Cloud changelog

More Precise Agent Builder Limits

Agent Builder now understands requests such as “get the first 100 records.”

It adds a Max items limit so the agent stops after collecting the requested number, even across multiple pages.

Better Screenshot Workflows

Screenshot commands can now be executed and previewed without running the entire agent.

Captured screenshots appear in a downloadable list, and filenames can be customized using transformations.

Space-Level Access

Org admins can control which Spaces each user group can access. Permissions can be set for viewing, editing, running, scheduling, deleting, and approving agents.

Agent Approvals

Organizations can require approval before agents are scheduled or run through the API. Approval requests are managed from a central Pending Approvals screen.

Filter Selection

The Editor now makes it easier to refine selections using element attributes, without manually writing XPath.

Multi-Input Agent Builder

Agent Builder can create Data List workflows that process multiple inputs, such as searching several keywords one at a time.

Remote MCP

Our remote MCP server now speaks OAuth 2.1 and streamable HTTP. Connect the agentic tool of your choice without managing credentials by hand — clients register themselves and discover how to authenticate.

Resources and Prompts

MCP now exposes Resources and Prompts alongside tools, so your assistant can pull context about your agents and spaces. Start an Agent Builder build from your assistant and get notified when it finishes.

Improvements & Bug Fixes

• Faster Run History and Usage pages
• Leaner memory use on Parquet exports
• Element-level capture in the Screenshot command
• Sharper auto-generated agent descriptions
• More reliable Auto Scroll on lazy-loading sites
• Improved organization filtering and search
• Editor and mobile polish across Control Center

Agent Builder (Beta)

Sequentum Cloud's Agent Builder is now available. Describe what you want to automate — or paste a URL with instructions — and our AI assembles the agent for you.

Use a prompt to build an agent in minutes. No AI at runtime to keep your data reliable and repeatable. When a build finishes, jump into Review in Editor to inspect or customize the agent further, or just hit Run or Schedule to put it to work.

Kick off a build and run agents from your own code. POST a prompt to /api/v1/agent-builder, poll for status, then start the agent. cURL, Python, and TypeScript snippets are right in the UI. Integrate directly with your preferred AI tooling using our MCP Server.

Improvements & Bug Fixes

• More reliable pagination across a wider range of sites
• Sharper detail-page detection on deep, multi-level catalogs
• Magic Wand reliability improvements on complex sites
• Smoother exports on parallel, separated runs
• Editor polish across the board
• Clearer error messages when an agent fails
• Quieter, more useful logs behind the scenes

Smart Pagination

Added a dedicated pagination command, compatible with most pagination styles including infinite scroll. Automatically added in the magic wand.

Bug Fixes

• Improved parquet file compresssion
• Improved python script engine
• Optimized memory usage of run servers

Remote MCP

Sequentum Cloud remote MCP is now live. Add it to your agentic tool of choice to monitor and manage your Sequentum Cloud agents all in one place.

Infinite Scroll Detection

Infinite scrolling has been added to the magic wand extraction in the editor. On sites where pagination is handled through infinite scrolling, the extractor now detects and automates scrolling to capture all items on the page.