# AI Inference

Call a language model to generate text from a prompt, for tasks such as summarization, rewriting, or explaining a classification.

Use [Information Extraction](information-extraction.md) when you need fixed fields and their available source references.

## Configure generation

| Setting | Description |
| --- | --- |
| Prompt | Required and nonempty; combine instructions with upstream content |
| System prompt | Optional role, style, and output requirements |
| LLM model | Optional; defaults to the workspace AI inference service configuration |
| Temperature | Optional sampling parameter passed to the model |
| Maximum tokens | Optional; if supplied, must be a positive integer |

Parse files first, then include their text in the prompt.

## Use the output

`text` and `result` contain the same generated text. `model` identifies the model used. Bind either text field to the next step.

For extraction governed by a JSON Schema, use [Information Extraction](information-extraction.md).

## Example: summarize meeting action items

Flow: Parse Audio → AI Inference → Save to MOI Volume.

Include the transcript's `text` in the prompt. Ask for the task, owner, and due date, and instruct the model to leave missing information unspecified. Save the generated text as a file.

## Related pages

- [Variables and data flow](../../variables-and-data.md)
- [Run and debug](../../run-debug.md)