DSPy overview and key facts (October 2026)

DSPy is an open-source Python framework that lets developers describe what a language model should do in code, then optimizes the prompts and examples automatically against a metric. It suits engineering teams that want repeatable prompt quality instead of manual trial and error.

DSPy has been developed by the Stanford NLP Group since December 2022 and is MIT licensed. Signatures declare typed inputs and outputs; modules such as Predict, ChainOfThought and ReAct decide how a signature is executed. Optimizers, including GEPA, MIPROv2 and BootstrapFewShot, compile a program against a metric by tuning instructions and few-shot examples, and can also fine-tune weights. The project reports 5.2 million monthly downloads and production use at companies such as Databricks, Shopify and Dropbox. As of September 28, 2026 the repository had about 38,400 GitHub stars.

• Optimizes prompts against a metricOptimizers try instructions and examples, measure the result with your metric and keep what works. Prompt tuning becomes a repeatable process instead of manual guesswork.
• Ports tasks between modelsBecause tasks are defined as signatures, a program can be recompiled for a different model. Switching providers does not mean rewriting every prompt by hand.
• Measures quality systematicallyEvaluation primitives score a program on a dataset. Teams compare versions and track whether changes improve the target metric.

• Treats prompts as compiled output of code rather than hand-written text.
• Backed by published research papers, including GEPA from July 2025.
• Free and model-agnostic, with no hosted service to buy.

Research project of the Stanford NLP Group, used worldwide as open source, without a country focus.

Optimization runs make many model calls, so budget for API costs during compilation. You need a labeled dataset and a clear metric for the optimizers to work. The framework is code only; your team runs and maintains it and chooses where model calls go.

No reviews yet. Be the first to share how your team uses it.

Every review shows its verification level. We read each one before it goes live. How we check reviews