DSPy

Overview

DSPy is a Python framework for building AI systems. Developers declare tasks as typed signatures and compose modules, then use optimizers to tune prompts and weights against a metric. The documentation covers tools, RAG, agents, multimodal fields, evaluation, and deployment topics.

Key Features

• Structured signatures declare tasks as typed inputs and outputs instead of prompts.
• Reusable modules such as Predict, ChainOfThought, and ReAct change execution strategy without rewriting the task.
• Optimizers compile a program against a metric to improve output quality.
• Programs can compose multiple signatures and modules using plain Python control flow.
• Documented capabilities include tool use, RAG pipelines, agent loops, and image fields for multimodal tasks.

Use Cases

• Building simple classifiers, sophisticated RAG pipelines, and agent loops.
• Extracting structured information from emails, support tickets, or other text.
• Creating tool-using agents with ReAct and function tools.
• Handling multimodal tasks that use images as signature fields.

Getting Started

• Install DSPy with pip install dspy.
• Follow the Getting Started documentation to build a first program and expand signatures.

Deployment & Requirements

• Requires Python 3.10 or newer.

Before You Adopt

• License: MIT. Review its terms before using, modifying, or distributing the project.
• Some optimization capabilities are explicitly labeled experimental.