DSPy

What it does

DSPy is a Python framework for building AI systems. Developers write the task as a structured signature with typed inputs and outputs rather than a hand-written prompt, then wrap it in a module that sets the strategy, such as direct prediction, step-by-step reasoning or a tool-using ReAct agent. You connect a language model through the library and call the resulting program like ordinary Python.

To improve quality, you supply examples and a scoring metric and run an optimizer such as GEPA, which tunes the prompts or weights of the program against that metric. The output is a saved, reusable program. DSPy also covers retrieval, classification, extraction, multimodal inputs and agent loops, with tools and MCP support.

It is for developers and researchers comfortable writing Python. It is free under the MIT license; you pay your model provider for the calls your programs and optimization runs make. Recent releases changed the language-model integration, so check the migration guide when upgrading.

What it can do

• Declare tasks as typed signatures instead of managing prompts
• Compose modules for direct prediction, reasoning and tool-using ReAct agents
• Optimize programs against a metric with optimizers such as GEPA
• Save and load compiled programs
• Handle image inputs in signatures
• Use tools and MCP from agent modules

Works with

• MCP

Before you choose

• You supply the language model access; optimization runs make many model calls, so cost depends on the provider and model you choose.
• Recent versions use bundled native LM engines without a LiteLLM dependency, with breaking changes to provider options, custom LMs, text completions and embeddings.
• It is a developer library used from Python code, not a hosted app or visual builder.

Listed under

• Coding
• MCP and agent tools
• Developers
• Developer tools
• LLM pipelines
• LLM-pipelines
• Open source
• Programming-framework

Get started

Install from PyPI and follow the Getting Started guide; the source is on GitHub under the MIT license.

1. Install with pip install -U dspy (Python 3.10 or later).
2. Connect a language model with dspy.LM and define a signature and module for your task.
3. Add examples and a metric, then compile with an optimizer and save the program.

The library is MIT-licensed; no paid plans are listed on the project site.

Source: dspy.ai, checked 1 Nov 2026. Prices can change; check before you buy.

Sources for this overview

Product pages, documentation, and pricing checked 1 Nov 2026. Each source supports part of the overview; availability and prices can change.

• DSPy: Program, don't prompt, your LLMs ↗dspy.ai
• GitHub - stanfordnlp/dspy: DSPy: The framework for programming—not prompting—language models · GitHub ↗github.com
• FAQ - DSPy ↗dspy.ai

Company website

dspy.ai: product, pricing and policy pages

Web archive

When the site first appeared

YC directory mirror

Batch and company name

Social accounts

Accounts the company links, and its posts on X

Human review

Checked by a person

knowledge_base

manual_review

Last checked 1 Nov 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.