Prompt inputs are named placeholders, or input variables, that let an AI Builder prompt receive dynamic values at runtime so the same prompt can be reused across many records, files, and scenarios instead of being hardcoded. Without inputs a prompt would only ever process the literal text you typed; with them, an app, a flow, or an agent can pass fresh content each time the prompt runs. AI Builder supports several input modalities. Text inputs carry text content for actions such as summarization, classification, information extraction, sentiment analysis, and generation. Image or document inputs carry file content (PNG, JPG, JPEG, and PDF) for file summarization, categorization, and extracting both textual and visual information, and you can combine text and file inputs in one prompt. Power Fx inputs let you augment the prompt with formulas, using non-connector functions and referencing other inputs with the syntax Input.'Your input name'. You add an input in Prompt builder, reached through AI hub then Prompts, by placing the cursor in the Instructions panel and typing a slash or selecting Add content, then choosing Text, Image or document, or Power Fx. The inserted token is referenced inline within the instruction text, you can rename it from the default such as Text input in the Name field, and you supply Sample data so you can test the prompt before publishing. Inputs must not contain instructions, which are blocked for security reasons. Callers supply real values when consuming the prompt: in Power Apps the prompt is added as data and called as a Power Fx function with Predict, passing arguments such as TextInput1.Text, and in Power Automate the Run a prompt action shows each defined input so you bind it to dynamic content from earlier steps. Limits include 25 MB total for files, fewer than 50 pages per document, and region availability, which all shape how reusable parameterized prompts behave.