Direct LM Calls with dspy.Predict

Guide the user through using dspy.Predict -- the simplest and fastest DSPy module for calling a language model. It takes inputs, calls the LM, and returns typed outputs. No intermediate reasoning, no extra steps.

Step 1: Gather context

Before writing code, clarify:

1. What is the task? Classification, extraction, formatting, or Q&A?
2. What output type is needed? A plain string, a constrained Literal list, or a structured Pydantic model?
3. Do you need batch processing? Thousands of items at once, or single calls?
4. Is there an existing signature, or does it need to be designed from scratch? (If from scratch, see /dspy-signatures.)

What is dspy.Predict

dspy.Predict is the atomic building block of every DSPy program -- one LM call, no reasoning chain, no tool loops. It takes a signature and calls the LM once to produce the output fields. Every other DSPy module (ChainOfThought, ReAct, etc.) builds on top of it.

When to use Predict vs ChainOfThought