DSPy
used in
0
stacks
Open-source Python framework for programming, rather than prompting, language models.
A Python framework from Stanford NLP that lets you express LLM pipelines as composable modules with typed signatures instead of hand-written prompt strings. Provides general-purpose modules such as ChainOfThought and ReAct, and optimizers that compile programs by automatically tuning prompts (and optionally weights) against a metric. Open source and free, with no paid product from the maintainers.
Strengths
- Declarative, modular programs replace brittle prompt strings
- Automatic prompt optimization against a defined metric
- Portable across models and providers (uses LiteLLM under the hood)
- Research-backed, with an active academic and OSS community
Tradeoffs
- Optimizer-driven workflow has a distinct learning curve
- Compilation and optimization can add cost and iteration time
- Less of a turnkey agent runtime than orchestration-focused frameworks