DSPy

Framework for programming language models instead of prompting them

Editorially reviewedChecked Oct 2026How we review

agent-frameworks#prompt-optimization#llm-programming#rag#open-source

Free plan Self-hosted

Toolglade’s take

Best for teams who want to program LLM pipelines as optimizable modules and let the framework auto-tune prompts and weights, rather than hand-crafting strings. The caveat is a steep learning curve and heavy abstractions that are overkill for simple one-shot prompting.

Should you use DSPy?

Best for: ML and NLP engineers · Researchers building RAG systems · Teams wanting reproducible LLM pipelines
Not ideal for: No-code users · Simple one-off prompt tasks · Non-Python stacks
Biggest strength: Removes brittle manual prompt engineering
Biggest weakness: Steeper conceptual learning curve
Pricing: Free plan, paid from $0
Top alternative: Semantic Kernel · see all

Facts & pricing last checked October 2026. How we review

About DSPy

DSPy is a Stanford NLP open-source Python framework that treats LLM development as programming, compiling declarative modules into automatically optimized prompts and weights.

DSPy (Declarative Self-improving Python) reframes LLM development as programming rather than prompting. Developers write declarative modules with input-output signatures, specify how the model should reason and define metrics for quality, and DSPy's optimizers search for the best few-shot examples, instructions and configurations automatically. This shifts brittle, manually tuned prompts into a compiled, testable software artifact. Maintained by Omar Khattab and the Stanford NLP group with contributions from a broad research community, DSPy has grown to roughly 34,000 GitHub stars and is on its 3.x line as of 2026. It is MIT-licensed and free to use. The framework is popular for RAG systems, structured extraction, classification and agent pipelines where reproducible, optimizable behavior matters more than one-off prompt crafting.

Weighing your options?See how DSPy compares to the alternatives.

TL;DR

DSPy is Stanford NLP's open-source framework that compiles declarative modules into optimized prompts, replacing manual prompt engineering.

Company overview

DSPy is a research-driven open-source project from the Stanford NLP group rather than a commercial company. It is led by Omar Khattab with a large community of contributors, and is distributed free under the MIT license.

By 2026 the project reached roughly 34,000 GitHub stars and the 3.x release line, cementing its role as a leading framework for optimizable LLM programs.

Product features

DSPy lets developers define signatures and modules describing tasks, then uses optimizers to search for the best instructions and few-shot examples against a metric. This turns prompt engineering into a compiled, reproducible workflow.

It supports classifiers, RAG pipelines and agent loops, integrates with major model providers and vector stores, and emphasizes measurable quality over ad hoc prompt tweaking.