AI and Human Assistance Actions

These actions bring intelligence into an AI Employee job: they send a natural language prompt to ProSeer AI along with the right context (a screenshot, a file, plain text, or a live browser page), or hand the work to a person when human judgment is required. All of the AI prompt actions share a few settings: the Natural Language Prompt itself (the more detailed and specific, the better and more reliable the results - and it supports {{Field}} tokens, so process values can be woven in), an AI Model choice that trades capability against cost, an optional Do Not Cost Optimize switch, and How to Determine Success? (Optional) - a semicolon-separated list of acceptable responses or a formula, so the job can treat unexpected answers as failures. The AI’s response is stored so later actions can use it.

Tip

Ask ProSeer AI to respond using specific, predictable words that you can then list in How to Determine Success? (Optional). That turns a free-form answer into a dependable pass/fail check.

AI - Screen Related Prompt

Sends ProSeer AI a screenshot of the target region (by default, the currently focused window) along with your prompt, and lets it operate the mouse and keyboard to carry out the task - looking at the screen again after each step until it is done. Use it for on-screen work that deterministic actions cannot handle: unlabeled icons, unusual widgets, or judgment calls about what is displayed. Prefer Click On Visible Text for simple clicks on stable text - it is cheaper and more reliable. Key settings: Action Target Region, up to three Approved AI Input selections that limit what the AI may do (for example, read only, left clicks only, typing, or scrolling), and an Example Image (Optional) to show the AI what it is looking for.

AI - Text Prompt

Sends a plain text prompt to ProSeer AI and stores the single response. No screenshot or file is involved, making this the fastest and least expensive prompt type. Use it to summarize, classify, extract, reformat, or make a decision about text the job has already gathered - for example, deciding which category an email belongs to before branching.

AI - File Related Prompt

Sends a file to ProSeer AI along with your prompt - for example, to extract fields from a PDF invoice or answer questions about a document. Key settings: File Path and File Name (both support {{Field}} tokens), and AI File Type. Leave the file type blank for single-turn analysis of documents such as PDFs, images, and text files. Choose Excel Spreadsheet (.xlsx) to let the AI explore and even update a workbook step by step; that unlocks Excel-specific settings - Sheet Structure (table with headers, table without headers, or form/freeform), Target Sheet Name to restrict the AI to one sheet, Header Row for sheets whose column names are not on row 1, and Read-Only to guarantee the AI never writes to the file.

AI - Browser Open URL / Initialize

Starts (or reuses) the AI-controllable browser and navigates to the specified URL - this action must run before any AI browser prompt. It verifies the page actually loaded (a blank page counts as failure), brings the browser to the foreground, and saves the page title for later actions. Key setting: Open URL, which supports {{Field}} tokens; include the protocol (https://) in the address.

AI - Browser Related Prompt

Gives ProSeer AI control of the AI Employee’s managed browser window: it reads the structure of the current page, performs one step at a time (clicking, typing, selecting), and re-reads the page after each step until the task is complete. Because it works from the page’s actual structure rather than a screenshot, it is usually the most reliable AI option for websites - prefer it over AI - Screen Related Prompt whenever the work happens in the browser. Use the AI - Browser Open URL / Initialize action first to open the managed browser on the right page. Like the screen prompt, it supports up to three Approved AI Input selections to limit what the AI may do.

Human Assistance

Pauses the job and notifies a person that their help is needed - for example, to complete a step that requires judgment, credentials, or physical action. The job waits until the assistance is provided. Key settings: Who to Notify About this Task? (a semicolon-separated list of user IDs, user group IDs, role IDs, or email addresses; field tokens and formulas can be used) and Task Description, which is included in the notification email - it supports {{Field}} tokens, so the helper sees the specifics of this process instance.