContentType.

paas_cms_update_display_template – Updates an existing display template. Only provided fields are updated. Omitted fields are unchanged.

paas_cms_upload_media – Downloads a file from a publicly accessible URL and uploads it as a media asset in CMS.

June 24

Agent Platform platform

Added bulk delete, which lets you select multiple chat threadschat threads, specialized agentsspecialized agents, workflow agentsworkflow agents, or skillsskills and remove them in one action.

Added model provider selection at the instanceinstance and specialized agentspecialized agent level, letting you choose the model that best fits each use case.

Mark chat

Updated the activity stream so you can follow what Mark does while it responds:

Added inline tool status, which shows the tools Mark uses in real time.

Added context pickup pills, which show the context Mark pulls into a response. See Access and explore Mark chatAccess and explore Mark chat.

Added a keyboard shortcut that stops a response when you press Escape, so you can halt Mark without clicking Stop. See Stop a responseStop a response.

Agents

Added execution-log export for specializedspecialized and workflowworkflow agents, so you can export a custom date range to analyze run history.

Added skill preview in the agent editor, so you can review what a skill does before adding it. See SkillsSkills in the Prompt details documentation.

Skills

Added skill watching, so you receive a notification when another user updates or deletes a skill you watch. See WatchersWatchers.

Added skill import and export, which lets you reuse skills across instances or keep an external backup. See Import and export skillsImport and export skills.

Added a scope switcher to the skills, which lets Opal Administrators move a skill between Personal and Organization without recreating it. If the shortcut is already in use, Agent Platform appends -personal or -org so it stays unique. See Switch a skill between personal and organizationSwitch a skill between personal and organization.

Added file attachment support to the skill creation tool, so agents can attach files when building skills for you. See the Skills managementSkills management tools.

RAG for CMP

Improved retrieval-augmented generation (RAG) asset search for Content Marketing Platform (CMP), which lets you filter assets by library folder name or pathfilter assets by library folder name or path. Mark returns content from the folder you specify, giving more precise results.

Added relevancy explanationsrelevancy explanations to the assets Mark suggests from your content library. Each explanation shows why Mark selected the asset, helping you decide whether it fits your request.

Released RAG support in the European Union (EU) region. Customers with EU data residency requirements can now use RAG while Optimizely keeps their CMP data within EU-based infrastructure. See Retrieval-Augmented Generation (RAG) overviewRetrieval-Augmented Generation (RAG) overview for information.

June 8

PowerPoint and Word document creation – Released the ability for Mark to read, create, and edit .docx and .pptx files when you enable the code_execution tool. Create and update on-brand PowerPoint and word documents without leaving Agent Platform. See Generate PowerPoint presentations and Word documents with MarkGenerate PowerPoint presentations and Word documents with Mark.

June 4

Agent Platform

EU app hosting – Added an EU region for Agent Platform hosting so European customers can keep data within the EU. See Data residencyData residency for details.

Notification types – Added five notification types under new Skill updates and Guardrails categories. Stay informed about changes to skills your agents use and quality issues with agents you watch. See Notifications overviewNotifications overview for details.

Permissions for sharing canvases – Added space-level sharing permissions to canvases. Grant teammates view, edit, or manage access to a canvas without leaving Agent Platform. See Manage canvasesManage canvases for details.

Limitless 1:1 Personalization – Released Limitless 1:1 Personalization, which uses Mark to generate landing pages for each target account or audience segment. Scale mid-funnel content without manual production limits. See Limitless 1:1 PersonalizationLimitless 1:1 Personalization for details.

Agents

Experimentation Program Overview agent – Released the Experimentation Program Overview agent, which generates a Quarterly Business Review-style report on your experimentation program performance. Identify successful strategies, areas for improvement, and stakeholder-ready insights. See Experimentation Program OverviewExperimentation Program Overview for details.

Experiment Value Estimator agent – Released the Experiment Value Estimator agent, which projects annualized impact and revenue lift from winning experiments. Quantify return on investment and support confident shipping decisions for winning variations. See Experiment Value EstimatorExperiment Value Estimator for details.

Experiment Backlog Prioritization agent – Released the Experiment Backlog Prioritization agent, which scores testing ideas using the Potential, Importance, Ease (PIE) framework. Get a ranked plan that eliminates subjective decision-making and identifies high-impact ideas first. See Experiment Backlog PrioritizationExperiment Backlog Prioritization for details.

Standard Audience Builder agent – Released the Standard Audience Builder agent, which builds and troubleshoots standard Optimizely Data Platform (ODP) audiences through natural language. Translate plain-language goals into validated audience configurations and remove the technical barriers of audience creation. See Standard Audience BuilderStandard Audience Builder for details.

Real-time Audience Builder agent – Released the Real-time Audience Builder agent, which creates, updates, and troubleshoots Optimizely Data Platform (ODP) real-time audiences through natural language. Target customers in real time without manual configuration. See Real-time Audience BuilderReal-time Audience Builder for details.

Connector tools

Microsoft Teams tools – Added Microsoft Teams tools to Agent Platform. Create chats, send messages, reply to channels, and check team member availability without switching apps. See Microsoft 365 connectorMicrosoft 365 connector for details.

Microsoft Outlook tools – Added Microsoft Outlook tools to Agent Platform. Manage email, schedule calendar meetings, check attendee availability, and configure mailbox settings without switching apps. See Microsoft 365 connectorMicrosoft 365 connector for details.

Google Ads tools – Added Google Ads tools that connect Agent Platform to your Google Ads account. Review campaigns, performance, and ad data through Mark chat without switching tools. See Google Ads toolsGoogle Ads tools for details.

MCP connectors

Gamma remote MCP connector – Added the Gamma remote Model Context Protocol (MCP) connector. Generate AI-driven presentations and webpages, customize templates, and manage folders without leaving Agent Platform. See Gamma remote MCPGamma remote MCP for details.

GitHub remote MCP connector – Added the GitHub remote Model Context Protocol (MCP) connector. Search repositories, read source code, manage issues and pull requests, and automate development tasks without leaving Agent Platform. See GitHub remote MCPGitHub remote MCP for details.

CMS (SaaS) tools

Decommissioned CMS GEO tools – Removed cms_geo_apply and cms_geo_analysis, which the CMS content tools now replace. Update any agents or workflows that referenced them. See System tools for Content Management System (SaaS)System tools for Content Management System (SaaS) for details.

June 1

Quality tab – Released the Quality tab for specialized agents. Use it to score every run, detect drift from established behavior, and recover runs that go wrong. This means you can scale agents across your organization without manually reviewing every execution. The Quality tab contains the following three features:

Output Evaluation uses an AI model to score each run against criteria you define. Configure up to 10 evaluation criteria and a baseline evaluation score between 75% and 95% in 5% increments. Provide up to five output examples or conversation samples that represent expected quality.

Execution Guardrails detect runs that deviate from your agent's normal behavior. The guardrails progress through Learning, Watching, and Enforcing modes based on successful runs observed. You control sensitivity and whether Agent Platform stops flagged runs automatically.

Execution Advisor recovers flagged runs with corrective guidance and retries failed tool calls with adjusted input. It also reviews sensitive tool calls (such as email, publish, or delete) with inline Approved or Blocked verdicts.

See Quality tab overviewQuality tab overview for details.

May 27

Added a Skills section to the specialized agent edit and create pages. You can now apply reusable organizational context, such as brand voice, product taxonomies, and compliance language to your agent without rewriting prompts. Mark selects organization skills automatically based on your agent's prompt or tools. You can also insert specific skills at exact points in your prompt using the Choose skills picker. The picker lists every organization skill that has a shortcut. See Create a specialized agentCreate a specialized agent for details.

May 26

Mark chat

Added automatic scroll to the latest message. When you reopen a previous conversation, Agent Platform now scrolls to the most recent message so you can pick up where you left off. See Access and explore Mark chatAccess and explore Mark chat.

Improved the Stop button responsiveness after sending Mark a message. The Stop button now responds immediately when you click it, so you can interrupt a running agent without delay. See step 17 in through the Agent Platform websitethrough the Agent Platform website or step 14 in an Optimizely productin an Optimizely product. Also see Stop a responseStop a response.

Canvas

Added a raw markdown editing mode. You can now toggle to a raw source view on markdown files in Canvas and edit the underlying markdown directly to control your document structure. See steps 3 and 4 in the text/markdowntext/markdown section.

Added a loading state for Canvas files. Canvases now display a loading skeleton while your files load, so you know files are loading instead of seeing an empty state. See Canvas overviewCanvas overview.

Agents

Added View Agent button in workflow agent execution logs. When viewing a workflow execution, you can now click through to the agent editor to view or modify the agent's configuration, making it easier to iterate on your workflows. See step 17 in the completed and failedcompleted and failed section of the Workflow agent logs documentation.

Improved workflow loop validation. The workflow editor now prevents you from creating multiple starting steps inside a loop, catching a common configuration error before you run your workflow. See Workflow agent loopsWorkflow agent loops.

Added a start time and a no-end-date option for scheduled workflows. When configuring a recurring schedule, you can now specify when the schedule begins and optionally let it run indefinitely, so the workflow runs on the cadence you need without manual end-date management. See Configure a scheduled triggerConfigure a scheduled trigger.

Added array and list input variables. Specialized agents can now accept array or list inputs, so you can pass multiple values to a single parameter when running an agent. See step 3 in the Define input variablesDefine input variables section.

Added filters for agent creator and modifier. You can now filter the agents list by who created or last modified an agent. The lists sort alphabetically and surface agents you created at the top. See Find your agentFind your agent.

Added raw markdown editing for agent prompts. You can now toggle to raw markdown mode when editing agent prompts to control formatting directly. See Core skillCore skill.

Skills

Added a code-editor-style file explorer for skill files. The skill file upload experience now uses a file explorer widget, making it easier to manage multiple files attached to a skill. See step 9 in Create an organization skillCreate an organization skill or Create a personal skillCreate a personal skill.

Updated the skill creation screen. The skills interface now has a clearer layout and helper text, making it easier to find the skill you need. See Skills overviewSkills overview.

Added support for PowerPoint and Word files. Skills can now work with PPTX and DOCX files, so you can process more document types with code execution. See step 6 in Components of a skillComponents of a skill.

Notifications

Launched browser notifications. You now receive browser notifications when important events happen in Agent Platform, such as when an agent you are watching is updated or a scheduled workflow completes, so you do not need to keep the Agent Platform tab open to know when things finish. See Enable browser notificationsEnable browser notifications.

May 22

CMS 13

Released tools to help you create and manage content and content types, optimize for search engines, and improve AI discoverability. See Available tools for CMS 13Available tools for CMS 13.

paas_cms_list_content_types – Lists all content types in CMS.

paas_cms_get_content_type_details – Retrieves detailed information about a specific content type, including all property definitions with types, validation rules, and group assignments.

paas_cms_create_content_type – Creates a content type with a key, display name, base type, and property definitions.

paas_cms_update_content_type – Modifies an existing content type's display name, description, or property definitions.

paas_cms_delete_content_type – Deletes a content type from CMS. The content type must not have any existing content instances.

paas_cms_list_property_groups – Lists all property groups in the CMS with their names and sort order.

paas_cms_create_property_group – Creates a property group with a unique key and optional sort order.

paas_cms_get_content_data – Retrieves a content item's data including all non-metadata properties. Fetches a specific version or the latest published version.

paas_cms_update_content_item – Updates properties on a content item. Accepts a JSON object of property name and value pairs. If no contentkey is specified, creates a content item as a draft. Returns a contentKey (GUID) and contentVersion for subsequent operations.

paas_cms_delete_content_item – Moves a content item to the waste basket. An editor can restore the item.

paas_cms_publish_content_item – Publishes a content item immediately or schedules it for a future date and time. Requires a content version number.

paas_cms_create_content_variation – Creates a locale variation of a page or experience content item.

paas_cms_get_content_preview_url – Generates a preview URL for a specific content version, letting Mark fetch and analyze the rendered HTML.

paas_cms_seo_analysis – Analyzes content SEO performance and provides AI-powered recommendations.

paas_cms_seo_edit – Applies SEO recommendations to CMS content.

paas_cms_geo_analysis – Performs an AI-powered audit of content for LLM discoverability.

paas_cms_geo_apply – Applies JSON-LD schema templates to the content's JsonLdTemplates property. Creates a draft version with the schema markup.

May 18

Released the following agents to the Agent DirectoryAgent Directory:

Experiment Backlog PrioritizationExperiment Backlog Prioritization – Transforms your list of testing ideas into a ranked, actionable plan. It scores each idea against the PIE (Potential, Importance, Ease) framework.

Experiment Value EstimatorExperiment Value Estimator – Transforms your winning test results into a clear, stakeholder-ready business case. It calculates projected annual impact, quantifies revenue lift, and applies an optional conservative discount to anchor expectations

May 14

Optimizely Product Recommendations

(Beta) Released Mark chat for Product Recommendations. Agent Platform is an AI assistant that provides contextual guidance directly within the product. Take Mark's help to manage high-level workspaces, retrieve algorithms, define website layouts, manage recommendation blocks, and analyze the impacts of your product recommendations. See Use Optimizely Agent Platform with Product RecommendationsUse Optimizely Agent Platform with Product Recommendations.

May 11

CMS (SaaS) Agent Platform tools

Use the CMS (SaaS) Agent Platform toolsCMS (SaaS) Agent Platform tools to list, view, create, and update stylesstyles for rows, columns, sections, and content types, directly through Mark chat. Combined with Mark's existing ability to create and manage CMS content, you can now handle the full content creation and editing experience without leaving Agent Platform.

List available styles – Ask Mark to return all styles available in your instance, including what content they apply to, whether they are set as default, and a summary of their settings.

View style details – Ask Mark for the full details of any style, including its settings and available choices.

Create a style – Describe the style you want, including its name, what it applies to, settings, and choices. Mark creates it for you.

Update an existing style – Change a style's name, default state, or settings without rebuilding it from scratch.

May 7

Agent Platform

Released memorymemory so Mark remembers your preferences and past interactions, and each output builds on the previous one without you repeating yourself.

Released the Artifacts pageArtifacts page, which organizes everything you created in Agent Platform in one searchable location so you can find earlier work without manually tracking it down.

Added notificationsnotifications so you can respond to agent changes and execution failures without checking execution logs or monitoring long-running agents.

Added multi-turn conversation modemulti-turn conversation mode so you can iterate with specialized agents in Mark chat until the output matches what you need.

Added chat pillschat pills, which are pre-populated prompts you can run with a single click when you open Agent Platform.

Released action cardsaction cards which are rich, inline components that replace plain text responses with something you can see, interact with, and act on right inside Mark chat.

Added Content Marketing Platform (CMP) context with RAG so Agent Platform has full context of your campaigns and content. You can ask questions in Mark chat, surface assets, and get recommendations tailored to your work. See RAG for Content Marketing PlatformRAG for Content Marketing Platform for more information.

Added Chat modesChat modes, so you can match Mark's output depth to your task.

Renamed Instructions to Skills in the Agent Platform UI. Any agent instructions you previously authored now display in Context > Skills with no change in behavior. See Skills overviewSkills overview.

Added the ability to importimport and exportexport workflow agents directly so you can share and scale agents across your Agent Platform instances.

Added Idea BuilderIdea Builder, which generates test ideas informed by your previous experiment results. (Web Experimentation only).

Agents

Added the Page Performance Evaluation agentPage Performance Evaluation agent, which runs a Lighthouse audit and visual screenshot analysis on any live URL and identifies what is affecting your scores along with specific fixes.

Added the Competitor PageSpeed Analysis agentCompetitor PageSpeed Analysis agent, which runs Lighthouse audits on your page and a competitor's page and returns a side-by-side gap analysis with prioritized actions.

Added the Page Conversion Optimization agentPage Conversion Optimization agent, which audits any live page and returns a list of fixes ranked by conversion impact so you know what to address first.

Added the Page Copy Optimization agentPage Copy Optimization agent, which analyzes your live page copy and generates versions optimized for clarity, conversion, and market fit.

Added the Experiment Conflict Checker agentExperiment Conflict Checker agent, which checks a new test against every live experiment in Web Experimentation, Performance Edge, or Personalization before you launch it, so audiences and variables do not overlap.

Tools

Dynamic tool lookup – Removed the 128-tool limit on dynamic tool lookups, so Agent Platform agents can use all of your configured dynamic tools in a single conversation instead of hitting a cap.

Google Workspace tools – Added connector tools that give Agent Platform access to data from your Google WorkspaceGoogle Workspace, so you can create, update, and act on documents, spreadsheets, presentations, and calendar events without leaving Mark chat. Supported tools include Google Docs, Google Sheets, Google Slides, and Google Calendar.

Remote MCP servers – Added remote MCP servers that connects Agent Platform to the tools your team already uses, including content platforms, project management, and analytics. Supported integrations include the following:

NotionNotion

ContentsquareContentsquare

ContentfulContentful

Gamma remote MCP connectorGamma remote MCP connector

GitHub remote MCP connectorGitHub remote MCP connector

SanitySanity

ClickUpClickUp

RAG tools – Released the following Retrieval-Augmented Generation (RAG) system toolsRetrieval-Augmented Generation (RAG) system tools:

rag_retrieve_from_cmp – Searches your Optimizely CMP data, campaigns, tasks, assets, work requests, and briefs, using semantic similarity and SQL-based queries. Returns results as structured data for use in conversation responses.

rag_retrieve_from_cmp_as_csv – Searches your Optimizely CMP data, campaigns, tasks, assets, work requests, and briefs, using semantic similarity and SQL-based queries. Returns results as tabular CSV data.

rag_retrieve_from_optimizely_graph – Search for relevant information in your CMP and CMS data using RAG for Optimizely Graph. The tool provides powerful search, filtering, and content delivery capabilities for your applications. The tool can return both structured documents (like content items) and unstructured analytical data, depending on your query.

April 30

Workflow agents

Added new options to the workflow agent menu for downloading and copying workflow agents as JSON. See Manage workflow agentsManage workflow agents.

Added trigger details and agent information to the workflow execution log for improved visibility into run behavior. See Workflow agent logsWorkflow agent logs.

Added an option to configure case sensitivity for condition evaluation in workflow agents. This lets you control whether conditions are matched exactly or regardless of letter casing. See step 8 in Workflow agent conditionsWorkflow agent conditions.

Moved the View Agent option in the workflow panel from the footer to the top of the panel for quicker access. See Create a workflow agentCreate a workflow agent.

Improved error messages across workflow agent operations to help you resolve issues faster. For example, validation now clearly indicates when a condition value is missing before saving, or when an email address is already in use. See Create a workflow agentCreate a workflow agent.

Mark chat

Updated Optimizely Chat to automatically focus the input field on load, so you can start typing immediately without clicking into the field first. See Access and explore Mark chatAccess and explore Mark chat.

Added the ability to pin conversations to the top of the sidebar for quick access to frequently used threads. See step 7 in Through the Agent Platform websiteThrough the Agent Platform website and step 9 In an Optimizely productIn an Optimizely product.

Updated Agent Platform to collapse streaming responses by default during generation, reducing visual noise while content loads. See Access and explore Mark chatAccess and explore Mark chat.

April 22

Optimizely Web Experimentation

Added Idea builder, a dedicated tool in Web Experimentation that generates experiment ideas using Agent Platform. Provide context about your page, goal, and research, and Mark returns specific, actionable test ideas you can save, refine, and act on. Optional inputs such as heatmap images, qualitative research documents, or analytics notes further improve the quality of generated ideas. See Idea builder for Optimizely Web Experimentation Idea builder for Optimizely Web Experimentation for information.

April 14

Mark chat

Improved chat history sorting so your most recently active conversations display at the top, making it faster to resume recent work without scrolling. See Access and explore Mark chatAccess and explore Mark chat.

Added undo support for pasted content in the chat input, so you can quickly revert accidental pastes without manually deleting content. See step 12 in the Mark chat user interfaceMark chat user interface.

Added message queuing in chat, which lets you type and send follow-up messages while Mark is still responding. Messages display as queued cards and send automatically once the current response completes, so you can stay in your flow without waiting. See Queue messages while Mark worksQueue messages while Mark works.

Agents

Added the ability to export agents as JSON files, making it easy to back up or share your agent configurations across Agent Platform instances. See Download or copy a specialized agent as JSONDownload or copy a specialized agent as JSON for information.

Added JSON file import for agents, so you can restore or transfer agent configurations across instances. See Configure a specialized agent from JSONConfigure a specialized agent from JSON for information.

Improved the workflow agentworkflow agent editor to automatically center on the trigger node when you open a workflow, giving you a clearer starting point.

Added helper message support for chat-triggered workflows, which lets workflow authors provide guidance text that appears when the workflow is activated. See step 3 in Configure a chat input triggerConfigure a chat input trigger.

Improved workflow builder UI workflow builder UI consistency with better sidebar layouts, node sizing, and icon updates.

Skills

Improved skills disambiguation so that when you ask Mark to edit a skill from chat, it correctly identifies which skill you mean even when names are similar.

Redesigned the skill create and edit form with a new Activation Trigger section that uses progressive disclosure, making it easier to configure when skills activate. See the Activation triggerActivation trigger section in the skills overview documentation.

Tools

Improved the Optimizely documentation search tool with better pickup accuracy and more relevant results, helping you find answers faster. See System tools overviewSystem tools overview for information.

April 13

Released the following Optimizely Product Recommendations toolsOptimizely Product Recommendations tools in Agent PlatformAgent Platform to help you monitor and analyze product, search, and category performance directly in Agent Platform.

pr_get_category_report – Retrieves a category report containing performance metrics based on a provided date range and site configuration.

pr_get_search_report – Generates a search report detailing search performance and metrics for a specific site and timeframe.

pr_get_trend_report – Retrieves a trend report analyzing performance by product, brand, or category over a designated period.

April 7

CMS 13

Released the Optimizely.Cms.OpalChat 2.0.0 NuGet package that includes Mark chatMark chat, so editors can access AI-powered assistance without leaving CMS.

April 1

Agent Platform platform

Added file type icons when attaching certain filesattaching certain files (for example, .docx, .xlsx, and ppt) to an agent, making it easier to identify your uploads at a glance.

Agents

Updated the webhook triggerthe webhook trigger with a redesigned configuration form that streamlines configuration and lets you select product instances directly within the trigger configuration.

Updated webhook triggers to support default authentication header values for Content Marketing Platform (CMP), eliminating the need to configure custom values when the defaults meet your requirements. See Workflow agent triggersWorkflow agent triggers.

Added parameter generation logsparameter generation logs to workflow execution logs. When agents include tool calls, the logs display within thinking chunks and capture how Mark interprets input, isolates required parameters, and constructs the JSON structure passed between agents.

Tools

Released nested type schema support for custom tool parameters. Custom toolsCustom tools now accept complex, deeply nested data structures, including objects within objects, arrays of objects, and enums, letting you create richer integrations with external systems. See for information.

March 30

Skills

Added the ability to create personal skills in Agent Platform. Personal skills let you customize how Mark responds to your requests, tailoring its tone, format, and behavior to your individual preferences. See Create a personal skillCreate a personal skill for information.

Added the ability to create and edit personalpersonal or organization-wideorganization-wide skills directly from a conversation. Enter save this as a skill or click Create skills at any point in a chat and Mark generates the skill, presents a review form, and saves it on confirmation. See step 9 in Through the Agent Platform websiteThrough the Agent Platform website for information.

Released skill mentions, which let you reference saved skills inline in Mark chat by entering / or clicking Use Skills. Use these to explicitly apply a specific skill to any prompt without moving away from the Agent Platform conversation. See step 15 in Through the Agent Platform websiteThrough the Agent Platform website or step 12 In an Optimizely productIn an Optimizely product for information.

Agent logs

Added the ability to view credit usage directly within agent logs, giving you visibility into how credits are consumed per agent run. See Specialized agent logsSpecialized agent logs and Workflow agent logsWorkflow agent logs for information.

Added the ability to export execution logs for specialized agents, letting you download and analyze agent run history outside of Optimizely Agent Platform. See Specialized agent logsSpecialized agent logs for information.

Workflow agents

Released the ability to view specialized agent details directly in the workflow builder, giving you immediate access to agent configuration without leaving the workflow context. See Create a workflow agentCreate a workflow agent for information.

Enhanced support for multiple logic outcomes within a single workflow agent conditionworkflow agent condition, letting you build more precise and flexible workflow branching without creating additional logic blocks.

Streamlined navigation so that clicking a workflow agent opens the workflow directly, reducing the number of steps needed to access and edit agent configurations. See Create a workflow agentCreate a workflow agent for information.

Added the ability to stop agent execution directly from the logs, giving teams greater control over running agents without navigating away from the log view.

Specialized agent logsSpecialized agent logs

Workflow agent logsWorkflow agent logs

Agent Platform agents

Released the following agents to the Agent DirectoryAgent Directory:

Content Ideation agentContent Ideation agent – Generates 10 content and campaign ideas for any topic, audience, or quarter by combining live web research with competitor gap analysis and seasonal opportunity mapping.

Social Post Research agentSocial Post Research agent – Researches trending angles, competitive context, and hashtags for any campaign topic, then packages findings into a structured brief ready to hand off for social post creation.

Social Post Generation agentSocial Post Generation agent – Takes a topic, URL, or key message and generates platform-native social posts for LinkedIn, X, Instagram, Facebook, and TikTok.

Creative Brief Generation agentCreative Brief Generation agent – Takes a campaign description, asset type, and key message and generates a complete, structured creative brief covering strategy, audience, messaging, and visual requirements, ready to hand off to a creative team or agency.

E-E-A-T Checker agentE-E-A-T Checker agent – Evaluates any piece of web content against Google's E-E-A-T quality signals, rating experience, expertise, authoritativeness, and trustworthiness with concrete evidence pulled directly from the content.

Agent Platform tools

Released the following connector tools:

Amplitude connector toolsAmplitude connector tools

Microsoft 365 connector toolsMicrosoft 365 connector tools (Depending on configuration)

Mixpanel connector toolsMixpanel connector tools

WordPress connector toolsWordPress connector tools

Released user-level authentication for certain connector tools, letting individual users link their personal third-party accounts to Agent Platform. Mark can now perform actions using each user's specific identity and permissions, ensuring accurate action attribution, granular data privacy, and a more personalized experience. See Connectors page overviewConnectors page overview for information. See the following documentation for connector tools that support user-level authentication:

Adobe Analytics connector toolsAdobe Analytics connector tools

Google Ads connector toolsGoogle Ads connector tools

Microsoft 365 connector toolsMicrosoft 365 connector tools (Depending on configuration)

Salesforce CRM connector toolsSalesforce CRM connector tools

WordPress connector toolsWordPress connector tools

Released support for remote Model Context Protocol (MCP) connector tools, extending Agent Platform's capabilities by connecting to approved external MCP servers. When an Agent Platform administrator enables a remote MCP for the organization, individual users authenticate their personal accounts so Mark can act on their behalf with their specific permissions and data access. See Connectors page overviewConnectors page overview for information

Atlassian Remote MCP connector toolsAtlassian Remote MCP connector tools

Conductor Remote MCP connector toolsConductor Remote MCP connector tools

March 27

Released a new version of the GA4 Web Traffic Report Generation agentGA4 Web Traffic Report Generation agent. This release includes improvements to custom date ranges, the comparison period, and exact dates are now printed before the report runs and shown in the header, so anyone pulling GA4 data manually can match them. See New version availableNew version available in the Agent Directory overview documentation for steps to update your agent.

March 18

Removed the creativity settingcreativity setting in the specialized agents configuration. Newer AI models no longer support user-configuration of this. Existing specialized agents that had a creativity value set continue to function normally. The setting is ignored by new models that do not support it.

March 16

Released a new version of the Real-Time Audience Inspection agentReal-Time Audience Inspection agent which answers questions about Optimizely Data Platform (ODP) real-time audiencesOptimizely Data Platform (ODP) real-time audiences. This version updates the tools called in the agent. See New version availableNew version available to update the agent if already installed.

March 13

Agent Platform platform

Released Commerce Connect as a product connectionproduct connection in Agent Platform. This lets an Opal Administrator connect your Optimizely Commerce Connect instance with an Agent Platform instance.

Workflow agents

Updated the steps to create a workflow from scratchcreate a workflow from scratch. When you create a workflow agent, you now land directly in the workflow builder with a new sidebar for agent details.

Updated the workflow agent editorworkflow agent editor UI. The workflow page now matches the Agent Directory visual style with updated header, scrollable tabbed content with a fixed footer, and the @Agent ID displayed in the sidebar header for quick reference.

Added the ability to switch between the workflow builder and setupswitch between the workflow builder and setup of a workflow agent.

Updated where you view the workflow agent logsview the workflow agent logs.

Updated Workflow agent conditions Workflow agent conditions to support multiple logic branches. You no longer need multiple separate nodes for each condition, just add one condition and various branches.

Added the following to Email triggers for workflow agents:

Ability to copy the email address. See step 2 in the Configure an email triggerConfigure an email trigger section.

Pre-populated sender and subject filter rows so you are not starting from scratch. See steps 3 and 4 in the Configure an email triggerConfigure an email trigger section.

Tooltips on subject filters. See step 4 in the Configure an email triggerConfigure an email trigger section.

Smarter delete behavior on filter rows. See the Configure an email triggerConfigure an email trigger section.

Specialized agents

Added the ability to download a record of your specialized agent's execution historydownload a record of your specialized agent's execution history as a CSV or Excel file, which is useful for tracking agent usage, reviewing inputs and outputs, and measuring ROI.

Added the ability to directly add example shotsexample shots by pressing / or { in the prompt template editor.

March 12

Effective March 12th, 2026 Optimizely may allocate Agent Platform workloads to any of its approved LLM models. Currently, Optimizely's Agent Platform utilizes Google Gemini's family of models as its LLM.

As of March 15th, 2026, Agent Platform's LLM models will include Anthropic Claude's family of models accessed via Google Vertex AI. For the purposes of Optimizely's Data Processor Agreement, Anthropic will become an Optimizely Sub-processor as from March 15, 2026. Optimizely reserves the right to add, remove, and/or substitute Agent Platform's underlying LLMs. All approved LLMs for Agent Platform are detailed in Optimizely's Software Service Services Description. Optimizely Agent Platform's allocation of workloads to different LLMs and the associated model will not alter how Agent Platform credits are consumed from Agent Platform usage.

March 4

Released a new version of the AI variation development agentAI variation development agent for Web Experimentation which helps you modify and update existing website elements, create new ones, and generate and apply enhancement suggestions, while maintaining brand consistency by automatically retrieving page styles.

March 3

Released a new version of the Competitive Insights agentCompetitive Insights agent, which provides a thorough review of competitive intelligence and market research on competitors. This version updates the tools called in the agent. See New version availableNew version available to update the agent if already installed.

March 2

Released a new version of the GEO Auditor agentGEO Auditor agent, which audits any web page for AI search readiness. This version adds the browse_web tool to the agent. See New version availableNew version available to update the agent if already installed.

February 24

Agent Platform platform

Introduced the new Agent Builder roleAgent Builder role, letting users create and manage specialized and workflow agents. This role is for users who need to customize or extend Agent Platform's AI capabilities without requiring administrator privileges.

Released customizable permissionscustomizable permissions to create custom rolescreate custom roles for Agent Platform users for more precise access control.

Added user names for the Created By, Triggered By, and Modified By fields for agentsagents and skillsskills.

Released the Optimizely Agent Platform app as a verified application in the Slack Marketplace, streamlining integration and enhancing collaboration for teams. See Optimizely Agent Platform in SlackOptimizely Agent Platform in Slack.

Mark chat

Standardized terminology, replacing references to threads and conversations with chats for a cleaner experience. See step 6 in the Through the Agent Platform appThrough the Agent Platform app section or step 8 in the In an Optimizely productIn an Optimizely product section in the Mark chat documentation.

Improved the responsiveness of the chat interface by showing a thinking... message immediately after you send a message.

Added the ability to upload Microsoft Excel (.xlsx) and Word (.docx) files in Mark chatMark chat, eliminating the need for file conversions.

Canvas

Added the ability to download canvasesdownload canvases letting you access your canvas content offline and share with stakeholders. The file types you can download depend on the canvas' MIME type. See Available download formatsAvailable download formats.

text/markdown

Microsoft Word – .docx

PDF document – .pdf

Plain text – .txt

Rich text format – .rtf

Markdown – .md

text/html

HTML – .html

PDF document – .pdf

Plain text – .txt

application/x-reactjs

JavaScript – .js

JavaScript JSX – .jsx

application/x-nextjs

JavaScript – .js

JavaScript JSX – .jsx

text/csv

Comma Separated Values – .csv

Microsoft Excel – .xlsx

PDF document – .pdf

application/vnd.ms-powerpoint

Microsoft PowerPoint – .pptx

PDF document – .pdf

Added an Edit in canvas button directly within Mark's response on the last message in a chat, as long as that message is not an agent output or canvas output already. See Create a canvas.Create a canvas.

Agents

Updated the Agent DirectoryAgent Directory to show a second description line on agent tiles, making it easier to understand what each agent does at a glance.

Added the ability to copy variables directly from the execution log sidebar, making it easier to reference and debug agent runs. See step 7 in the Workflow agent log overview section of the workflow agent logsstep 7 in the Workflow agent log overview section of the workflow agent logs and step 11 in the specialized agent log overview sectionstep 11 in the specialized agent log overview section.

Launched the ability to add Example shotsExample shots to your agent's prompt template. Example shots help the agent better understand the desired output format, tone, style, and specific requirements of the task. This reduces ambiguity and can lead to more accurate and relevant responses, especially for complex or nuanced tasks.

Released the following agents to the Agent DirectoryAgent Directory:

Competitive Webpage Analysis agentCompetitive Webpage Analysis agent – Scans your competitors' websites, captures screenshots, and delivers an actionable report with recommended changes to outperform competitors so you can easily see what is working in your market and how to stay ahead.

Email Content Translation agentEmail Content Translation agent – Retrieves and translates the full content (body, subject, preview, and alt-texts) of a specific email in Optimizely Campaign, ensuring accuracy, cultural relevance, and consistent brand voice. It provides the original and verified translated versions side-by-side in a two-column format.

Email Optimization agentEmail Optimization agent – Analyzes provided newsletter content in Optimizely Campaign to give structured feedback and detailed recommendations for improvement. It evaluates readability, engagement, and visual appeal, offering specific sections for optimization tips, brand consistency checks, A/B testing suggestions, and a dynamic Content Score.

GEO Auditor agentGEO Auditor agent – Audits any web page for AI search readiness. You can get a comprehensive Generative Engine Optimization (GEO) report covering AI crawler accessibility, Core Web Vitals, schema markup, content structure, and citation readiness, with a prioritized action plan to improve your visibility in AI-powered search engines like ChatGPT, Perplexity, and Google AI Overviews.

Subject & Preview Text Ideation agentSubject & Preview Text Ideation agent – Analyzes provided email content from Optimizely Campaign to create three highly effective pairs of subject lines and preview texts (focusing on urgency, curiosity, and benefit). It optimizes these pairs to boost open rates, adheres to character limits, and provides a marketing rationale for each suggestion.

Web Experimentation and Personalization tools

Released the following tools in Agent Platform to help you use the AI variation development agent in Web ExperimentationWeb Experimentation and PersonalizationPersonalization.

exp_ve_apply_change – Applies a pending change to the Visual Editor.

exp_ve_element_glob – Finds elements in the Visual Editor DOM that match a glob pattern.

exp_ve_element_grep – Searches for elements in the Visual Editor DOM whose content or attributes match a regular expression.

exp_ve_element_read – Reads the content or attributes of a specific element in the Visual Editor.

exp_ve_element_tree – Retrieves a partial DOM tree for a specified element and its descendants in the Visual Editor.

exp_ve_get change_schema – Retrieves the JSON schema for a specific type of Visual Editor change.

exp_ve_get_selector – Generates a robust CSS selector for a given element in the Visual Editor.

exp_ve_list_pending_changes – Lists all pending Visual Editor changes for the current experiment or personalization.

exp_ve_revert_change – Reverts a specific pending Visual Editor change, undoing its effects.

exp_ve_validate_change – Validates a Visual Editor change object against its schema to ensure it is well-formed.

Content Marketing Platform (CMP) tools

Released the following toolstools in Agent Platform to help you orchestrate and streamline your content marketing workflows in CMP:

create_campaign_from_work_request – Creates a new campaign in Agent Platform based on an existing work request.

create_milestone – Creates a new milestone within a specified campaign or project in CMP.

create_task_from_work_request – Creates a new task in CMP based on an existing work request.

create_work_request – Create a new work request in CMP using a specified template.

get_form_template_by_id – Retrieves a specific form template using its unique identifier.

get_form_templates – Retrieves a list of all available form templates in CMP.

get_object_templates – Retrieves a list of available object templates in CMP.

get_saved_view_resources – Retrieves resources associated with a specific saved view in CMP.

search_cmp_resources – Searches for various resources within CMP based on specified criteria.

canvas_to_structured_content – Converts the content of an AI canvas into a specified structured content (an asset type) within a specified CMP task.

update_asset – Updates various properties of an asset in the CMP Library, including metadata, tags, location, and archival status, using partial updates.

update_library_asset_field – Update a single field value on a CMP library asset, supporting various CMP field types and value formats.

update_milestone – Modifies the details of an existing CMP milestone, such as its name, due date, status, or description.

update_work_request_resource_link – Modifies an existing resource link associated with a specific work request, such as updating its URL, description, or type.

Optimizely Analytics tools

Released the following toolstools in Agent Platform to help you create explorations in Optimizely Analytics:

oa_find_metrics – Searches and finds existing metrics from Optimizely Analytics. Uses fuzzy matching on metric names and semantic search on descriptions to find relevant metrics.

oa_list_apps – Lists all available applications in Optimizely Analytics.

oa_get_metric_data – Retrieves and executes a specific saved metric in Optimizely Analytics, returning its calculated table data.

oa_find_datasets – Searches and lists available datasets in Optimizely Analytics. It uses fuzzy matching on dataset names and semantic search on descriptions to find relevant datasets.

February 5

Added the ability for you to upload .xlsx and .docx files to Mark chat, letting you directly share and collaborate on common document types without having to switch between applications. See the Mark chat sectionMark chat section in the Supported file types in Optimizely Agent Platform documentation.

Added the Status column to the Skills pageSkills page to quickly see if a skill is active or inactive.

Updated skills to be inactive by default. You must now toggle Active to On when creating a skillcreating a skill. This prevents unintended activation of skills and ensures only fully prepared and reviewed skills go live.

Agent Directory

Updated the organization of agentsorganization of agents in the Agent Directory to be structured by use cases, improving the discoverability and navigation while making it easier to find the right agent for a specific task.

Added a visual indicator displaying if an agent is newnew. This indicator helps you quickly identify and explore newly added agents, ensuring you are aware of the latest features available to you.

January 27

Agents

Released the following agents to the Agent DirectoryAgent Directory:

Content Refresh AnalysisContent Refresh Analysis – Identifies, sorts by last edit date, and helps resolve duplicate and outdated content (in terms of content guidelines as reference) across a website in CMS (SaaS) to improve Search Engine Optimization (SEO) performance, site credibility, compliance, and content maintenance efficiency.

Experimentation Program OverviewExperimentation Program Overview – Creates a report on how your Optimizely Experimentation program is performing using top experiment and win rate data.

FAQ CreationFAQ Creation – Analyzes existing content to generate contextually relevant, Generative Engine Optimization (GEO)-optimized Frequently Asked Questions (FAQs).

GEO Schema OptimizationGEO Schema Optimization – Lets you enhance your CMS (SaaS) content's visibility and (LLM) discoverability by automatically identifying opportunities for structured data markup.

Product PromotionProduct Promotion – Builds and configures product promotions in Commerce Connect, seamlessly handling the assignment of products and variants and the application of specific promotion types.

Profound Citation Gap AnalysisProfound Citation Gap Analysis – Analyzes competitive citation performance for any topic using Profound's AI search tracking data.

Real-Time Audience InspectionReal-Time Audience Inspection – Answers expert questions about an Optimizely Data Platform (ODP) real-time audience, its purpose, how it overlaps with other audiences to achieve your stated goal, and other general troubleshooting queries.

SEO Metadata ImplementationSEO Metadata Implementation – Evaluates existing SEO properties, identifies optimization opportunities, recommends improvements, and upon your approval, creates or updates SEO metadata fields in your CMS (SaaS) instance.

SEO Metadata OptimizationSEO Metadata Optimization – Evaluates a URL for existing SEO properties, identifies optimization opportunities, and recommends improvements.

OCP connector tools

Released the Crunchbase connector toolsCrunchbase connector tools in Agent Platform, which provide AI-powered access to the world's most comprehensive dataset of companies, investors, funding rounds, and key business personnel.

Released the Zoom Webinar connector tools Zoom Webinar connector tools in Agent Platform, which let you summarize webinar transcripts, get and update webinar details, and get poll results.

CMS (SaaS) tools

Released the following tools in Agent Platform to help you manage your CMS (SaaS) content. See Content items section in the System tools for Content Management System (CMS) documentation Content items section in the System tools for Content Management System (CMS) documentation for more information.

cms_create_content_item – Creates a new empty content instance in CMS (SaaS) based on an existing content type.

cms_get_content_data – Retrieves CMS (SaaS) content data for a specific content item.

cms_delete_content_item – Deletes a content item from CMS (SaaS) using its unique content key

cms_publish_content_item – Publishes a specific content version in CMS (SaaS), making it visible to site visitors.

cms_update_content_item – Updates the value for any property of a content item using a "JSON Merge Patch" algorithm.

Optimizely Graph tools

Released the following tools in Agent Platform to help you manage your Optimizely GraphOptimizely Graph content in CMS (SaaS)CMS (SaaS) and CMS 12.

graph_content_type_schema – Your starting point to understand the content structure within your Optimizely Graph instance. The tool lets you and Mark discover available content types and their fields, which is crucial for constructing accurate GraphQL queries.

graph_content_graphql_executor – Lets you execute custom GraphQL queries against your Optimizely Graph instance. You can retrieve specific content items, filter results, sort, and get aggregated data (facets).

graph_content_search_tool – Use as a simpler alternative for searching content in Optimizely Graph when constructing complex GraphQL queries might be excessive. It lets you search for content based on a search phrase across a specified content type.

January 22

Personalization

Released the following system tools in Optimizely Agent Platformsystem tools in Optimizely Agent Platform to help you with your Personalization projects:

exp_get_schemas – Retrieves detailed schemas for various Optimizely entities relevant to Personalization. This is a mandatory first step before using the exp_execute_query tool to retrieve actual data about these entities.

exp_execute_query – Executes a template-based query to fetch specific data from your Optimizely Personalization instance. It lets you retrieve detailed information about various entities.

Experimentation

Released the following Feature ExperimentationFeature Experimentation and Web ExperimentationWeb Experimentation system tools in Optimizely Agent Platform to help you with your Experimentation projects:

exp_get_schemas – Retrieves detailed schemas for various Optimizely entities relevant to Experimentation. This is a mandatory first step before using the exp_execute_query tool to retrieve actual data about these entities.

exp_execute_query – Executes a template-based query to fetch specific data from your Experimentation instance. It lets you retrieve detailed information about various entities such as feature flags, rules, environments, attributes, and events, based on your specified criteria.

exp_program_reporting_top_experiments – Retrieves experiments that have generated the highest positive or negative lift values within a specified timeframe. It focuses on the performance of individual tests, letting you see which variations of your features are performing best (or worst) against your chosen metrics.

exp_program_reporting_underperforming_experiments – Helps you identify A/B tests that are not yielding significant or positive results, letting you quickly decide whether to stop, re-evaluate, or iterate on them.

exp_program_reporting_win_rate – Gives you a high-level overview of the success of your experimentation program by calculating the win rate over a specified period. It is a key metric for understanding the efficiency and impact of your product development cycle.

January 14

Agent Platform platform

Added support for .csv file upload in Mark chatMark chat and in agentsagents.

Added appearance setting (font size and dark mode) for usersusers and administratorsadministrators.

Mark chat

Added a toggle to show or hide the Mark chat history panel. See Mark chat user interface through the Agent Platform appMark chat user interface through the Agent Platform app.

Updated the Chat input editor and added support for basic Markdown (bold, italics, inline code). See Mark chatMark chat.

Tools

Updated the Agent Platform platform to disable new tools and registries by default, requiring Agent Platform administrators to explicitly enable them.

Released system tools for Campaignsystem tools for Campaign, helping you easily get information about your Smart campaigns.

Updated the analyze_pagespeed tool'sanalyze_pagespeed tool's results to display the title of the score returned.

Agents

Updated workflow agentsworkflow agents to let you add a triggertrigger to your workflow first.

Added the ability to add existing specialized agent executions as Output Evalsadd existing specialized agent executions as Output Evals.

Added versioningversioning in specialized agents, including the ability to viewview and restorerestore specialized agents to previous versions. You can also create a new specialized agent based on a previous versioncreate a new specialized agent based on a previous version.

Updated the Creativity selectionCreativity selection to a slider and updated the options to not include 0.0.

Updated the Prompt Template to include the ability to add variables, tools, and skills by entering / [, or \ respectively. See Additional prompt template configurationAdditional prompt template configuration.

Updated variables in specialized agentsvariables in specialized agents.

Variables are now case-insensitive, making it easier for Mark to correctly map variables regardless of casing. For example, [[Test]] in the prompt template now matches a variable named test.

Variables cannot have duplicate names. So, if you create a variable named test, you cannot add another variable named test, Test, or some other casing of the same variable name. This ensures data integrity and consistency.

Updated the tools selection page and added the ability to view a description of the tool before adding it. See Define toolsDefine tools.

Added a description of the tools listed on the specialized agent page. See Define toolsDefine tools.

Removed the ability to change an agent's ID for agents installed from the agent directoryagent directory, making it easier to update these agents in the future.

Added a link to the agent on the confirmation pop-up when installing agents from the agent directorinstalling agents from the agent director, letting you access the agent quickly and review the newly imported agent.

Product connections

Added Optimizely Product RecommendationsOptimizely Product Recommendations as an option to select from the Product connectionsProduct connections.

Moved the Connections tab to Settings. See Product connectionsProduct connections.

Product Information Management (PIM)

Released Mark chatMark chat, letting you interact with Agent Platform from PIM.

Released system tools for PIMsystem tools for PIM, which help you retrieve general statistics from the Dashboard.

January 12

Optimizely Reporting for Agent Platform

Added Agent Platform Agent, the primary agent for chat and feature interaction, to the Opal Agent UsageOpal Agent Usage dashboard.

Note

An administrator can toggle generative AI availability off for all users of an instance from the Opti ID Admin Center. See Turn generative AI offTurn generative AI off.

2026 Optimizely Agent Platform release notes

October 13

Agent Platform

Virtual Teammates

Agents

Connectors

October 6

Connectors

October 5

September 30

September 21

September 18

Connectors

MCP connectors

September 17

Agent Platform﻿ platform

Workflow agents

Optimizely Product Recommendations tools

September 13

September 2

August 27

August 25

August 14

Agent Platform﻿ platform

Optimizely Graph tools

July 29

Agent Platform﻿ platform

Workflow agent updates

UX improvements

Agents

Connectors

July 22

Agent Platform﻿ platform

Tools

June 30

Opal Agent Usage dashboard

Agent Platform﻿ platform

Agents

Tools

CMS 13 Tools

June 24

Agent Platform﻿ platform

Mark chat

Agents

Skills

RAG for CMP

June 8

June 4

Agent Platform

Agents

Connector tools

MCP connectors

CMS (SaaS) tools

June 1

May 27

May 26

Mark chat

Canvas

Agents

Skills

Notifications

May 22

CMS 13

May 18

May 14

Optimizely Product Recommendations

May 11

CMS (SaaS) Agent Platform﻿ tools

May 7

Agent Platform

Agents

Tools

April 30

Workflow agents

Mark chat

April 22

Optimizely Web Experimentation

April 14

Mark chat

Agents

Agent Platform platform

Agent Platform platform

Agent Platform platform

Agent Platform platform

Agent Platform platform

Agent Platform platform

CMS (SaaS) Agent Platform tools

Agent Platform platform

Agent Platform agents

Agent Platform tools

Agent Platform platform

Agent Platform platform

Agent Platform platform

Optimizely Reporting for Agent Platform