Input control defines how you pass data from a flow to an AI agent. You use it to build clear, reliable inputs from flow records, plain text, and supported files, while staying within model capabilities and platform limits. Proper input control ensures the agent behaves consistently for every record processed.
Configure inputs to make sure the agent receives the right context for each run, with the following benefits in mind:
-
Generate accurate outputs from structured flow data
-
Add human-readable context when records alone are not enough
-
Include documents or images when the selected model supports them
You can choose from four input modes. The platform automatically determines the mode based on what you map.
-
Record (JSON): Choose records when decisions depend on structured flow data. When you pass the current flow record as structured data, the agent receives a stringified JSON version of the record.
-
Text: Allows for freeform instructions or summaries that do not depend on structured data. You must send plain text only.
-
Blob (files): Send the model documents, such as PDFs, or images to analyze. Multiple files are allowed per run, subject to limits.
-
Mixed: Map records, text, and files in a single request. This option is common when you want structured data plus explanatory text or reference documents.
To access and modify a mapping, navigate to the flow and then click the mapping icon next to the AI agent. The Edit mapping panel opens.
Mappings allow you to define how data is handled between two applications. Each mapping interface features an Input and Output panel, complete with dropdown lists, to facilitate the visualization and definition of record mappings. You define a mapping by selecting one field from the input and one from the output.
For example, in an import mapping, you can match a field from your source record to a corresponding field in your destination record. If you want to further adjust the mapping output, click the Preview button whenever you want to see the effect of your recent changes.
Note
Edit mapping clearly separates destinations for record, text, files, and an optional conversation identifier.
-
Supported file types are images (png, jpg, webp) and PDFs
-
Each file is subject to a per-file size limit, and the total size per agent run cannot exceed 50 MB
-
If any file reference is invalid or inaccessible at runtime, the agent fails immediately with guidance
Note
Unsupported file types result in record failure.
Before you run an agent in production, you can preview exactly the input that will be sent. It helps you confirm that the agent receives the correct inputs and that your prompt is composed as expected.